Challenges and Solutions for Sustainable Smart City Development (EAI/Springer Innovations in Communication and Computing) 3030701824, 9783030701826

Citation preview

EAI/Springer Innovations in Communication and Computing

R. Maheswar · M. Balasaraswathi Ravi Rastogi · A. Sampathkumar G. R. Kanagachidambaresan  Editors

Challenges and Solutions for Sustainable Smart City Development

EAI/Springer Innovations in Communication and Computing Series editor Imrich Chlamtac, European Alliance for Innovation, Ghent, Belgium

Editor’s Note The impact of information technologies is creating a new world yet not fully understood. The extent and speed of economic, life style and social changes already perceived in everyday life is hard to estimate without understanding the technological driving forces behind it. This series presents contributed volumes featuring the latest research and development in the various information engineering technologies that play a key role in this process. The range of topics, focusing primarily on communications and computing engineering include, but are not limited to, wireless networks; mobile communication; design and learning; gaming; interaction; e-health and pervasive healthcare; energy management; smart grids; internet of things; cognitive radio networks; computation; cloud computing; ubiquitous connectivity, and in mode general smart living, smart cities, Internet of Things and more. The series publishes a combination of expanded papers selected from hosted and sponsored European Alliance for Innovation (EAI) conferences that present cutting edge, global research as well as provide new perspectives on traditional related engineering fields. This content, complemented with open calls for contribution of book titles and individual chapters, together maintain Springer’s and EAI’s high standards of academic excellence. The audience for the books consists of researchers, industry professionals, advanced level students as well as practitioners in related fields of activity include information and communication specialists, security experts, economists, urban planners, doctors, and in general representatives in all those walks of life affected ad contributing to the information revolution. Indexing: This series is indexed in Scopus, Ei Compendex, and zbMATH. About EAI EAI is a grassroots member organization initiated through cooperation between businesses, public, private and government organizations to address the global challenges of Europe’s future competitiveness and link the European Research community with its counterparts around the globe. EAI reaches out to hundreds of thousands of individual subscribers on all continents and collaborates with an institutional member base including Fortune 500 companies, government organizations, and educational institutions, provide a free research and innovation platform. Through its open free membership model EAI promotes a new research and innovation culture based on collaboration, connectivity and recognition of excellence by community. More information about this series at http://www.springer.com/series/15427

R. Maheswar  •  M. Balasaraswathi Ravi Rastogi  •  A. Sampathkumar G. R. Kanagachidambaresan Editors

Challenges and Solutions for Sustainable Smart City Development

Editors R. Maheswar VIT Bhopal University Bhopal, India Ravi Rastogi University of Bisha Bisha, Saudi Arabia G. R. Kanagachidambaresan Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology Chennai, India

M. Balasaraswathi Saveetha Institute of Medical And Technical Sciences (SIMATS) Chennai, India A. Sampathkumar VIT Bhopal University Bhopal, India

ISSN 2522-8595     ISSN 2522-8609 (electronic) EAI/Springer Innovations in Communication and Computing ISBN 978-3-030-70182-6    ISBN 978-3-030-70183-3 (eBook) https://doi.org/10.1007/978-3-030-70183-3 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Preface

Smart city development is the present focus of all the countries to save the nonrenewable resources and for creating a better future. The technologies like wireless communication, Internet of Things (IoT), and machine learning based system design aids in improving the architecture of the smart city environment. The sensors, actuators and processing units working in smart cities are exposed to rapid changing harsh environments, creating distrust in data being sent or received. Machine learning and deep learning based approaches solve this problem by providing confidentiality and high trust to the data. The challenge is not only in the harsh environment but also on attacks to such networks. The networks in the present environment are highly heterogeneous and the present research contributions mainly concentrate on homogenous networks issues and solutions. Implementing heterogeneous solutions in smart cities ends up with new issues and problems. This book discusses many aspects of smart cities in a wide scale and provides a clear view about smart and sustainable development of smart environments. The challenges faced in developing smart applications are also addressed along with the solutions in terms of reliability, security and financial issues. The smart city development which includes smart transportation, smart management, intelligent wearable electronics for smart health care and other services and smart homes are explored in a brief manner. The Internet of Vehicles (IoV) plays a vital role in creating three-dimensional smart city communication. IoV provides smart and intelligent applications that consume a significant amount of computing power and network resources. The effective management of road vehicular congestion via adaptive traffic signal scheduling policy, on the other hand, depends on proper measuring and assessment of the vehicular congestion. Hence, a significant number of road traffic congestion monitoring, predicting and assessment techniques using various types of Information and Communication Technologies (ICTs) have been devised over the past few decades, which are detailed briefly in this book. The role of deep learning in smart city development involves major sophistication and accurate predictions. Deep learning algorithms are also involved in applications like waste management and the healthcare domain. Specifically, the significance of deep architectures to classify the waste into recyclables or not are widely explored ­nowadays. This book also features about the state v

vi

Preface

of the art and emerging trends in IoTs for smart cities. The Strengths, Weaknesses, Opportunities, and Threats (SWOT) analysis for intelligent transportation system in smart cities is also highlighted. Bhopal, India  R. Maheswar Chennai, India   M. Balasaraswathi Bisha, Saudi Arabia   Ravi Rastogi Bhopal, India   A. Sampathkumar Chennai, India   G. R. Kanagachidambaresan

Acknowledgement

I would like to express my sincere thanks to Prof. Chlamtac, President, European Alliance for Innovation (EAI), and Eliška Vlčková, Managing Editor, EAI, for providing the opportunity of editing the book under the title, “Challenges and Solutions for Sustainable Smart City Development”. As a lead editor for this book, I would like to thank the Management, VIT Bhopal University, India, for giving me all kind of support and facilities to complete this book successfully. I also thank all my co-­ editors for their constant involvement and quick responses in the preparation of this book. I am highly grateful to all the authors for submitting their valuable research and also my heartfelt thanks to all the reviewers for giving their precious feedback then and there to improve the work. I would also like to thank all the supporting staff from Springer who really helped a lot and their extended support with quick and efficient efforts made the book finally a successful one.

vii

Contents

 Fog and Edge Computing for Automotive Applications ������������������������������    1 Sandhya Devi R. Subburaj, V. R. Vijay Kumar, P. Sivakumar, B. Vinoth Kumar, B. Surendiran, and A. Neeraja Lakshmi  Intelligent Transportation System in Smart City: A SWOT Analysis��������   17 Tejas Parekh, B. Vinoth Kumar, R. Maheswar, P. Sivakumar, B. Surendiran, and R. M. Aileni  Deep Learning in Smart Applications: Approaches and Challenges����������   49 M. Sowmiya, B. Banu Rekha, and R. Kanthavel  Smart Metering Using IoT and ICT for Sustainable Seller Consumer in Smart City����������������������������������������������������������������������������������������������������   75 L. Sathish Kumar, M. Ramanan, Jafar A. Alzubi, P. Jayarajan, and S. Thenmozhi  Deep Learning-Based Activity Monitoring for Smart Environment Using Radar������������������������������������������������������������������������������������������������������   91 N. Susithra, G. Santhanamari, M. Deepa, P. Reba, K. C. Ramya, and Lalit Garg  GIS-Based Air Quality Index Spatial Model for Indian Cities��������������������  125 Navneet Munoth and Neha Sharma  Intelligent Wearable Electronics: A New Paradigm in Smart Electronics����������������������������������������������������������������������������������������  169 Ribu Matthew, Jyotirmoy Dutta, R. Maheswar, and Kawsar Ahmed  Road Traffic Congestion Monitoring in Urban Areas: A Review����������������  199 Pampa Sadhukhan, Sahali Banerjee, and Pradip K. Das  Smart Waste Management Model for Effective Disposal of Waste Management Through Technology����������������������������������������������������������������  213 Ramalatha Marimuthu, M. Shanthi, Supavadee Aramvith, and S. Sivaranjani ix

x

Contents

 Edge Analytics and Deep Learning for Sustainable Development��������������  231 Santosh Kumar Singh, Supernova Chakraborty, and Vishal Soodan  Markov Model-Based Smart Home Assistance for Geriatric Care������������  253 V. P. Arul Kumar, V. Dineshbabu, and G. R. Kanagachidambaresan  State-of-the-Art and Emerging Trends in Internet of Things for Smart Cities������������������������������������������������������������������������������������������������  263 B. Gopinath, M. Kaliamoorthy, U. S. Ragupathy, R. Sudha, D. Usha Nandini, and R. Maheswar Index������������������������������������������������������������������������������������������������������������������  275

About the Editors

R.  Maheswar  has completed his B.E. (ECE) from Madras University in the year 1999, M.E. (Applied Electronics) from Bharathiyar University in the year 2002 and Ph.D. in the field of Wireless Sensor Network from Anna University in the year 2012. He has about 19 years of teaching experience at various levels and is presently working as Dean-Research (Assistant) and Dean In-Charge for the School of EEE, VIT Bhopal University, Bhopal. He has published around 70 papers at International Journals and International Conferences and published 4 patents. His research interest includes wireless sensor network, IoT, queueing theory and performance evaluation. He has served as guest editor for Wireless Networks Journal, Springer, and is serving as editorial review board member for peer-reviewed journals. He has also edited 4 books supported by EAI/Springer Innovations in Communications and Computing book series. He is presently an associate editor in Wireless Networks Journal, Springer, Alexandria Engineering Journal, Elsevier, and Ad-hoc Sensor Wireless Networks Journal, Old City Publishing. M. Balasaraswathi  is an associate professor at the Graduate Program in Electronics and Communication Engineering (ECE) of Saveetha School of Engineering at Saveetha Institute of Medical And Technical Sciences (SIMATS), Chennai, India. She has completed her B.E. (ECE) from Madras University in 1999, M.E. (Applied Electronics) from College of Engineering, Guindy, Anna University in 2005 and Ph.D. in the field of Wireless Communications from Anna University in 2017. She has about 18 years of teaching experience at various xi

xii

About the Editors

levels. Her research interests include wireless communications, wireless sensor networks, IoT, signal and image processing, and nature inspired computing. She has authored and edited few books published by reputed publishers and coauthored and published research articles in science citation index journals, conference proceedings, presentations, and book chapters. She served as a member and committee chair of various international conferences and also co-authored 2 patents. She is serving as editorial review board member for peer-reviewed journals. She has received South Indian Women Achievers Award in 2019. Ravi Rastogi  received his Ph.D. degree in Computer Science and Engineering in December 2011 from Uttarakhand Technical University, Dehradun, Uttarakhand, India, M.S. in Computer Science degree in September 2005 from Fairleigh Dickinson University, Teaneck, New Jersey, USA, Bachelor of Science degree in 1999 from Ch. Charan Singh University, Meerut, Uttar Pradesh, India. He has worked as database developer, in Microsoft, Seattle, USA and ING Philadelphia, USA.  He has also taught and worked in Jaypee University of Information Technology, Waknaghat, Solan, Himachal Pradesh, India, for almost 4 years. He has also taught and worked as department coordinator in Sharda University, Greater Noida, Uttar Pradesh, India, for 4 years and Galgotias University, Greater Noida, Uttar Pradesh, India, for more than 2 years. At present, he is working in University of Bisha, Bisha, Saudi Arabia. His research interests are in the fields of interconnection networks, stable matching problems, data mining and intelligence in systems, Internet of Things and cloud computing. A.  Sampathkumar  received his Bachelor’s in Information Technology in 2009; Master’s in Mainframe Technology in 2012 and completed Ph.D. Degree in Anna University Chennai in 2019. He has 10  years of academic experience and is currently working as program chair for B.  Tech CSE with Specialization in Health Informatics in School of Computing Science and Engineering at VIT Bhopal University, Bhopal, Madhya Pradesh, India. He has been involved in organizing several International conferences and workshops. He was the Conference Technical Chair and Publication Chair of several IEEE conferences. He has published more than 16 articles in peer-reviewed journals, and has been reviewer for few reputed international journals and member of CSI, ACM societies. His research interest includes artificial intelligence, data mining, machine learning, wireless sensor networks, data analytic and optimization techniques.

About the Editors

xiii

G. R. Kanagachidambaresan  received his B.E. degree in Electrical and Electronics Engineering from Anna University in 2010 and M.E.  Pervasive Computing Technologies in Anna University in 2012. He has completed his Ph.D. (Wireless Body Area Networks, Healthcare) in Anna University Chennai in 2017. He is currently an associate professor, Department of CSE, Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology. His main research interest includes body sensor network and fault tolerant wireless sensor network. He has published several reputed articles and undertaken several consultancy activities for leading MNC companies. He has also guest edited several special issue volumes and books in SPRINGER and is serving as editorial review board member for peer-­ reviewed journals. He is presently working on several government-sponsored research projects like ISRO, DBT and DST.  He is the managing director for Eazythings Technology Private Limited.

Fog and Edge Computing for Automotive Applications Sandhya Devi R. Subburaj, V. R. Vijay Kumar, P. Sivakumar, B. Vinoth Kumar, B. Surendiran, and A. Neeraja Lakshmi

1  Introduction Traffic congestion is a worldwide problem. It impacts the transport sector’s efficiency by introducing delays that cause travel time increase and result in increased fuel consumption and associated emissions of vehicles [1]. This is anticipated to escalate considerably in severity due to the significant rise in the number of vehicles. To enhance the transport sector’s productivity and the related economic and environmental impacts, an efficient real-time approach is expected to be used in successful traffic management schemes that increase traffic flow and road infrastructure utilization.

S. D. R. Subburaj Department of EEE, Kumaraguru College of Technology, Coimbatore, India e-mail: [email protected] V. R. Vijay Kumar Department of Electronics and Communication Engineering, Anna University Regional Campus, Coimbatore, India P. Sivakumar (*) · A. Neeraja Lakshmi Department of EEE, PSG College of Technology, Coimbatore, India e-mail: [email protected] B. Vinoth Kumar Department of IT, PSG College of Technology, Coimbatore, India e-mail: [email protected] B. Surendiran Department of Computer Science and Engineering, National Institute of Technology, Pondicherry, India e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 R. Maheswar et al. (eds.), Challenges and Solutions for Sustainable Smart City Development, EAI/Springer Innovations in Communication and Computing, https://doi.org/10.1007/978-3-030-70183-3_1

1

2

S. D. R. Subburaj et al.

The Internet has become the world’s largest and most popular communication mechanism between people, companies, and governments. In addition, a variety of computers are connected to the Internet, collect and produce data, and provide a range of computer services. Many fixed or mobile computer systems and facilities today rely on centralized infrastructures, such as cloud storage, to store information and/or process data. Data centers will typically be further away from the application’s end-user, resulting in an increased delay in access to data and larger server turnarounds. The most suitable method to avoid traffic congestion and improving road safety is through the use of the Internet of Vehicles (IoV)-connected vehicle technologies. IoV is vehicle applications focused on the Internet of Things (IoT), developed to accomplish intelligent transportation (ITS) road safety and traffic control goals. The IoV solution uses a wireless sensor network (WSN). The WSN contains vehicle nodes that serve as sensory nodes as shown in Fig. 1. Stationary route points and service-related nodes function as roadside units (RSUs), which are the link to network infrastructure through a central server-operated operator node. For exchanging data between vehicular sensory interfaces and cars, the WSN uses vehicle-to-vehicle (V2V) communication and uses vehicle-to-infrastructure (V2I) to exchange data between car nodes and RSUs. Vehicular ad hoc networks (VANETs) are primarily cloud computing-based. The cloud offers unified processing and storage facilities via a cloud server or remote servers. Data collected can be recovered from any location without the need for large vehicles for storing and processing. Consumers will also exchange a significant amount of data between cars [2]. When dealing with cloud computing, one important impact is the delay in transferring data from cars to the cloud server, as well as in the processing. Because of this aspect and with a growth in the number of vehicles with improved mobility, technologies are required to ensure low latency and reliable services. Additionally, the connectivity between the vehicle and the cloud server requires high bandwidth. The high network load often induces more energy consumption in different wireless devices and has a significant effect on the cost of bandwidth. The number of smart IoV vehicles will increase drastically in the coming years. Therefore, careful consideration must be taken to effectively manage the IoV infrastructure. Thus, vehicular edge computing (VEC) emerged with the advent of new

Fig. 1  Components of vehicular WSN in IoV

Fog and Edge Computing for Automotive Applications

3

technologies in smart vehicles to satisfy the computing demands of the cars. VEC consists of computing devices residing on roadsides or in the vicinity of the vehicles. Such architecture helps vehicles to communicate, compute, and share data. It removes the communication delay with the cloud and is more useful for complex systems like IoV [3]. While edge computing can solve several IoT-related problems, there are some drawbacks to them. Edge node control is rather user-centric and only integrates sensitive fault-tolerant facilities. Fairness is also tedious in edge environments to ensure between multiple users. Fog computing overcomes these cloud and edge constraints by exploiting comparatively efficient user-level tools and reducing user-level pressures of resource and application service management.

2  S  cope of Edge and Fog Computing in Automotive Environment Vehicle networks demand that vast volumes of data be stored and analyzed in real time. The information being shared is time-sensitive due to high node latency, as there is restricted bandwidth in the communication channels. Also, the network needs to be aware of the context location. Through the year 2030, it is projected that up to 960 million operating connected vehicles will be processing huge bytes of data per day. The communication mechanism must therefore be robust for a technically feasible solution that meets the requirements of quality of service (QoS) and quality of performance (QoP) [1]. To ensure confidentiality and trust to transactions, fog and edge computing frameworks provide IoT systems with added encryption [4]. For instance, today’s wireless sensors installed in outdoor environments frequently need remote upgrading of wireless source code to address security-related problems. However, due to various environmental factors such as unstable signal strength, interruptions, constraint bandwidth, etc., the remote central backend server may face challenges to perform the update quickly, thereby increasing the likelihood of a cybersecurity attack. In comparison, the backend will configure the fastest route through the network across different nodes if the fog and edge computing (FEC) frameworks are used to easily upgrade the device protections for the wireless sensors. The FEC helps them to consider their consumers’ goals by promoting autonomous decision-­making as far as computation, storage, and control functions are concerned [5]. In turn, FEC’s understanding of many self-adjustment, self-organization, self-healing, self-expression processes, and so on transforms the function of IoT devices from passive to smart active devices that can continuously run and respond to consumer needs without relying on decision-making from the distant cloud.

4

S. D. R. Subburaj et al.

3  Internet of Vehicle (IoV) The Internet of Things (IoT) refers to the thousands of physical devices worldwide which are connected to the Internet and collect and exchange data. Connecting and attaching sensors to any of these devices brings a degree of artificial intelligence to machines that might help them to exchange real-time data without needing human intervention. IoV is the convergence of the Internet and IoT along with the cars, allowing smart connectivity between cars and other heterogeneous networks to offer a range of services to drivers and passengers. First, vehicles deliver their onboard units (OBU) for computing purposes [6]. The OBU will gain in particular from the data provided by embedded sensors (e.g., camera, GPS, radar) to help the driver determine. Sensors may, for instance, distinguish potential accidents (e.g., pedestrian or bike) with an obstacle. The driver should also be warned that unnecessary intervention is precautionary.

3.1  IoV Architecture As depicted in Fig. 2, there are several types of communication networks in IoV topology:

Fig. 2  IoV architecture

Fog and Edge Computing for Automotive Applications

5

• Vehicle-to-Vehicle (V2V): In V2V, the vehicles are wirelessly linked to exchange information and establish a 360-degree view of other surrounding vehicles. • Vehicle-to-Infrastructure (V2I): The V2I is known to be a network with two aspects with connectivity. In bidirectional communication, the infrastructure serves the car. In the intelligent transportation system (ITS), travelers are connected to the local VEC network available, e.g., roadside (RSU) module and base station (BS). • Infrastructure-to-Infrastructure (I2I): The I2I is defined by connecting infrastructures to promote computing, network load balancing, and knowledge sharing. In such a contact network, the components primarily consist of a series of BSs, RSUs, fog servers, and cloud servers.

3.2  Challenges Faced in Smart and Secure IoV Some of the challenges faced in implementing smart and secure IoV systems are: • Latency: The cloud faces the difficulties of meeting the requirement that end-to-­ end latency is regulated within tens of milliseconds. In particular, the reasons resulting from IoT’s latency cannot be addressed by urban smart grids, self-­ driving vehicle networks, interactive and augmented reality systems, real-time financial exchange apps, healthcare applications, and eldercare. • Uninterrupted: The long distance between cloud and front-end IoT devices can confront issues arising from unreliable and inconsistent connections to the network. For instance, due to the disconnection happening at the intermediate node between the vehicle and the distant cloud, an IoT-based vehicle would be unable to operate properly [7]. • Security: A large number of front-end constraint devices may not have enough resources to protect themselves against the attacks. Front-end outdoor devices that rely on the remote cloud to maintain them up-to-date with the applications may be targets for attackers, as attackers can conduct disruptive behavior on the edge network where the front-end devices are situated and the cloud does not have complete control over them. The intruder can even destroy or monitor the front-end computer, as well as submit fake data to the cloud [8]. • Resource-constrained: Most front-end devices are typically resource-constrained in which they are unable to execute complicated computing activities and thus IoT applications generally allow front-end devices to transmit their data to the cloud continuously. However, in many devices that operate with battery power, such a design is impractical, because end-to-end data transmission via the Internet can still consume a great deal of energy.

6

S. D. R. Subburaj et al.

4  Vehicular Edge Computing (VEC) Contrary to the clustered vehicular cloud computing infrastructure, VEC is looking for systems of widely dispersed implementations. Edge computing is a more suitable approach for the development of computational capacities in automotive environments. Through edge computing, data collection and review were carried out near the end computers. Edge serves as an intermediate between the cloud and cars. In the vicinity of the vehicle networks, servers with computational and storage capabilities (edge nodes) are deployed [9]. Because the processing and storage resources are provided near to the customer (on the bottom), the resources provide better QoS through edge computing. Besides, to help the contemporary applications in-vehicle networks, a powerful communication and computation mechanism is needed.

4.1  Edge Computing Architecture The VEC architecture can be comprised of three layers: cloud layer, an edge cloud layer, and smart vehicular layer as shown in Fig. 3: • Cloud Layer: The most significant benefits of the cloud layer include data collection, data extraction, database filtering, storage, batch processing, and dynamic data computation that goes beyond the edge node’s computational capacity. Also, the cloud can calculate in a very short time, even the gigantic amount of data and the complex computations [2]. There are two components of the cloud infrastructure: storage and computation. The data gathered from different edge nodes, which may be used for a later/longer duration and do not need real-time computation, will be sent to the cloud layer, where it will be permanently processed for future review. The computation component computes and analyzes the numerical complexity. • Edge Cloud Layer: The edge cloud layer guarantees that the smart vehicle layer is linked to the cloud network. The cars include systems that utilize wireless networking protocols, such as 802.11p, 3GPP, 3G, 4G, LTE, and 5G, to do this. The aim is to include low latency, position recognition, disaster response, routing, user discovery, and networking and to increase service efficiency because it is near to vehicles and is used for real-time interaction [10]. The edge cloud layer also deals with certain applications that require very low latency fast response. • Smart Vehicular Layer: Vehicles are expected to carry out more communication, share facilities aboard, and provide storage [11]. The smart vehicle layer is responsible for abstracting knowledge that vehicles can possess from embedded sensors, GPS, camera, radar, lidar, and other equipment. The knowledge gathered may be submitted for storage to the edge cloud layer, or it can act as feedback for multiple resources in the application layer. The vehicle would have connectivity, transport, knowledge, and learning capability in a smart vehicle layer to predict the driver’s intentions.

Fog and Edge Computing for Automotive Applications

7

Fig. 3  Components of edge computing architecture

4.2  Impact of VEC on Smart Vehicles VEC and the idea of utilizing smart cars as networks also opened up an environment with numerous relevant automotive technologies, such as driving safety, augmented reality, infotainment, and video streaming services as shown in Fig.  4. The VEC networks play a significant role in computing for applications where high numerical processing is the need, thus mitigating the delay as if an incident happened, it needs us to devise a solution for rescheduling traffic lights and effectively dissipating broad traffic backlog. It makes an exceptional need for computing resources [2].

5  Fog Computing 5.1  Overview Fog computing or fog networking, also known as fogging, moves limits of computing applications, data, and resources away from the centralized cloud to the network edge’s logical source [10]. Fog computing is a cloud extension to the edge of the

8

S. D. R. Subburaj et al.

Vehicle to Vehicle

Vehicle to Infrastructure SMBS

Vehicle to Pedestrian

Vehicle to Infrastructure

SRSU

Fig. 4  Communication in smart vehicle

network and brings computing resources and services closer to end-users. Such proximity provides some benefits such as decreased lag, which increases user experience.

5.2  Fog Computing Architecture Fog computation does not have a standard architecture, and study works sometimes focus on various architectures. Figure  5 provides a high-level description of the computational system for fog. It consists of three main layers, in the most typical scenario: cloud, fog, and end devices: • Cloud layer: The cloud layer is where actual data center nodes are put. Each node has a CPU(s), primary memory, and network bandwidth, which is used to fulfill resource demands from users. Clouds are linked to wide area networks (WANs) which have economic advantages, scalable infrastructure, data-­intensive end-user analysis [12], consistency of the services provided, and a broad degree of tolerance for faults. However, because cloud computing operates centrally, it is unable to support context-aware computing for IoT applications. • Fog layer: Any computer that is capable of downloading, saving, and linking to the network may be regarded as a fog processing system. Some devices can be considered as IoT and fog devices in this sense, and smartphones start the best example of this. The fog layer is a series of antennas, gateways, and networked equipment (routers and switches) positioned between the edges of the network and the clouds. Fog devices shape a distributed system that provides services at a specific location for a certain set of end devices and handles data that are transmitted by such devices [13].

Fog and Edge Computing for Automotive Applications

9

Fig. 5  Components of fog computing platform

• End devices layer: This ground layer covers end devices such as sensors, actuators, or items; handheld smartphones and desks, smart meters, aircraft, and smart cars; and desktop PCs. End computers may be viewed as human resources that have a variety of computational capabilities. Both of these components form a contact network, and their data is transmitted through the fog layer to the cloud. As shown in Fig. 5, the fog nodes must be mounted on such connected vehicles, such as self-driving buses and taxis, and the versatility of such nodes (called vehicular fog nodes) be used to provide the computation and communication resources where they are required. Such vehicular fog nodes along with the static fog nodes located at the edge of cellular networks (called cellular fog nodes) create a hybrid fog computing device. The fog computing platform supports automatic discovery and networking of vehicular fog nodes, capability registration and configuration, and provisioning of dynamic applications.

5.3  A  pplication and Data Management in Fog Computing Networks Fog computing provides a broad range of applications and network resources [14]. Infrastructure services provide computation, networking (bandwidth and firewalls), and storage infrastructure on-demand utilization, while application services enable runtime environments, operating systems, and programming interfaces. Fog

10

S. D. R. Subburaj et al.

Fig. 6  Hierarchy in application management for fog computing

Fog Application Management

Application Architecture

Application Placement

Application Maintanence

resource management denotes functional operations such as the installation, virtualization, and control of fog nodes that facilitate the network and application resources focused on fog. In turn, fog resource management implements load management, automatic provisioning, and auto-scaling to maintain multi-tenancy and continuity of operation [5]. An application management technique typically refers to a set of algorithms, mathematical models, scientific research, and guidelines that govern program development, deployment, and execution in a computer system. In addition, application management approaches incorporate entry monitoring, position accessibility, data maintenance, and process durability as per the respective device specifications [15]. An application management approach includes defining the characteristics of apps such as their development model, graphical architecture, type of operation, style of workload to be compatible with the network, and framework resources dependent on fog as shown in Fig. 6. This needs identifying appropriate putting solutions in fog environments for the applications. At the same time, a balance between application-­centric QoS requirements must be struck. In fog environments, it also requires facilitating security and resiliency support during application execution.

5.4  Simulation Tools for Fog Computing Several simulations of fog models were used in actual structures to determine or explain how such systems operate. Models include components that represent the various parts or states of the real system and the relationships between them that define the fog system’s behavior [16]. The core components for fog computing systems include the networks (i.e., the computers and the network), the architecture (which involves tools, utilities, and their management), and applications (which operate in the fog and have certain requirements). Some of the simulation tools for fog computing includes: • iFogSim is the first used simulator for fog computing [17]. It’s built-in Java as an extension to CloudSim15, the most common cloud computing simulator. iFogSim enables the validation of resource and device positioning approaches concerning traffic latency, device throughput, network use, energy consumption, and operational costs. iFogSim provides a GUI for describing topologies of the fog

Fog and Edge Computing for Automotive Applications

11

network (i.e., sensors, actuators, cloud data centers, and interconnections between them). • FogNetSim++ is a simulation system that can manage fog networks and computers [18]. This is focused on OMNeT++, a C++ simulation platform, and system built on open-source modules, commonly used in academia. FogNetSim++ is an event-driven simulator that aims to provide a static or dynamic environment that supports sensors, fog nodes, distributed data centers, and a broker node. The latter has the role of managing the other devices and their requests. FogNetSim++ does support resource execution. • FogDirMine [19] is a Python-based simulation tool designed to model the CISCO FogDirector behavior, which is a tool for managing IoT applications on fog systems. • ModFogSim is an extension of iFogSim that needs significant effort to model accessibility and migration approaches for users. ModFogSim [20] provides accessibility and synchronization features to iFogSim, as well as wireless links. As regards expense, it inherits the metrics of iFogSim, enhancements to the metric of latency.

6  Comparison Between Edge and Fog Computing Edge network’s computing components do not abound in processing, networking, and energy capacity. They are less effective [21] of operating large-scale and dynamic systems over a prolonged period. Fog computing overcomes these edge constraints by exploiting comparatively efficient user-level tools and rising user-­ level responsibilities of resource control and application support. Fog computing ensures smooth connectivity with cloud data centers and ultimately offers a comprehensive IoT device execution infrastructure [22]. Table 1 describes the significant variations between cloud, edge, and fog computing. In some cases, edge computation is considered a service model offered by various paradigms, namely, dew, mist, and fog computing. Dew computing occurs in IoT systems and mist computing takes place at the IoT gateways. However, fog computing is deemed highly feasible within such contemporary paradigms, owing to its widespread support for IoT applications.

7  Discussion Vehicle congestion identification and enhancement of road safety include contact-­ based approaches that rely on tactile network management and noncontact methods to ensure precision. IoV-based systems are ideally adapted for time-sensitive traffic congestion prediction applications involving high sensory data precision. Accuracy rates of traffic intensity calculations focused on V2V and V2I are within reasonable

12

S. D. R. Subburaj et al.

Table 1  Comparison among cloud, vehicular edge, and vehicular fog computing Features Location of data processing Main hardware required Latency Computing capabilities Storage capability Resource assignment Purpose

Security

Cloud computing Central cloud server Microcontroller

Vehicular edge computing Devices

Vehicular fog computing IoT gateway or fog nodes Single board computer

High Advanced processing capabilities High

Programmable logic controller Medium Low Limited processing power Limited processing power Low Medium

Dedicated

Shared

Shared or virtualized

Best suited for long-term in-depth analysis of data

Best suited for quick analysis for real-time response Increased security via encryption at source

Best suited for quick analysis for real-time response High

Low

limits. The specifications for IoV systems that collect large amounts of data for processing include low response time, low communication latency, mobility support, and location awareness. VEC- and vehicular fog computing (VFC)-dependent solutions meet the aforementioned criteria [11]. Nonetheless, the VFC-dependent solution utilizes lightweight algorithms and streamlined application connectivity to increase network response time and accommodate a larger range of apps. Extremely low contact latencies provided by the fog computing-focused solution with mobility and centralized infrastructure can be helpful for applications needing high-precision vehicle monitoring and power [23].

8  Conclusion This paper discusses and analyzes the importance of fog and edge computing in automotive environments. It provides a comprehensive overview of the theory, architecture, implementations, and applications within each of these facets. Cloud computing provides multiple benefits to the IoT platform for smart vehicles. This computing which centralizes the conventional approaches suffers from long latency and unreliable vehicular connections and may congest a large amount of data into the network backhaul. So, to avoid these limitations and to meet future possibilities, fog and edge technologies provide inevitable solutions to many IoT challenges occurring in the vehicular environment. Fog and edge computing is an advanced architecture that migrates the computing from the cloud to the edge of the network.

Fog and Edge Computing for Automotive Applications

13

Fog and edge computing can be a more appropriate platform to boost computing capacities in-vehicle networks through the intelligence that travels back to the vehicles. It allows processing to be carried out at roadside units, wireless access points, or base stations situated at the edge of the network.

References 1. A.  Thakur, R.  Malekian, Fog computing for detecting vehicular congestion, an internet of vehicles based approach: a review. IEEE Intell. Transp. Syst. Mag. 11(2), 8–16 (2019) 2. R. Naqvi, S. Salman, S. Wang, M. Ahmed, M. Anwar, A survey on vehicular edge computing: architecture, applications, technical issues, and future directions. Wirel. Commun. Mob. Comput. 2019, 1–19 (2019). https://doi.org/10.1155/2019/3159762 3. R. Mahmud, A.N. Toosi, K. Rao, R. Buyya, Context-aware placement of Industry 4.0 applications in fog computing environments. IEEE Trans. Ind. Inf. 16(11), 7004–7013 (2020) 4. R.S.  Sandhya Devi, V.R.  Vijaykumar, P.  Sivakumar, Neeraja Lakshmi A, Vinoth Kumar B, in Edge Architecture Integration of Technologies: Cases on Edge Computing and Analytics (2020), https://doi.org/10.4018/978-­1-­7998-­4873-­8.ch001 5. R.  Mahmud, K.  Ramamohanarao, R.  Buyya, Application Management in Fog Computing Environments: A Taxonomy, Review and Future Directions (University of Melbourne, Melbourne, 2020) 6. A. Hammoud, H. Sami, A. Mourad, H. Otrok, R. Mizouni, J. Bentahar, AI, blockchain and vehicular edge computing for smart and secure IoV: challenges and directions. IEEE Internet Things Mag. 3(2), 68–73 (2020) 7. H.A.  El Zouka, A secure interactive architecture for vehicular cloud environment, in Proceedings of the IEEE Conference on Smart Cloud (2016), pp. 254–261 8. M. Anwar, S. Wang, M. Zia, A. Jadoon, U. Akram, R. Naqvi, S. Salman, Fog computing: an overview of big IoT data analytics. Wirel. Commun. Mob. Comput. 2018, 1–22 (2018). https:// doi.org/10.1155/2018/7157192 9. J. Sun, Q. Gu, T. Zheng, P. Dong, Y. Qin, Joint communication and computing resource allocation in vehicular edge computing. Int. J. Distrib. Sens. Netw. 15, 155014771983785 (2019). https://doi.org/10.1177/1550147719837859 10. Y. Xiao, C. Zhu, Vehicular fog computing: vision and challenges (2017), pp. 6–9, https://doi. org/10.1109/PERCOMW.2017.7917508 11. Y. Ai, M. Peng, K. Zhang, Edge cloud computing technologies for internet of things: a primer. Digit. Commun. Netw. 4(2), 77–86 (2017). https://doi.org/10.1016/j.dcan.2017.07.001 12. M. Saad, Fog computing and its role in the internet of things: concept, security and privacy issues. Int. J. Comput. Appl. 180, 7–9 (2018). https://doi.org/10.5120/ijca2018916829 13. R. Hussain, J. Son, H. Eun, S. Kim, H. Oh, Rethinking vehicular communications: merging VANET with cloud computing, in Proceedings of the IEEE conference on Cloud Computing Technology and Science (2012), pp. 606–609 14. X.  Hou, Y.  Li, M.  Chen, D.  Wu, D.  Jin, S.  Chen, Vehicular fog computing: a viewpoint of vehicles as the infrastructures. IEEE Trans. Veh. Technol. 65(6), 3860–3873 (2016) 15. P. Hu, S. Dhelim, H. Ning, T. Qiu, Survey on fog computing: architecture, key technologies, applications and open issues. J. Netw. Comput. Appl. 98, 27–42 (2017) 16. C.-H. Hong, B. Varghese, Resource management in fog/edge computing: a survey on architectures, infrastructure, and algorithms. ACM Comput. Surv. 52, 1–37 (2019). https://doi. org/10.1145/3326066

14

S. D. R. Subburaj et al.

17. H.  Gupta, A.  Vahid Dastjerdi, S.K.  Ghosh, R.  Buyya, iFogSim: a toolkit for modeling and simulation of resource management techniques in the internet of things, edge and fog computing environments. Software: Pract. Exper. 47(9), 1275–1296 (2017). https://doi.org/10.1002/ spe.2509 18. T. Qayyum, A.W. Malik, M.A.K. Khattak, O. Khalid, S.U. Khan, FogNetSim++: a toolkit for modeling and simulation of distributed fog environment. IEEE Access 6, 63570–63583 (2018) 19. S.  Forti, A.  Ibrahim, A.  Brogi, Mimicking FogDirector application management. Softw. Intensive Cyber Phys. Syst 34, 151–161 (2019) 20. C. Puliafito, D.M. Gonçalves, M.M. Lopes, L.L. Martins, E. Madeira, E. Mingozzi, O. Rana, L.F. Bittencourt, MobFogSim: simulation of mobility and migration for fog computing. Simul. Model. Pract. Theory 101, 102062 (2020) 21. Y. Mao, C. You, J. Zhang, K. Huang, K.B. Letaief, A survey on mobile edge computing: the communication perspective. IEEE Commun. Surv. Tutor. 19, 2322–2358 (2017) 22. S. Margariti, V. Dimakopoulos, G. Tsoumanis, Modeling and simulation tools for fog computing—a comprehensive survey from a cost perspective. Future Internet 12, 89 (2020). https:// doi.org/10.3390/fi12050089 23. A.V. Dastjerdi, H. Gupta, R.N. Calheiros, S.K. Ghosh, R. Buyya, Fog computing: principles, architectures, and applications, in Internet of Things: Principles and Paradigms (Morgan Kaufmann, 2016), pp. 61–75 R. S.  Sandhya Devi  received her B.E. (Electronics and Communication Engineering) from Avinashilingam University, Coimbatore, Tamilnadu, India. She completed her M.E. (Embedded Systems) in Anna University of Technology, Coimbatore, Tamilnadu, India. She is Asst. Professor at Kumaraguru College of Technology. Her areas of interest include embedded system design and ARM processor. She published her project in a journal and international conference.

V. R.  Vijay Kumar  is currently working as an Associate Professor in the Department of Electronics and Communication Engineering, Anna University Regional Campus, Coimbatore. He received his Ph.D. degree from Anna University Chennai in the area of nonlinear filtering and masters and bachelors from Thiagarajar College of Engineering, Madurai, and Thanthai Periyar Govt. College of Technology, Vellore, respectively. He has 20  years of teaching experience, and his area of research includes image processing, signal processing, and VLSI design. He has published more than 85 research papers in international journals and conferences.

Fog and Edge Computing for Automotive Applications

15

P.  Sivakumar  received his B.E. degree in Electrical and Electronics with I class in 2006 from Anna University. He completed his M.E. degree in Embedded System Technologies with I class in 2009 from Anna University Coimbatore. He completed his Ph.D. in Electrical Engineering with a specialization in Automotive Embedded Software in the year 2018 from Anna University, PSG College of Technology. His research interests include embedded system, model-based design, model-based testing of automotive software, automotive software development, and fog and edge computing. He has around 14  years of teaching experience. He has published 14 papers in reputed international journals. He has also published 20 national and international conferences papers and has organized national seminar-workshop funded by DRDO, DST, and MNRE. B.  Vinoth Kumar  is working as an Associate Professor with 16  years of experience in the Department of Information Technology at PSG College of Technology. His research interests include soft computing, blockchain, and digital image processing. He is the author of more than 26 papers in refereed journals and international conferences. He has edited three books with reputed publishers such as Springer and CRC Press. He serves as a guest editor/reviewer of many journals with leading publishers such as Inderscience and Springer.

B.  Surendiran  is currently working as Associate Dean Academic and Assistant Professor in the Department of Computer Science and Engineering at National Institute of Technology Puducherry, Karaikal. He had completed his Ph.D. from National Institute of Technology Trichy. He has more than 30 publications in international conferences and journals. He had reviewed more than 200+ papers for various journals and conferences. His research interests include medical imaging, machine learning, dimensionality reduction, and intrusion detection systems.

A. Neeraja Lakshmi  is pursuing her master’s degree at PSG College of Technology, and she has completed her B.E. in Electronics and Communication Engineering from Saranathan College of Engineering in the year 2017. She has published papers on the automotive domain at international conferences. Her main research interest includes IoT, in-vehicle software development, and autonomous or connected vehicle.

Intelligent Transportation System in Smart City: A SWOT Analysis Tejas Parekh, B. Vinoth Kumar, R. Maheswar, P. Sivakumar, B. Surendiran, and R. M. Aileni

1  Introduction For quite a long time, transportation has been viewed as a connect to all parts of life around the globe. The world’s regular habitat, social government assistance, and monetary improvement for the most part rely upon transportation frameworks. By and large, sheltered, perfect, supportable, and evenhanded transportation frameworks help nations, particularly urban communities and urban focuses, to flourish [1]. One of the primary gear driving consequences is stable transportation, a pattern that has been around for quite a while. Accomplishing economical advancement in savvy urban communities requires the reconciliation of solid transportation frameworks. The driving variable of keen urban communities is to guarantee the association of human capital, framework, and social money to accomplish more practical monetary development and a better lifestyle for those living in these cities. T. Parekh · B. V. Kumar (*) Department of IT, PSG College of Technology, Coimbatore, India e-mail: [email protected] R. Maheswar School of EEE, VIT Bhopal University, Bhopal, India P. Sivakumar Department of EEE, PSG College of Technology, Coimbatore, India e-mail: [email protected] B. Surendiran Department of CSE, National Institute of Technology, Pondicherry, India e-mail: [email protected] R. M. Aileni Faculty of Electronics, Telecommunication and Information Technology, Politehnica University of Bucharest, Bucharest, Romania

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 R. Maheswar et al. (eds.), Challenges and Solutions for Sustainable Smart City Development, EAI/Springer Innovations in Communication and Computing, https://doi.org/10.1007/978-3-030-70183-3_2

17

18

T. Parekh et al.

Vehicles, trucks, and other rough terrain vehicles make up just a little piece of the transportation framework, including trains, planes, bikes, and vessels. Smart cooperation between these frameworks is required and is just conceivable through a genuine start to finish ITS [2]. This technology serves as a “multi-model coefficient” of transportation for smart cities and enhances the public transportation impact. Digitizing public transportation with real-time information, mobile apps, and other technologies will improve the customer experience and allow passengers to make more informed and efficient mobility decisions. Increasing environmental awareness, technology, increasing life expectancy, and long-term working families are contributing to fundamental changes affecting the mobility of millennials entering the workforce. The city’s transportation system serves as a lifeline to the city’s delicate performance. When there are no proper incoming channels, life stops for individuals living in urban territories. Legitimate courses and the executives of transportation characterize the personal satisfaction in present-day innovative urban communities. The ITS will change the manner in which individuals travel in metros and shrewd urban communities. ITS offers novel methodologies in giving various methods of transport, propelled foundation, traffic, and the versatility executive arrangements. It utilizes many savvy, remote, and correspondence advances to give clients access to shrewd, protected, and quick travel. The important features of the intelligent transport network and the advantages of the ITS are discussed in detail. The government, transport operators, and service providers should cooperate in the planning and implementation of smart transport. Transforming existing transportation systems into different touch points and activities—with the command to implement vehicle tracking and emergency response systems—it is important to implement smart payment systems or go for green options. Implementing mobility as a service (mass) in Finland is a good example of how a public-private effort can ensure model success at all levels. Proper planning is important, and countries need to spend enough time developing a comprehensive and comprehensive plan. National and local administrations should be careful to get involved quickly in smart transport technology or pit planning. Despite the many technical solutions, each city has its own specific complexities and problems. This challenge is well suited for a city that addresses all the core goals of smart transportation—a convenient, green, secure, integrated solution. In addition, smart transportation should be integrated with the city’s “smart” plans so that development is sustainable and profit is maximized. Implementing smart city transportation as a whole is a complex approach. Internet of Things is the significant innovation in technology, which leads to existence of smart city projects. IoT devices, sensors, and “contents” of applications make data that make technical solutions viable. For instance, the new water meters (smart) report the quality of water and utilization and alert a leaking water company or potential contamination [3]. Big data also has a significant job in city management, and now most urban areas are the data executives in the organization. The mix of analysis of big data and smart city solutions can assist urban communities with improving administration in basic critical zones.

Intelligent Transportation System in Smart City: A SWOT Analysis

19

As the city’s population explodes and private car ownership grows, the city’s transportation becomes unstable—and not just from an environmental point of view. If the current trend continues, economic growth and quality of life will also be severely affected. These can appear as smart transport framework, versatile applications, and computerized administrations, which take into consideration more effective utilization of existing methods of transport and totally new methods of transportation, for example, self-driving vehicles. SCT can have a major effect in the manner suburbanites travel in thickly populated urban territories and assist regions with sparing expenses, offer better types of assistance to residents, and better oversee security and well-being.

2  Intelligent Transportation System (ITS) The ITS is an optimal transport management and service framework that plans to offer inventive types of assistance identified with various methods of transport. ITS consolidates high innovation and upgrades in data frameworks, information transmissions, sensors, regulators, and progressed scientific techniques with customary vehicle frameworks, and it is the most significant element of ITS [4]. At the point when the transportation framework is coordinated into foundation and vehicles, these innovations decrease clog, improve security, and increment efficiency [5].

2.1  ITS Structure It has a multilevel structure: • • • •

Physical layer Communication layer Operation layer Service layer

This four-tiered structure guarantees connectivity, manufacturing, and services. All those layers support each other. 2.1.1  Physical Layer The physical layer contains all aspects of the transportation framework, including infrastructure, vehicles, and people. With the development of information technology, almost any kind of agent [6] can be seen as: 1. Able to experience their environment. 2. Have some control over their actions.

20

T. Parekh et al.

3. Interaction with other agents. It provides the ability to collect basic traffic data and respond to environmental changes. Data on the physical layer is collected through generalized sensors and their platforms such as roadside sensors, onboard sensors, and online social media platforms. They all reflect traffic conditions from some angle. 2.1.2  Communication Layer ITS provides a communication layer for proper and timely communication between subsystems. According to the American National ITS Architecture, the communication layer provides four main types of communication: • Field-vehicle communication. Communication between vehicles and instant infrastructure • Fixed point-fixed point communications. Communication between static entities • Vehicle-to-vehicle communication. Short-distance wireless communication link between vehicles • Wide-area wireless (mobile) communication. Communication systems that enable communication with vehicles and passenger mobile devices The combination of those options goes through the gaps in the physical world and supports: (a) Data exchange within the physical layer (b) Data exchange between the physical layer and the operation layer

2.1.3  Operation Layer The operation layer that is the backbone of the ITS architecture collects data and translates it into information and knowledge. Information collected from all elements of the transportation system is disposed of or distributed [7]. The results of data disposal are in response to the physical layer as services at the service level. Operation layer consists of three basic components: Advanced Transportation Management Systems (ATMS), Advanced Traveller Information Systems (ATIS), and Advanced Vehicle Control System (AVCS). ATMS refers to the overall system management. ATIS provides information to passengers. AVCS is a new level of control technology applicable to vehicles and infrastructure. They all form the minimum operation function; in other words, they are the core of the transport operation, which is emphasized in it. In the structure of the US National ITS, ATMS and ATIS play an important role in the operation of all its applications, interacting directly with them.

Intelligent Transportation System in Smart City: A SWOT Analysis

21

2.1.4  Service Layer Service layer is where services are deployed and run. The result of operation layer will be combined in order to provide better transportation services. The user of it might be the public or a system operator. The American National ITS Architecture defined user service bundles, which consist of 33 user services. This classification may not cover every requirement, but it meets the demand of daily travel. ITS improves transportation well-being and versatility, decreases natural effect, advances feasible transportation advancement, and upgrades efficiency through the coordination of cutting-edge correspondence-based data and electronic advances into all transportation components.

2.2  Autonomous Vehicles A completely autonomous vehicle can be characterized as a vehicle which can see its condition, choose which path to take to its goal, and drive it. As it is where, we can say self-sufficient vehicles are smart vehicles or robocars which utilize an assortment of sensors, PC processors, and information bases, for example, guides to assume control over a few or the entirety of the elements of driving from human administrators. Vehicles furnished with this innovative technology will have its own advantages. It will probably diminish crashes, vitality or energy utilization, and significantly contamination. As of late significant original equipment manufacturers (OEMs) have declared their arrangements to start selling such vehicles in a couple of years from now. 2.2.1  Classification of Autonomous Vehicles There are numerous OEMs which have just executed this into their leaving creation vehicles to think of a model for testing reason. This incorporates numerous prestigious automakers. This is classified into four unique levels [8]: 2.2.1.1  Level 1: Operation Specialized Automation It incorporates computerization of explicit control operations, for example, voyage control, path direction, and programmed parking. The drivers are completely connected with and liable for the whole vehicle control.

22

T. Parekh et al.

2.2.1.2  Level 2: Joint Operation Automation It mirrors the mechanization of many incorporated control functions, for example, perfect journey control with path focusing. Drivers are liable for street observing and are required to be accessible for control consistently, yet might be uprooted from vehicle activity in specific situations. 2.2.1.3  Level 3: Customized Automation Drivers can play out all well-being basic undertakings in specific circumstances and depend on the vehicle to screen changes in circumstances that require a progress back to driver control. Drivers don’t hope to continually screen the street. 2.2.1.4  Level 4: Complete Self-Driving Automation Vehicles can play out every driving obligation for the full voyage and screen the state of the streets and in this way work with the individuals who can’t drive and work without human proprietors. In any case, numerous automakers have begun testing their models yet at the same time it’s far to go to achieve that exactness and certainty where we can indiscriminately put our confidence on autonomous vehicles. The company Google has set an objective of 2018 to industrially dispatch its self-driving vehicle. These autonomous vehicles will have a drawn-out effect on the general public, and it will be a special change on how we drive. In any case, this change from customary to automated vehicles ought to be a progressive one with the goal that individuals will have that certainty and it tends to be utilized everywhere scale. 2.2.2  Fundamental Objectives of Autonomous Automobiles In this section the main objectives of autonomous vehicles have been discussed. The main objectives of an autonomous vehicle are listed below [8]: (a) (b) (c) (d)

Recognition Movement planning Movement navigation Performance

(a) Recognition It is the capability of the vehicle to comprehend its prompt condition. This will assist the vehicle with avoiding impact among vehicles and furthermore to keep a beware of any sort of obstructions which may appear in the vehicle way. Presently a day there are numerous electronic gadgets accessible in the market which can be utilized for this reason. LIDAR utilizes laser assortment locaters

Intelligent Transportation System in Smart City: A SWOT Analysis

23

to emanate light beams over hindrance removal and compute flight time until the picture is reestablished to mirror the article’s climate. Unlike LIDAR, RADAR give radar frameworks different abilities and constraints utilizing radio waves. Other regular gadgets utilized in self-driving/self-­ sufficient vehicles for mindfulness are monovision cameras. These cameras imply that they have a wellspring of vision. These cameras are straightforward gadgets and video takes care of are ordinarily used to decide fundamental situations, for example, path checking and speed limit signal. Car monovision cameras are less modern and have a lower pixel thickness than cameras in cell phones. (b) Movement Planning Movement includes performing lower-level activities toward accomplishing an elevated-level objective. It’s mind-boggling to design the way of a self-ruling vehicle in a powerful domain, particularly when the automotive is required to use its full abilities. There have been ongoing advances in computational abilities both as far as equipment and calculations and correspondence models which can help us in making a mistake-free self-ruling vehicle. Movement planning comprises of path factors which are to be controlled to keep away from any sort of setback. Those way factors are:

1. Vehicle’s steering (direction): The movement planning ought to be sufficiently proficient to direct the vehicle through static or dynamic automotive traffic. 2. Speed: Movement planning module ought to be productive enough to have control over the speed of the automotive as per surroundings. In the wake of considering previously mentioned boundaries, self-sufficient vehicle ought to have the option to adapt with these boundaries and produce neighborhood ways to be followed.

It ought to likewise have the option to allocate expenses to paths dependent on time consumed, fuel utilization, good ways from hindrances, and different imperatives. When all the accessible ways are acquired, it ought to pick the best way based on time (schedule), expense, traffic, and different limitations. (c) Movement Navigation Vehicles additionally utilize sensors which suit localization, for example, following one’s position on the planet Earth. Localization utilizes the GPS. Automotive GPS get signals through satellites revolving in orbit to locate their coordinates. The acquired coordinates are cross verified with street maps to find vehicles on streets. (d) Performance Once the vehicle has perceived its environment, completed the movement planning, and navigated the route, it’s time to act. So, on the basis of all detailed parameters, an autonomous vehicle takes decision. There are many challenges faced by an autonomous vehicle; few of them are listed and discussed:

24

T. Parekh et al.

1. Pathway Analysis In the vision-based path identification, several strategies have been introduced in the past. These strategies utilize various sorts of path designs/patterns, an alternate kind of path model, and furthermore different sorts of methods. Until now self-ruling vehicles have only been driven at streets which are very much arranged and organized. The abilities of this vehicle on the unstructured street are unknown. 2. Surpassing Surpassing has consistently been related as the primary driver of mishaps around the world. In a self-ruling vehicle, the worry turns out to be high. Surpassing in a self-ruling vehicle can be achieved by creating programmed calculations. The hardware gadgets fitted in the vehicle which chip away at the rationale given in the calculation to finish its given task. In the beginning, the source vehicle with the assistance of sensors recognizes the automotive ahead and gauges the separation between them [9]. With the assistance of this, it gauges its relative speed. Sound system-enabled cameras are utilized to recognize the moving articles which utilized edge discovery innovative technology. On the off chance, during surpassing if any vehicle comes toward, the source vehicle identifies and moves the vehicle to safe separation and overrides the choice of surpassing. There is by all accounts a great deal of potential for self-governing vehicles across the fields. In any case, the key lies in the sheltered execution of this innovation. Furthermore, safety must be accomplished by implementing exacting guidelines and norms. It requires some investment for individuals to believe a totally new innovation, and one single mishap can totally cut down the notoriety of this industry. But again, this innovation will be convenient in tackling many traffic-related issues, for example, parking and gridlock and mishaps by impressively lessening the movement stress.

2.3  ITS Algorithm 2.3.1  Approach and Methodology This subsection depicts how the ITS functions in algorithmic documentation [10]. The required database in the server contains traffic lights, IP, statues, streets, and position. 2.3.2  Algorithm   1: While true   2: A car sends a packet that includes {IP address, GPS position, impact sensor and accelerometer sensor fields reading) every 2 s to the server

Intelligent Transportation System in Smart City: A SWOT Analysis

25

  3: The server increments the street counter according to the received GPS location   4: If the fields of the accelerometer and impact is false then   5: Check the count of cars at the intersection.   6: If count of cars is less than or equal to 82 then   7: Prioritize roads according to the highest counters and calculate the time when the traffic light will stay turned on.   8: Else if the number of cars is greater (or more) than 82 and for more than 90 s one of the traffic lights has been turned off then   9: Turn On traffic lights that have been turned off for too long. 10: Else 11: set the priority asper the street counter. 12: End If 13: Else If fields of impact or accelerometer are true then 14: To the nearest police station send the accident packet and also send it to the hospital, contains accident street position 15: Mark the particular street as an accidental street 16: If minimum of one lane of all lane’s is open and is safe for passing the vehicles then 17: The police send a packet to the server. 18: delete the mark from street and then return it back to the calculation methods 19: Else 20: wait for the police’s packet to arrive and then open the street 21: End if 22: End if 23: Compare the specific GPS coordinates and decrease street counter by 1. 24: Return back to step 2. The ITS can be extended to include dealing with the lanes such that they manage the control of traffic lights and go across streets at the same time with traffic lights in the whole city so as to make it simple and snappy for crisis vehicles to arrive at the goal as quick as could be expected under the circumstances.

2.4  Smart Public Transportation Public transportation is a service shared by the general public for transportation. Unlike modes of transportation such as carpooling, rickshaws, and taxis, strangers share the system here. The reason people leave after public transportation is its cost, environment, and its role on accessibility. Above all else, PT is efficient. Less utilization of transports such as buses, railways than utilizing private vehicles. In the event that individuals own a vehicle, they should pay a ton of cash for services, fixing, and protection. It costs all the cash individuals acquire. What’s more, there are numerous special cases for certain people, for example, students, the highly aged people, and kids. The other thing is PT can secure the earth. This reduces pollution

26

T. Parekh et al.

Fig. 1  Functionality block diagram for smart PT system

and reduces smoke and reduces traffic jams as fewer cars move on the road. Also, large cities have good access to public transportation. Using public transportation can save both money and time lost in traffic jams. The functional diagram for smart public transportation is shown in Fig. 1. 2.4.1  Explanation All the steps will be in the order of execution based on the sequence shown in Fig. 2 [11]: • ARM (LPC2138): It acts as a CPU unit. It generates information for each sensor and displays it on the LCD and the reputable output gadget. It must be used as a result of the N number. The sensor is connected to the controller. It was chosen because of its ability to process parallel information and time to organize matches within the framework. This controller has a three-saturated pipeline which helps in brisk procedure. Since it has 40 pins of GPIO, it has the flexibility to attach (2 ^ 40) number of sensors. Here, it takes cooperation from various sensors (simple and sophisticated) and prepares it for the necessary yield. • Switch: Switches are utilized for various applications. This is a four-leg press-­ button switch. On its first “release” and “when pushed (pressed),” it is considered as input. The different applications utilized for switches are crisis switches, rash driving switches, incline switches, area switches, and framework on/off switches. This switch performs a constraint function for the controller, and the controller satisfies the need for a server switch. If its field changes to the following, then it acquires the territory and longitude of that zone to that specific second and shows it on the LCD screen display and yields extraordinary applications to follow this region at this specific time.

27

Intelligent Transportation System in Smart City: A SWOT Analysis

LM35 (bus temperature monitoring)

Ramp switch (on/off for handicaps)

Accelerometer(accident detecon and acknowledgement to hospital through GSM)

MQ3 (ALCOHOL DETECTION FOR DRIVER)

ARM Controller (LPC2138)

PIR(To Count No. Of Persons In And Out Of Bus)

Switch press(GSM acknowledgement to respecve registered nos). DC motor on/off switch

RFID Tag authencaon

Fig. 2  System execution process and methods Fig. 3  A PIR sensor unit

PIR Sensor input (analog)

ARM

• PIR sensor unit: This unit goes as contribution to the processor as appeared beneath in Fig. 3. It is associated with the ADC input stick of the regulator. As this sensor is a very basic sensor, it contains general data and offers those to control unit. By then, the control unit turns into a basic signal to the propelled ones and assists with indicating it on the LCD display unit and simultaneously depends on the person’s smart application to various access zones. 1. Accelerometer Unit: This unit is utilized for crash identification and receipt system. As it is an analog sensor, the control unit input will also be analog. Shocks/vibrations are received as input and then send it to the controller. It is programmed in a way that it detects 700 hazards on X- and Y-axis. The analog input is digitized by the controller, and on the GSM port and also on LCD display, the output is produced as shown in Fig. 4, respectively. 2. Alcohol-Level Sensor: This sensor is attached with ARM’s ADC port as shown in Fig. 5. As it is an analog sensor, it is also otherwise called as a carbon dioxide/a gas sensor. It distinguishes the change of % liquor noticeable all around. It goes about as the controller’s input and changes over the levels recognized by the controllers into computerized esteems and shows as yields on LCD unit and GSM frameworks. Alcohol detection % level is set at 80 mg. 3. Temperature Sensor Unit: It is commonly utilized to identify ambient temperature. It is a simple analog sensor, associated with the controller’s ADC port. The input of this sensor goes about as the input of the analog temperature value and the sensor controller. At that point, the controller changes the analog value of temperature to digital value and then displays it continuously on the LCD unit.

28

T. Parekh et al.

Accelerometer Input (analog)

ARM

Output to LCD and GSM (digital)

Fig. 4  Accelerometer unit (ADXL335)

Alcohol Sensor Input (analog)

ARM

Output to LCD and GSM Unit (digital)

ARM

Output to LCD and GSM Unit (digital)

Fig. 5  Alcohol sensor unit (MQ3)

Inputs from Various sensors Fig. 6  (SIM900) GSM/GPS unit

4. GSM or GPS Unit: They are associated with controller’s GPIO port as shown in Fig. 6. The GSM unit is used to accept numbers registered to the system via SMS. And GPS frameworks are utilized to collect ongoing coordination for the framework. These are directly connected to the resultant output of whole system. This unit provides output for the accelerometer unit, liquor-level sensor, harsh driving switch, crisis switch, position monitoring switch, and the PIR sensor unit, individually. This port is associated with the framework (system), individually. 5. RFID Scanner: It is utilized as an authenticator for the framework. The provided input to scanner will be RFID card. The scanner deciphers the sequential no. of card, and if it matches the code, then it gives us validation to work on the system, until which the system won’t start. To make the validation noticeable, a LED is mounted to gleam. 2.4.2  Methodology This section considers the five examples and plans for the hardware requirement, software, as well as software development and testing: (a) Compared to Indian technologies. As of 2010 there is a fully mechanized public transport. Speaking specifically about the bus as a transportation system, so far, the mechanical system is still used. (b) In any case, as technology progresses, there are critical changes that traditionally happen once. For the former, private transportation technology is far better than the public transportation that once existed with the government. Among them are automatic lighting control, GPS tracking, Wi-Fi, AC, TV unit, and anything.

Intelligent Transportation System in Smart City: A SWOT Analysis

29

(c) The public transport system is still mechanized. Now government buses are also being developed in Pune, Maharashtra. For the former, Volvo has similar features. There are a digital display and emergency gate AC, for announcements and the next stop signal. (d) Gujarat, Karnataka, West Bengal, Punjab, and other states have more developed public transport system than Maharashtra. • Hardware –– The equipment utilized is the fundamental innovation GPS or GSM innovation for transport area tracking and observing. –– Accidents discovery and GSM innovation ought to be utilized to screen vehicle mishap. –– PIR sensors are utilized to tally the quantity of individuals going in the transport and through the secondary passage. A temperature measuring sensor is utilized to keep track the temperature of the transport. –– The security switch with the SMS receipt to Ra enrolled police headquarters is utilized as a pointer for reckless driving, transport document, and crises. –– RFID card is utilized for verification of driver, and slopes are utilized to make it helpful and simple for individuals with incapacities to utilize public transportation. • Utilization of Software –– Track Bus needs to develop an Android application for smartphones to make tracking and monitoring easier and faster. –– The embedded C language controller is used for programming, i.e., for the ARM7 processor. • System Development and Testing –– For equipment: Flash Magic for program consuming, Dip follow for PCB planning, and an Android emulator for application testing –– Keil 4: Utilized for program investigating and simulation –– Java IDE platform: Utilized to build up the Java condition for sequential correspondence among models and PCs –– MySQL language: Utilized to make a database for versatile applications 2.4.3  Algorithm for System Execution This algorithm is divided into three parts: prototype setup, hardware algorithm, and software algorithm which overall combine to form an algorithm for execution of system [11]: • Setup for Model Prototype 1. The equipment model must be associated with the connector. 2. The antenna/receiving wire must be associated with the GPS unit, and the reception apparatus is free in the sky.

30

T. Parekh et al.

3. Associate the model to the PC. 4. Open the Java IDE arrangement. • Algorithm for Demonstration of Hardware 1. Initiate the equipment. 2. Give admittance to the framework utilizing RFID unit. 3. Start LCD and sensors. 4. LCD Display: Start estimation of temperature sensor, accelerometer esteem on X-hub, liquor sensor worth, and PIR sensor figuring. 5. On the off chance that the framework presses the on/off switch, start the framework second. 6. At the point when the estimation of the MQ3 sensor changes, for example, the estimation of the liquor content changes, the GSM framework sends a SMS receipt to the enrolled number. The driver is flushed and has ongoing coordination of position. Liquor substance can be changed by splashing scent or physically moving the liquor through the sensor. 7. At the point when the accelerometer vibrates or yanks, it identifies a mishap when the x-hub and y-hub turn above 300 g and 700 g, individually. The GSM unit at that point sends a receipt SMS to the enrolled portable number. Contains a message of continuous coordination of danger identification and area. 8. In the event that the area following switch is squeezed, it will follow the ongoing area to follow the framework. This area is shown on the guide in the brilliant application. 9. However, PIR sensors are introduced at the framework passageway and exit for singular count. An expansion in the quantity of individuals at the passage and an expansion in the individual at the leave door. At that point, the measure of additions and decrements is shown in the shrewd application. 10. At the point when the slope switch is squeezed, the incline opens and closes. 11. At the point when the emergency and rash driving switches are squeezed, they acknowledge SMS through GSM unit with GPS coordinate to enroll. • Algorithm for Software Demonstration 1. Open eclipse environment. 2. Select the main page and run it as a Java application. Then, select the com port to which the system is connected via USB port. 3. Set all the features, i.e., Bad Rate, Parity Bits, and Enable and Disable Bits accordingly. 4. It will not be displayed as soon as the features are set. When available people and the location tracking switch is pressed, it returns the position on the main page as latitude and longitude value. 5. Then, run the Run on Main project on the server and then complete it by selecting the Tomcat v8.0 server on the local host. 6. Then, the application page opens in the browser. 7. On the homepage, search options for search are provided.

Intelligent Transportation System in Smart City: A SWOT Analysis

31

8. When searching for a bus, it displays the description and location of the bus to track. 9. When the track is clicked, Google displays the current real-time location of the bus on the map. The designed system is a completely secure and smart support public system. System implementation ought to be accomplished for the transport. The ARM7 processor is utilized as a regulator to control the whole preparing [12]. This framework outperforms the essential mechanical, Volvo, and BRT frameworks. It is safer, keen, and modern. At the point when a crisis switch is included, the transport file switches and the accident recognition framework gets secure. The framework is brilliant and modern as it has different highlights like liquor recognition, GPS following, GSM receipt, and incapacitated inclines.

3  SWOT Analysis 3.1  Strength The ITS joins a wide scope of data and data innovation to make a system of frameworks that help oversee traffic, secure streets, and few more. Since transportation organized turns out to be more arranged network, ITS will modify the way drivers, organizations, and governments approach street transportation. These propelled frameworks can help in betterment of transportation in various manners. 3.1.1  Improves Traffic Well-Being (Safety) Hazardous speed, risky climate conditions, and overwhelming traffic would all be able to prompt mishaps and death toll; ITS help with the entirety of this. Ongoing climate observing frameworks gather perceivability, wind speed, rains, and street conditions, and the sky is the limit from there, giving traffic regulators minute data on driving conditions and status. On completely networked frameworks, this data can be utilized to keep on refreshing warning signals also to restrain the speed when vital, to make drivers aware of the conditions and situations around them. Crisis vehicles can react rapidly to mishaps since real-time ongoing traffic observing cautions them. ITS traffic light control assists with redirecting traffic from occupied or hazardous zones to maintain a strategic distance from gridlocks yet lessen the danger of mishaps.

32

T. Parekh et al.

3.1.2  Diminishing Infrastructural Harm/Damage Substantial vehicles put a great deal of focus out and about the road system, particularly when they are overburdened. Weighing stations and different other weight managements lessen the danger of overburdening, however wastage of time and the traffic delays. Wave-in-motion frameworks check the sort, size, and the capacity or weight of automotive as they are in motion and communicate the gathered information back to a central server. Overburden automotive can be recognized, and fitting move is made, which can prompt congestion and harm to streets in hauliers. These frameworks can streamline usage as well as diminish the expense of street fixes, permitting it to be designated somewhere else. 3.1.3  Traffic and Traffic Light Control Existing brought together traffic signal frameworks move in a manner to decrease gridlock and guarantee smooth development of vehicles through the street organized. Wise transportation frameworks, be that as it may, permit traffic signals to react to evolving strategies. Versatile traffic signal frameworks make shrewd crossing points that oversee traffic because of the types of automotive they use. Likewise they may lean toward explicit types of traffic, for example, crisis vehicles or PT. The enormous amount of custom crossing points cooperate to make a system where the lights modify because of traffic designs over a set timeframe, lessening weight time and moving traffic effortlessly. 3.1.4  Vehicle Parking Administration Illicit parking blockage adds to perilous city roads and makes get to issues for parking spots reserved for disabled drivers, city vehicles, and others. Unnecessarily bustling drivers hinder traffic to slither in occupied regions as guests can’t park themselves. Traditional way of parking systems is expensive and disorganized; they also participate in the masses themselves. Scan parked vehicles to detect smart parking violations, and interact with the placed parking meters to distinguish and record unlawfully left vehicles. Drivers realize that and instead of taking risk with the parking officer (human), drivers will be consequently referred to for illicit or broadened parking. These ­programmed systems help increase driver compliance and improve traffic flow through the smooth turnover of parking spaces.

Intelligent Transportation System in Smart City: A SWOT Analysis

33

3.1.5  Collecting Traffic Data Without accurate information on policies of street use, legitimate traffic flow planning is incomprehensible. Already existing traffic sensors can get familiar with a great deal about what number of vehicles utilize a specific vehicle or convergence; however ITS can do significantly more. Electronic traffic counters can gauge the number and sort of vehicles going to and fro in a particular territory of the city, just as the most extreme traffic. Electronic traffic counters can quantify the number and kind of vehicles going to or fro in a particular zone of the city, just as the greatest traffic times, travel length, and other information. This data causes officials to distribute their assets in the most conceivable manner. As the developing populace of urban drivers and suburbanites squeezes our organized street, urban areas need better than ever gear (equipment) in the progressing battle to diminish gridlock and protect drivers. The ever-developing ITS insurgency offers another perspective about traffic and street management.

3.2  Weakness Autonomous vehicles have many benefits such as reduced driver stress, increased mobility for nondrivers, increased safety, increased fuel efficiency, and reduced pollution. But there are still many challenges that serve as barriers to the implementation of this technology. Some key challenges are further discussed [8]. 3.2.1  Cost The organizations which are trying their self-governing vehicles have paid powerful sum in building those vehicles. If this technology gets proved, then it is estimated that it might boil down to half of the value which is yet an exceptionally immense sum. In the coming future, if the costs of self-governing vehicles come some place close to regular vehicles, JD Power late overview depicts 37% of individuals would absolutely or presumably yet an independent vehicle as their future vehicle. 3.2.2  Technology Challenges Companies testing autonomous vehicles are paid huge sums in the construction of those vehicles. Google paid about $80,000 for the AV module, which is not available to the common man. Once this technology is proven, its price can be halved, which is yet an exceptionally huge amount. In the forthcoming future, autonomous vehicle costs will come anywhere near conventional vehicles. A recent survey by JD Power found that 37% of people have their next vehicle as a perfect or autonomous vehicle.

34

T. Parekh et al.

3.2.3  Evacuation of Old Vehicles The greatest test in executing a self-ruling vehicle is to scrap all the old vehicles that are not outfitted with self-ruling modules since it causes a ton of unexpected outcomes and accordingly lessens the effect and well-being of independent vehicles. On the off chance that old vehicles can be brought back, at that point there will be an answer; however without seeing it once more, it will be a tremendous errand. Conventional vehicles are running on the streets. On the off chance, everybody is happy to keep that expenses. 3.2.4  Issue of Unemployment Albeit self-ruling vehicles have numerous advantages yet the greatest test, we predict it is the joblessness issue. The day self-driving technology will be completely proven, drivers’ requirement will be next to negligible. In this way, each one of those individuals who today are earning their work as expert drivers will no more have the option to do the same. The most significant businesses which would get impact with the presence of self-sufficient cars are taxi, heavy-load vehicles, and also marine freighting. 3.2.5  Security and Protection Concern In this day and age where everything is getting constrained by gadgets, security and protection concern is additionally a greatest issue. Electronic information isn’t sheltered today and it’s defenseless against data misuse. Indeed, even an independent vehicle can be used by any psychological oppressor outfit to do their self-­destruction missions. Likewise, as these vehicles will be associated through GPS, anybody can get the position, and it tends to be utilized to any sort of terrible reason. 3.2.6  Guidelines and Regulation The government should make guidelines and laws to smooth out the way toward bringing self-ruling vehicles into the street. There ought to be laws where the worry over security and assurance of safety and the maintenance of this enormous private information ought to be addressed. Self-sufficient vehicles ought not be utilized by any psychological militant gathering this ought to likewise be tended to by government [13].

Intelligent Transportation System in Smart City: A SWOT Analysis

35

3.3  Opportunities 3.3.1  Blockage and Climate Alerts Traffic and climate data can be communicated on vehicles, for example, instant messages or guides data. These frameworks are as of now in support and permit drivers to be educated not to drive in troublesome or hazardous circumstances. 3.3.2  Path Navigation Route frameworks are in assistance, navigating driver to the picked goal, providing early alert about intersections and essential moves, for example, path changes. At the time of the EDDIT venture [14], a driver depicted the route framework as a “co-­ driver or a partner who provides the instructions being aware about the paths.” Voice directions along with each turn visual guide lessen memory use and visual obstruction with essential driving work and significantly improve execution and blunders in experienced drivers. These frameworks will lessen driving pressure and limit clashes and mishaps while evading the requirement for path changes or modifications right away before the intersection. 3.3.3  Obstruction Acknowledgment Frameworks are as of now set up to identify obstructions behind and close to the automotive while in low-speed moves, for example, parking. Similar frameworks models may alert automotive to change vulnerable sides or during amalgamation and path changes. These things can alert the driver and forestall crashes. 3.3.4  Night-Perception Improvement Numerous frameworks are being created to enhance and process drivers’ perception around evening time. Some obvious range (i.e., bright or infrared radiation) enlightens the front of automotive with radiation and utilizations fluorescence to upgrade direct perceivability or utilize radiation sensors to make pictures of articles that are not rehandled on the sensor by individual’s eye. Images from the thermal energy beamers in the nature or surroundings are produced from various other systems. The scope of identifying people on foot to walk, hindrances, and road trajectories can be fundamentally expanded. A portion of these frameworks in military assistance are as of now dependent on night-perception technology. Some permit the driver to see legitimately through the windshield, while others make a processed picture on the head-up display (HUD) at the inner side of the windshield from which the outer world can in any case be seen. Nonetheless, more (old) experienced drivers are

36

T. Parekh et al.

especially helpless against psychological deliberation, which pictures a HUD and forestalls the identification of huge outside articles [15]. 3.3.5  Intelligent Cruise Control (ICC) and Lane Keeping Assistance ICC is a system that controls the speed of a vehicle and the distance behind a previous vehicle. Commercially fast control vehicle speed will soon be available at a foreordained cost while keeping up a protected route between automobiles. The forthcoming variants will tell the driver if the foreordained speed surpasses the speed limit and inquire as to whether the driver needs to go along. On the off chance that a perilous circumstance is discovered, they can control the accelerator as well as brake to forestall backside crashes. Later on, the ICC framework will automatically constrain vehicle speed to as far as possible and reduce the speed when moving toward turns, red traffic lights. Surroundings speed cutoff points may change contingent upon the day or climate conditions [16]. There are prototype systems to assist with lane keeping. These are applied in the middle of the lane following the vehicle to reduce the following small forces. They possibly work when the steering wheel is under control of the driver and if necessary, forces can be changed by driver. 3.3.6  Avoidance and Warning of Collision ICC provides safety from collisions at rear-end of the vehicle. A hindrance recognition framework to caution of contentions during mergers or path changes, which old or elder drivers find troublesome, will be accessible in a couple of years. Old drivers are bound to be hit by vision while driving in rush hour gridlock or at uncontrolled intersections. Preliminaries with a break-estimating framework to assist drivers with intersection traffic have yielded great outcomes [14], albeit any such framework has been in business activity for quite a long time. This framework will keep going for a long time later on to caution of perilous circumstances if basic moves at uncontrolled intersections become more entangled. Furthermore, the viability of a crash notice relies fundamentally upon whether the driver forms data accessible in the sequential or in parallel [17]. The suggestion here is that if the specialized troubles can, in the end, be settled, more established drivers can work better through programmed control intercession in case of an inevitable impact, as a crash alert meddles with the driver’s essential execution of vehicle control. While such a framework may not be very much acknowledged by drivers, elderly drivers have demonstrated a craving for controlling help.

Intelligent Transportation System in Smart City: A SWOT Analysis

37

3.3.7  Driver Status Monitoring This sentence depicts about the driver status first in the vehicles which are commercial and later on to the personal cars as well. These are likely to appear first on commercial vehicles, but this applies to cars at the right time. 3.3.8  Sign in to the Vehicle The framework will in the end be created, which will show street signs inside at the display in the vehicle and caution drivers of mishaps or unordinary circumstances. These require broad checking of a few transmitters to move data from street and rough terrain to vehicle. Interest in this foundation is probably going to postpone the execution of these frameworks. On the other hand, subtleties of street signs can be put away with an advanced route map for showing at a time suitable to the route framework. The issue for this situation is refreshing the database. The self-­governing framework doesn’t show transitory signals, for example, “redirection,” “street work,” or “floods.” 3.3.9  Occupant Safety Seat belts and airbags, which are system occupant protection systems, are intended to ensure a huge male inhabitant. For this reason, they are smaller and lighter and can exert more power than those sitting near the steering wheel. In particular, in a relatively minor accident, they can cause injuries to the fragile elderly, with no injuries if the convener meets the characteristics of a particular resident. An inhabitancy security framework utilizes sensors to gauge the right weight and seating location with respect to position and also a card to distinguish the proprietor’s age, gender, and different attributes (it is conceivable to outwardly situate the seat, mirror, and controlling wheel naturally to suit the individual). In a given accident, the framework alters the heap forced on the inhabitancy so it is in any event important to control them. Limitation frameworks are as of now in administration, which incorporate pre-pressure and lock safety belts to decrease a proprietor’s movement during a mishap. The keen control framework will be additionally evolved.

3.4  Threats There are a few ITS dangers and risks dependent on three expansive characterizations that spread a subset of utilizations and frameworks (A&S). It likewise incorporates the absolute prescribed procedures by which organizations and buyers can ensure its security [18].

38

T. Parekh et al.

3.4.1  Physical Strikes and Dangers While physical strikes tumble to the least level in Trend Micro’s danger figure, physical harm to offices can cause immense misfortunes in fund and recuperation. Its offices are uncovered on streets and expressways, which are truly available to anybody. Noxious entertainers can harm physical segments, for example, uncovered ports, measures, and organization radio wires, prompting disappointments in information transferring and detailing. Another conceivable method to genuinely assault a wide range of vehicles through the Controller Area Network (CAN bus) is to permit applications to control and convey between gadgets without a focal PC.  The security holes of the CAN transport have been found since 2016; however there are still examinations to discover different gadgets, segments, and innovations that can penetrate different frameworks, for example, savvy homes and public organizations to invade organizations. In the pattern of ride-sharing administrations, there are likewise physical segments of other public vehicles, for example, e-bicycles, which clients can’t re-visit appropriate dock stations, specialized and natural difficulties and batteries, and significantly more. Obliteration issues. Moreover, different methods of genuinely harming interconnected vehicles were likewise thought to be, for example, an obscure hypothetical situation or the establishment of extra parts to bargain vehicles or increment deals. 3.4.2  Organization Attacks and Threats As per Trend Micro exploration, network attacks on it are the greatest danger to developing shrewd urban areas and transportation frameworks. Organization attacks and dangers set focuses for the ordinary activity of gadgets and hardware, and disturb administrations, and for information penetration and information robbery. The malware is generally utilized by cybercriminals to circulate different noxious payloads to individuals with incapacities and private help enterprises that depend on data trade, for example, availability and government workplaces, coordination segments, and other related framework, and can empower offices, for example, other business structures, to hurt cybercriminals. Zones of living arrangement for additionally threatening assaults: Distributed Denial of Service (DDOS), Man-in-TheMiddle attack (MiTM) or upgrade of benefits, and so forth. The interruption of frameworks that control and work these transportation frameworks can prompt the pointless and unapproved utilization of significant assets, gotten wellsprings of touchy data to significant expense targets, lost income, and taken property. On account of shared or rental vehicle administrations, there are additionally situations where gadgets can be utilized to follow past clients’ information—a worry is whether the security of data abuses information protection laws.

Intelligent Transportation System in Smart City: A SWOT Analysis

39

3.4.3  Remote Attacks and Threats Remote correspondence for V2V, V2I, and I2I is the foundation of smart vehicle and city activities, particularly progressively information trade and reaction organization. As the notable Jeep Hawk 2015 appeared, vehicles that are distantly undermined by installed parts or associated gadgets are obscure and certainly feasible. Shortcomings in vehicles or peripherals, just as unlicensed public Wi-Fi associations, can help kill vehicles running out and about. As indicated by Georgia Tech’s exploration, just 20% of interconnected vehicles stranded in rush hour gridlock in focal Manhattan are reasonable for individuals with inabilities—including the arrangement of crisis responders and interchanges—or less whenever hacked. It is applied to different urban communities. Moreover, worker side security ought to be improved from the plan stage to conveyance, as noxious sites, feeble passwords, and application information search capacities uncover makers’ information and programming weaknesses used to control traffic or to open and start mechanized vehicles. On the off chance that producers consider gadgets that coordinate voice-­ empowered colleagues, known weakness, and the advancements of these shrewd and associated contraptions, these vehicles can likewise be utilized for bargain. Application weaknesses and openings can be utilized to deal with interruptions and examine associated transport and vehicles or utilized in basic foundation and businesses. 3.4.4  Suitable Practices Cyberattack and interruption anticipation procedures must be incorporated into the execution of the ITS offered by everybody in every nation: strategy producers, makers, residents and/or inhabitants, originators, organizers, and administrators. Infringement on open framework and foundations have gotten typical, and alleviation measures are fundamental to the general framework overall for dynamic assurance and safeguard. While no guard against planned assaults is strong, here are probably the most ideal approaches to lessen it against physical, organization, and remote cyberattacks [18]: 1. Implement and reinforce physical safety efforts around ITS hardware and additionally offices. Ensure no unapproved workers are around the territory to harm or harm explicit pieces of the office. Consider restricting the quantity of approved workers who have physical admittance to gear or office and the inside known and acknowledged breaks or spans. Cutoff associated computerized gadgets brought inside the extent of these gadgets just to gadgets needed to perform security reviews and upkeep checks. 2. Implement organization dividing, checking, recognition, and impeding frameworks, security frameworks. For example, firewalls and danger of the executive’s entryways can distinguish and forestall vindictive URLs, orders,

40

T. Parekh et al.

messages, and content endeavors to reach or execute malevolent activities before assaults. In case of an assault, security and IT groups can rapidly recognize, dismantle, and speed up to keep gate-crashes from entering organizations and endpoints, in this way evading possible loss of information. Observing assists with a multilayered security framework, permitting fast recognition of inconsistencies and excellent information trade, particularly explanatory methods, for example, AI and man-made reasoning—from gadgets to organized cell phones. 3. Perform customary security reviews to ensure there are no holes in the organization, equipment, programming, and firmware. Guide all the gadgets associated with the organization at home or in the association, and ensure that the default qualifications of the producers have been changed for all keen gadgets. Perform weakness sweeps to discover holes in the framework to rapidly execute suitable arrangements. Actualize a fix the board cycle that guarantees downloads to all associated frameworks without interference to administrations and activities. Utilize accessible apparatuses, for example, Shodan to examine for uncovered gadgets, and utilize all ports to forestall multifaceted verification (MFA) or unapproved approaching or active traffic that must be introduced on all appropriate stages. 3.4.5  Challenges The challenges of IoT application for public transport (PT) [19] fall into the accompanying classifications: plans of action, protection and uprightness issues, security, interoperability, versatility, convenience, information assortment, and organization. Such difficulties may have different effects on rural or urban environments. 3.4.6  Further Involvement of IoV (Internet of Vehicle) Current traffic operation strategies are largely based on a centralized control algorithm. When the landscape is complex, it consumes considerable computing resources. With the help of IoV technology, we can deliver a portion of the decisive work to each vehicle. By increasing the exchange of information between vehicles, a decentralized solution can be proposed that provides an effective traffic management service and ensures system stability. 3.4.7  Utilization of Multiple-Source Data in ITS Compared to a single source, multiple resources can provide complementary data, and a multisource data combination can better understand the situation observed by reducing the certainty of individual resources. In addition, the sensor suite is very inexpensive to install and maintain. If one or more sensors do not work due to some failure, it will slightly reduce the penetration rate and have little effect on system

Intelligent Transportation System in Smart City: A SWOT Analysis

41

performance. With the help of ITS, we can get real-time and multisource traffic data. As a result, it allows the transportation system to operate more efficiently. 3.4.8  Automated Driving With the advancement of communication and sensing technology, commercial automated driving is coming soon and will become a related technology research hotspot. In addition, traffic activity is also affected. In addition to making cars more secure, specialists are creating approaches to automate vehicle innovation to diminish blockage and fuel utilization. 3.4.9  Model Validation Although a large number of transportation models have been proposed, there is no proper method for their accuracy. On the one hand, the transportation system is sophisticated, and the simulation does not give a uniform assessment of the real world. On the other hand, large-scale reality verification is impossible. In the near future, with its development, we will be able to use more data we collect to reduce bias between simulation and the real world. 3.4.10  Security Electronic security is always a serious question. Although security measures for personal computer and Internet communications have been implemented after extensive scrutiny, infrastructure protocols have been developed from vehicle to vehicle and vehicle with security implemented at an early development stage. Computer hackers are usually more likely to target it, resulting in collisions and traffic interruptions or theft of personal information. To overcome those challenges, some tasks may become the focus of future research. For our part, the control strategy of the Cooperative Automated Vehicles (CAV) is seen as a major opportunity for its development. CAV is an IoV-based automated vehicle [20]. The CAV collects data from all transportation elements and can collaborate with all of them to build the entire traffic system efficiently. It differs from the centralized operating approach in traditional traffic management systems. The CAV gave us a new perspective on how to work in a decentralized way [21]. 3.4.11  Standard Measurements for Course Assessment Information assortment and examination ventures zeroing in on recognizing and choosing suitable highlights for course assessment ought to be supported and at last for the proposition of standard measures for course correlation [22]. Truth be told,

42

T. Parekh et al.

courses can be sorted by duplication of estimations that are not regularly estimated, or just considered in one contextual investigation, yet not in others. For the most part acknowledged essential models for course correlation permit better similar arrangements that show the correct ways for clients. Specifically, a way can be thought of as various indistinguishable areas, each with its own attributes. Some of them are entirely steady over the long haul (e.g., asphalt materials and conditions, street widths, inclines, paths), while others are momentary variable (e.g., vehicle-­ free zones with steep trains on either side, schedules). The right mix of essential networks will in the end prompt the meaning of a lot of standard lattices, so it very well may be utilized to handily assess courses and look at course arrangements. 3.4.12  Dynamic Ideal Path These days, the accessibility of metropolitan sensors considers significant variables for estimating an approach to reflect constant states of nature (i.e., traffic-disturbed neighborhood occasions, floods, antagonistic climate, contamination). The accessibility of significant measurements, the area of courses (e.g., metropolitan, rural, provincial) and explicit buyer needs (e.g., old and resigned people with various necessities from representatives and understudies) “ideal powerful meaning of” course “related to other people.” The course ought to be characterized as “naturally and purchaser situated, relevantly unique.” Setting up unique standards to decide the correct way can likewise consider normal obstructions set by nearby policymakers. (To plan, strict necessities can be characterized as far as contamination because of verifiable atmosphere [23]. Issues, more center, ought to be given to a portion of the other. Vehicle types, regardless of whether taken independently, can prompt more contamination than others and can give little clog and compaction.) Applying a positive meaning of the way in the end prompts appointing various loads to the recently indicated standard grid. 3.4.13  Arranging and Programming the Public Capital in Transportation Government interest in arranging and programming transportation coordination in the public, territorial, or neighborhood setting is convoluted by the decent variety of transportation courses and monetary, framework, or operational duties with numerous public, local, and/or nearby governments. IoT advancements can quicken information assortment and access on provincial transportation, empowering neighborhood specialists to give total data on current customer patterns in a specific region [24]. This will prompt more successful activity coordination among officials at various levels.

Intelligent Transportation System in Smart City: A SWOT Analysis

43

4  Results and Discussion 4.1  ITS Algorithm The ITS targets diminishing the mean wait time (AWT) for automobiles on traffic lights. AWT is referred to as the time it takes for a vehicle to go through a crossing point, equivalent to the entirety of the waiting times per vehicle, divided by the quantity of vehicles at the convergence as shown in Eq. (1). This part gives three solutions for its solution and portrays the outcomes, examination, and conversation of the investigations. A few tests were performed to acquire different outcomes, and afterward the best or most suitable arrangement was chosen in all cases: AWT =

∑ Waiting time for car Number of cars in the intersection

(1)

The first approach in solution is the counter arrangement, for example, sees that the plate light is controlled by the quantity of vehicles on every road, hence the street with the most elevated counter. The most priority need was found to open at beginning. There have been a few examinations with D experiments to contrast the counter outcomes and the typical position (each time the traffic signals open in similar request for various 30 s on all vehicles). The subsequent solution is the edge arrangement, which implies the controlling cycle for the traffic lights will be as per the period at which the traffic light has been turned off (red light). The arrangement fills in as follows: • Check the OFF an ideal opportunity for the four traffic lights. • Perform comparison between them. • The traffic signal with the more OFF time opens first. The past method was performed under one condition that this traffic signal had a particular wait time equivalent to 90 s (edge). The last solution is a mixture arrangement that joins the above two solutions (counter and edge arrangements). This arrangement fills in as follows: • The server contrasts the vehicle’s counter of all the traffic lights that crossed the limit estimation of 90 s. • If there are paths more than one past the holding time edge, open the road with the most elevated counter (blockage). The combined statistics of the three methods, i.e., counter, threshold, and threshold and counter, is presented as chart for the calculation of average waiting time using for each method as shown in Fig. 7.

44

T. Parekh et al.

Average Waing Time

160

AWT IN SECONDS

140 120 100 80 60 40 20 0

17

30

53

65

78

80

82

85

87

96

110

NO.OF CARS Counter Soluon

Threshold Soluon

Counter Threshold Soluon

Fig. 7  A bar graph representation of AWT for three approaches

5  Conclusion This chapter describes ITS structure, autonomous vehicles, its algorithm, and how to implement smart public transport to successfully implement smart transport in fast-growing smart cities. The challenges encountered throughout the preprocessing and process are illustrated with appropriate scenarios. The results of the methods promote in detail the clean and hygienic smart transportation process with SWOT analysis.

References 1. D. Bamwesigye, P. Hlavackova, Analysis of sustainable transport for smart cities. Sustainability 11, 2140 (2019). https://doi.org/10.3390/su11072140 2. S.W. Turner, S. Uludag, Intelligent transportation as the key enabler of smart cities, in IEEE/ IFIP NOMS 2016 Workshop: International Workshop on Platforms and Applications for Smart Cities (PASC) (2016), p. 1261 3. S. Muthuramalingam, A. Bharathi, S.R. Kumar, N. Gayathri, R. Sathiyaraj, B. Balamurugan, “IoT Based Intelligent Transportation System (IoT-ITS) for Global Perspective: A Case Study”, Internet of things and Bigdata Analytics for smart generation (2018), p. 279–300 4. J. Sussma, Perspectives on Intelligent Transportation Systems (ITS) (Springer Science Business Media, 2005) 5. X. Yan, H. Zhang, C. Wu, Research and development of intelligent transportation systems, in 2012 11th International Symposium on Distributed Computing and Applications to Business, Engineering Science (2012), pp. 321–327 6. M.  Wooldridge, N.R.  Jennings, Intelligent agents: theory and practice. Knowl. Eng. Rev. 10(2), 115152 (1995) 7. Y. Lin, P. Wang, M. Ma, Intelligent transportation system: concept, challenge and opportunity (IEEE 2017), p. 161, https://doi.org/10.1109/BigDataSecurity.2017.50. 978-1-5090-6296-6/17 8. M.V. Rajasekhar, A.K. Jaswal, Autonomous vehicles—the future of Automobiles (2015)

Intelligent Transportation System in Smart City: A SWOT Analysis

45

9. A. Patnaik, P. Tiwari, On the realization of autonomous cars in diverse conditions (IJAIEM, Sept 2014) 10. S. Dahbour, R. Qutteneh, Y. Al-Shafie, I. Tumar, Y. Hassouneh, A.A. Issa, Intelligent transportation system in smart cities (ITSSC), ed. by K. Arai et al., IntelliSys 2018, AISC 868, (2019), pp. 1157–1170 11. P. Gawade, A. Meeankshi, IOT based smart public transport system. Int. J. Sci. Res. Publ. 7(7), (July 2017) 12. M. Sakairi, Water-cluster-detecting breath sensor and applications in cars for detecting drunk or drowsy driving. IEEE Sensors J. 12(5), 1078–1083 (May 2012) 13. Autonomous Vehicle Technology: A Guide for Policymakers (RAND Corporation, 2014) 14. P.R. Oxley, C.G.B. Mitchell, Final report on elderly and disabled driver’s information telematics (Project EDDIT), Commission of the European Communities DG XIII, R & D Programme Telematics Systems in the Area of Transport (DRIVE II), Brussels (1995) 15. D.R. Tufano, Automotive HUDs: the overlooked safety issues. Hum. Factors 39(2), 303–311 (1997) 16. S.L. Suen, C.G.B. Mitchell, S. Henderson, “Application of Intelligent transportation systems to enhance vehicle safety for elderly and less able travellers”, 16th International Technical Conference on the Enhanced Safety of Vehicles (1998) 17. W. Najm, M. Mironer, J. Koziol Jr., J. Wang, R.R. Knipling, Synthesis report: “Examination of target vehicular crashes and potential ITS countermeasures”, U.S.  Department of Transportation Report DOT-HS-808-263, National Highway Traffic Safety Administration, U.S. Department of Transportation, Washington, DC (1995) 18. Out on a highway Run: Threats and Risks on ITS and smart vehicles, Posted in Internet of Things, Intelligent Transportation Systems, Industrial Internet of Things, Cybercrime, Cybersecurity article dated (03 Dec 2019) 19. P. Davidsson, B. Hajinasab, J. Holmgren, et al., The fourth wave of digitalization and public transport: opportunities and challenges. Sustainability 812, 1248 (2016) 20. M. Wang, W. Daamen, S.P. Hoogendoorn, B. van Arem, Cooperative car-following control: distributed algorithm and impact on moving jam features. IEEE Trans. Intell. Transp. Syst. 17(6), 1551–1563 (2016) 21. G.  Guo, S.  Wen, Communication scheduling and control of a platoon of vehicles in VANETS. IEEE Trans. Intell. Transp. Syst. 17(6), 1551–1563 (2016) 22. J.G.  Su, M.  Winters, M.  Nunes, et  al., Designing a route planner to facilitate and promote cycling in Metro Vancouver, Canada. Transp. Res. A Policy Pract. 44(7), 495–505 (2010) 23. S. Porru, F.E. Misso, F.E. Pani, C. Repetto, “Smart mobility and public transport: opportunities and challenges in rural and urban areas”, Journal of Traffic and Transportation Engineering (English Edition) 7(1), 88–97 (February 2020) 24. B.V.  Kumar, R.  Akash, A.  Karan, K.  Vishal, J.A.P.  Kumar, A smart public transportation system for reliable and hassle free conveyance in sustainable smart cities, in International Conference on Computer Communication and Informatics (ICCCI) (IEEE, 2020), pp. 1–6 Tejas Parekh  is currently pursuing B.  Tech degree in Information Technology final year in PSG College of Technology, Coimbatore. He has done few academic projects in the fields of image processing, machine learning, and operating system. He has also done a project in collaboration with Samsung Prism Program in the research area of networking, along with that he has also served as a campus ambassador for Samsung Prism. He has been passionately working and has been entitled as head to Public Relations of Entrepreneurs Cell, PSG College of Technology.

46

T. Parekh et al. B.  Vinoth Kumar  is working as an associate professor with 16  years of experience in the Department of Information Technology at PSG College of Technology. His research interests include soft computing, blockchain, and digital image processing. He is author of more than 26 papers in refereed journals and international conferences. He has edited three books with reputed publishers such as Springer and CRC Press. He serves as a guest editor/reviewer of many journals with leading publishers such as Inderscience and Springer.

Dr. R. Maheswar  has completed his B.E. (ECE) from Madras University in the year 1999, M.E. (Applied Electronics) from Bharathiar University in the year 2002, and Ph.D. in the field of Wireless Sensor Network from Anna University in the year 2012. He has about 19 years of teaching experience at various levels and is presently working as dean—research (assistant) and dean incharge for the School of EEE, VIT Bhopal University, Bhopal. He has published around 70 papers at international journals and international conferences and published 4 patents. His research interest includes wireless sensor network, IoT, queueing theory, and performance evaluation. He has served as guest editor for Wireless Networks journal, Springer, and served as editorial review board member for peer-reviewed journals and also edited four books supported by EAI/Springer Innovations in Communications and Computing book series. He is presently an associate editor in Wireless Networks journal, Springer; Alexandria Engineering Journal, Elsevier; and Ad Hoc & Sensor Wireless Networks journal, Old City Publishing. P.  Sivakumar  received his B.E. degree in Electrical and Electronics with I class in 2006 from Anna University. He completed his M.E. degree in Embedded System Technologies with I class in 2009 from Anna University Coimbatore. He completed his Ph.D. in Electrical Engineering with a specialization of Automotive Embedded Software in the year 2018 from Anna University, PSG College of Technology. His research interests include embedded system, model-based design, model-based testing of automotive software, automotive software development, and fog and edge computing. He has around 14 years of teaching experience. He has published 14 papers in reputed international journals. He has also published 20 national and international conferences papers and has organized national seminar and workshop funded by DRDO, DST, and MNRE.

Intelligent Transportation System in Smart City: A SWOT Analysis

47

B. Surendiran  is currently working as associate dean academic and assistant professor in the Department of Computer Science and Engineering at National Institute of Technology Puducherry, Karaikal. He had completed his Ph.D. from National Institute of Technology Trichy. He has more than 30 publications in international conferences and journals. He had reviewed more than 200+ papers for various journals and conferences. His research interests include medical imaging, machine learning, dimensionality reduction, and intrusion detection systems.

Dr. R. M. Aileni  (female) is a scientific researcher third degree in Computer Science and has obtained in 2012 the Ph.D. degree in Industrial Engineering at Technical University “Gheorghe Asachi” of Iasi. She is a Ph.D. student at Faculty of Electronics, Telecommunication and Information Technology, Politehnica University of Bucharest. She graduated Faculty of Textile Leather and Industrial Engineering Management and Faculty of Computer Science. In 2010 during her Ph.D., she obtained a research fellowship for doctoral studies at ENSAIT, Lille University of Science and Technology, France, where she specialized in 3D modeling and simulation for textiles, using the Kawabata system, 2D-3D design concept for the design, and simulation of technical textile articles. In 2015, she obtained the Excellence Fellowship Grant for doctoral studies in Belgium, Mons University.

Deep Learning in Smart Applications: Approaches and Challenges M. Sowmiya, B. Banu Rekha, and R. Kanthavel

1  Introduction The concept of smart applications is rising rapidly due to an increase in urbanization of developed and developing countries. The main goal of smart applications is to minimize human intervention and maximize the use of automated technologies. In the deployment of sensors, wearable devices, and actuators in various domains, a massive amount of data is generated at high speed. Also, smart cities are deploying intelligent technologies to automate the process. Mainly, in sectors such as transportation, environmental, healthcare, and agriculture, industries are currently presenting artificial intelligence (AI) techniques. Rapid developments in AI provide the platform to process the huge amount of data, analyze the pattern, and forecast the outcome [1]. This paper discusses the progress of deep learning (DL) methods in environmental and healthcare sectors. Specifically, this paper presents a review of the theoretical background, deep learning concepts, methodology involved in waste management, and retinal image analysis of diabetic retinopathy. Figure 1 highlights the era of using deep learning in smart city development. The main contributions of this work focus on the study of deep learning in two fields which are described as follows: 1. Waste segregation is a challenging part of making the environment smarter. For developing a sustainable economy, it is essential to use intelligent systems that

M. Sowmiya (*) Department of ECE, PSG Institute of Technology and Applied Research, Coimbatore, India B. B. Rekha Department of Biomedical Engineering, PSG College of Technology, Coimbatore, India R. Kanthavel Department of Computer Engineering, King Khalid University, Abha, Saudi Arabia © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 R. Maheswar et al. (eds.), Challenges and Solutions for Sustainable Smart City Development, EAI/Springer Innovations in Communication and Computing, https://doi.org/10.1007/978-3-030-70183-3_3

49

50

M. Sowmiya et al.

Fig. 1  Popularity of deep learning and smart city over the past 5 years (Source: Google trends)

require less manpower. The smart bins are emerging to segregate the waste and put it into corresponding waste categories. In recent days, deep learning algorithms have been used to automatically segregate the recycling waste to develop the prototype. This paper highlights the use of convolutional neural network (CNN) to classify and identify the properties of waste so that it can be employed in smart bins. 2. Artificial intelligence techniques applied in the healthcare domain allow the professionals to monitor, diagnose, and highlight the province of the problem and propose the possible required solution. Moreover, many smart devices are emerging to reduce the cost of public healthcare systems. Recently, smart contact lenses to diagnose and treat diabetic retinopathy [2] and smartphone-based DR detection systems have been developed [3]. Deep learning algorithms have rapidly turned out to be a methodology for analyzing and diagnosing the various medical pathological conditions. This paper provides the knowledge of abnormalities, lesions identified in retinal image to diagnose diabetic retinopathy (DR), and DL methods used to segment the features and identifies the issues in diagnosis. We also discuss the challenges in current research and provide the suggestions for future researchers.

2  Waste Segregation and Classification Waste management is one of the challenging issues for both developed and developing countries due to the rise in population growth, urbanization, and industrialization. According to the global survey given by the World Bank, waste generation is expected to increase from 2.01 billion tons (2016) to 3.40 billion tons (2050). Urban people in India generate 62 million tons of municipal solid waste (MSW) each year, and only 11.9 million tons is treated for disposal and recycling [DownToEarth

Deep Learning in Smart Applications: Approaches and Challenges

51

survey]. Wastes are often dumped in landfill sites or openly burned. These uncollected waste practices create the effect of health issues and environmental consequences like climate change and air and water pollution. On making the sustainable cities, municipalities focus on an integrated approach based on the 3Rs (reduce, reuse, and recycle) [4]. However, few efforts have been made to recycle the amount of waste generation at source due to the limitation of resources. Waste management is a shared activity of segregation, collection, transportation, recycling, and disposal of waste [5]. Waste segregation and recycling is an essential component for building a smart garbage system. Segregation of solid waste requires manpower, which brings many health issues for the waste sorters. Recently due to the COVID-19 situation, people are disposing of gloves and masks mixed with other wastes and making the waste collectors lives at risk. Additionally, this is a time-­ consuming and less efficient process due to the massive amount of waste generated each day. Recent advancements in technology have given the automated approach of segregating the waste from households to industries. Researchers have been exploring automated waste segregation techniques to advance the proficiency of the recycling process. This paper provides a comprehensive review of recent research on the segregation of recyclable wastes using artificial intelligence techniques. This paper focuses on  DL methods to identify and classify the waste from the image acquisition system, which can be further reused or recycled. Also, we discussed the widely used DL architectures, available datasets, and challenges yet to be explored in the future with fruitful suggestions.

2.1  Waste Materials Municipal solid waste (MSW) includes mostly recyclable waste (paper, plastic, glass, metals, etc.) and nonrecyclable wastes (food waste, medical waste, sanitation waste) [EPA, US]. From the survey given by the US Environmental Agency, recycling progression of materials like plastics, metals, and glass is limited as shown in Fig. 2. Recyclable waste can be processed to produce useful products, and recycling reduces the amount of waste sent to landfills and incinerators. This review work will primarily focus on the convergence of DL techniques in image-based recyclable waste identification systems, which can be implemented initially in the automated process to segregate the wastes.

2.2  Database Information The main aim of the study discusses the deep learning models to classify the recyclable wastes like paper, plastic, metal, glass, and others. But there are only a few datasets available for the research openly. Waste image datasets comprise of either collected from Google search, acquisition systems, or existing open image

52

M. Sowmiya et al.

Fig. 2  Recycling and composting proportion of materials (Source: United States Environmental Protection Agency, 2017) Table 1  Examples of waste image categories Waste class Paper Plastic Metal Glass Fruit/vegetable Kitchen waste Trash

Waste items Cups, books, newspaper, cardboard Shampoo bottles, bags Can, pen, cap Bottles Apple, orange, carrot, onion Lunch box, egg, cooked rice Other items

Recyclable/others Recyclable Recyclable Recyclable Recyclable Others Others Others

databases. There are few open databases available such as TrashNet [6], Flickr Material Database (FMD) [7], and Materials in Context Database (MINC-2500) [8]. Our focus is on image-based recyclable waste identification, and most of the researchers used the TrashNet dataset for related works. TrashNet contains a collection of 2527 images from 6 categories, paper, plastic, metal, glass, cardboard, and trash. The trash images are taken from the Stanford area with white background and lighting variations using mobile devices. Each picture is resized down to a spatial resolution of 512 × 384 pixels because of the resource limitations. Researchers also created their own dataset with specific categories. Table 1 shows a few examples of waste classes and their subgroup. Waste images can be generated through an image acquisition system or through smartphones. The camera is connected to a controller like raspberry pi 3 which implements a Python script that captures images [9] of waste types, to achieve better knowledge during training.

Deep Learning in Smart Applications: Approaches and Challenges

53

Fig. 3  Example of waste images

Fig. 4  Example of metal cans with different shapes

Figure 3 shows a sample of recyclable waste categories. The waste image has slight variations than the original image, due to change in shape and properties of waste as a result of being thrown away. As shown in Fig. 4, the same metal can have three different shapes. Therefore, precise classification and identification of the above said cases are challenging ones. Deep learning algorithms are able to extract those features and accurately identify waste types.

2.3  Methodology in Waste Segregation Waste segregation using computer vision techniques involves obtaining images from the camera, object detection, feature extraction, and classification as recyclable or not. Since the real-time image is noisy and random, pre-processing the data will improve the performance of the model. Various pre-processing steps such as image resize, removal of noise, and enhancing the brightness [6] have been employed. Data normalization methods such as Z-score and Min-Max can be used if necessary, to make the model less prone to bias.

54

M. Sowmiya et al.

2.3.1  Object Detection Multiple waste items and non-waste items might be present in real-time captured garbage images. During this scenario, each waste item has to be identified separately and classified into a particular class. Few kinds of research have addressed object detection in waste segregation so far. Region-based convolutional neural network (R-CNN) models have been used to detect and identify the size of the waste. It may require the size of all the images in the training set to be the same. Authors have developed a model with Faster R-CNN to classify waste into paper, recycling, and landfill from multiple objects in a single image with precision accuracy of 68.3% [10]. Similarly, E-waste can be detected using Faster R-CNN with an accuracy of 90% [11]. 2.3.2  Feature Extraction and Classification Size, shape, texture, and the color of the waste item are the essential features that can be extracted during the training of deep neural models like CNN. Principal component analysis (PCA) might be used to reduce the dimension of the feature vectors prior to the classification stage. The hybrid system can be developed by using the features extracted from the CNN model and process it to a classifier as shown in Fig. 5. Features are extracted using pre-trained models and a  multilayer perceptron is trained to identify the waste calss [12]. From the study, we found that SVM along with CNN can improve prediction accuracy [13]. Moreover, the additional information regarding the waste properties can be given by the sensing unit which enhances the CNN prediction [12]. 2.3.3  Supervised Architectures for Waste Image Classification Convolutional neural network (CNN) is one of the most familiar deep learning algorithms that performs sequential operations of convolution and pooling to extract the features from the data. CNN provides the spatial low-dimensional representation with robust features. Applications like image classification, segmentation, and Waste Image Database

Feature Extractor (Pre-Trained CNN Model)

Classifier

Predicted Waste Class

Convolution

Pooling

Convolution Pooling

Fig. 5  Predicting the waste class using pre-trained model

Deep Learning in Smart Applications: Approaches and Challenges

55

detection are extensively using CNN to achieve state-of-the-art results. Many researchers used CNN architectures to perform waste image classification and Faster R-CNN to detect the waste object in the image. Deep architectures, namely, AlexNet, GoogleNet, VGGNet, and ResNet [14], pre-trained on ImageNet are commonly used in the research work. AlexNet is a collection of 8 layers, GoogleNet consists of 22 layers, while VGGNet and ResNet are available in several layer configurations. 2.3.3.1  AlexNet Architecture for Waste Image Classification AlexNet architecture outperforms the CNN in ImageNet LSVRC-2012 competition with minimal error rate. It consists of eight layers including a convolutional layer and fully connected layers as shown in Fig. 6. To extract the deep features, each layer used kernels such as 11 × 11, 5 × 5, and 3 × 3. After convolution, ReLU improves the convergence followed by the overlapping pooling layer. Finally, 1000 class labels are prophesied by the Softmax layer. To reduce overfitting, data augmentation, and dropout in fully connected layer was implemented [13]. Similarly, other architectures used in waste segregation are detailed in Table 2. Table 3 summarizes the variant of the CNN model instigated for various waste image datasets and highlights the practice used. AlexNet pre-trained on the ImageNet database can act as a feature extractor. ImageNet database contains 15 million high-resolution images with 22,000

Fig. 6  AlexNet architecture Table 2  Characteristics of CNN architectures Architecture AlexNet VGG16 GoogleNet ResNet50 Inception-v3

Year 2012 2014 2014 2015 2015

No. of layers 8 16 22 50 (residual connections) 48

Parameters ~60 M ~138 M ~6 M ~26 M ~23 M

AlexNet, VGG16, GoogleNet, and ResNet VGG, Inception, ResNet

Inception-ResNetV2, DenseNet121

CNN

[14]

[16]

[17]

[15]

Recyclable (paper, plastic, metal, glass) and nonrecyclable (fruits/ vegetables, kitchen waste) 2527 images with 6 different classes Paper, plastic, metal, glass, cardboard, general trash Paper, plastic, metal, glass, cardboard, trash

Own

CNN (AlexNet) + multilayer perceptron (MLP)

[12]

Garbage In Images (GINI)

TrashNet

Own (garbage images) TrashNet

Plastic, paper, and metal

Own

CNN (AlexNet) and SVM

[9]

Waste item Paper, plastic, metal, glass

Dataset TrashNet (1800)+ own (1302)

References Architecture [6] VGG16, ResNet18

Table 3  Outline of deep learning architectures for waste image classification

Inception—ResNet produced the best performance Transfer learning, fine-tuning the parameters, optimizers like adam and adadelta are used Reduction in prediction time, memory usage

Classifier—Softmax, SVM

Data augmentation, identifying multiple wastes in images using Faster R-CNN

Observations Used data augmentation and transfer learning, overfitting occurred in VGG16, ResNet18 produced the best outcome Overfitting is reduced

87.69%

InceptionResNetV2—87% DenseNet121—95%

GoogleNet + SVM— 97.86% 88.6%

CNN—83% SVM—94.8% 90%

Accuracy 87%

56 M. Sowmiya et al.

Deep Learning in Smart Applications: Approaches and Challenges

57

categories [13]. Waste datasets are tested with the pre-trained model to classify the waste items. Comparison of VGG, Inception, and ResNet models has been analyzed for the TrashNet dataset. From the analysis, combining Inception-ResNet models produced more accuracy than the individual model [15]. In addition, it is found that some models are biased toward particular waste item due to imbalance in data categories; low inter-class variations like the plastic bottle are confused with a glass bottle. Development of diverse architectures such as GarbNet [18] or OscarNet [18], which are based on pre-trained CNN architectures such as AlexNet or VGG-19, has been made for waste classification. However, researchers choose to train the model on a massive dataset like ImageNet and then extract the features and used it for image classification. 2.3.3.2  Faster R-CNN Faster R-CNN is a variant of the R-CNN family developed in 2015. It is widely used in object detection applications, where the bounding box for each object is generated and features of those objects are extracted using backbone CNN. Faster R-CNN is implemented to detect the objects in food tray images along with CNN that produced the mean average precision of 86% [19], as shown in Fig.  7. It can be employed in a multi-waste image to abstract the single waste and fed for further process.

2.4  Network Training Methods 2.4.1  Transfer Learning In applications, research using transfer learning has gained more attention due to limited computational resources and a lack of large-scale datasets for training. However, the number of images used to train the deep CNN from scratch is very less due to limited availability of open-source waste dataset. Identifying an accurate waste classification model is a critical task because of the varying nature of wastes and their method of disposal. Most of the researchers have used a pre-trained network for identifying wastes. The two approaches used for training and classification are: 1. Using a pre-trained network as a feature extractor and fine-tuning a pre-trained network on waste data 2. Training a CNN model from scratch using the available databases Pre-training a model using large datasets like ImageNet [13] and features are extracted, fine-tuning the model with the specific application data. Since, pre-training a network makes the model to learn the parameters, so that it can be used for any similar applications.

58

M. Sowmiya et al.

Fig. 7  Example for bounding box for each object in food tray [19]

2.4.2  Data Augmentation Overfitting is the challenging issue confronted in deep learning methods for image-­ based applications. Deep learning models produce accurate results based on the training set. For some applications, due to a smaller number of images available for training, the overfitting phenomena will arise. To overcome the above issue, data augmentation has been developed to deep neural networks on smaller datasets [20]. Image augmentation methods like rotate, crop, horizontal flip, and the vertical flip are commonly used to generate a greater number of data for training. Figure 8 shows the sample augmented images. During the validation phase, k-fold cross-validation method can be employed to evaluate the performance of the model before testing. Cross-validation can effectively improve the generalization ability of the model and overcome the overfitting [21]. For example, using tenfold cross-validation, divide all datasets into ten copies, take one copy each time for testing and take nine copies for training, and finally obtain the average.

2.5  Challenges and Future Research Ideas In this section, the major research gap identified from the existing work for segregation of waste categories and suggestions the researcher can take up to develop their ideas. From the survey, it is understood that there are still many challenges ahead for this developing area owing to the limited datasets, real-time constraints, and complex nature of deep learning:

Deep Learning in Smart Applications: Approaches and Challenges

59

Fig. 8  Example of flipped augmented images

1. A limited number of data reduce the training accuracy and cause overfitting. Techniques like transfer learning, data augmentation, and dropout can improve the performance on smaller datasets. 2. Imbalanced dataset, i.e., some of the classes in the dataset have a larger number of images which make the model to become unfair. The uneven distribution of data will affect the model. 3. Fine-tuning the hyperparameters like learning rate, epochs, and optimizer helps in achieving better testing accuracy. 4. Multiple deep models can achieve an advantage over individual, i.e., individual CNN models can be trained and features are extracted to form a new feature vector. Using fusion techniques [22], the feature vector can be analyzed and classified to improve the overall performance of the waste classification problem. Also, feature vectors can be directly fed to fully connected layers. 5. Ensemble classifiers proved to perform well when compared to single base classifiers [23]. Three types of ensemble methods are bagging, boosting, and stacking. Researchers can implement ensemble methods for waste image ­ identification. The following are the main constraints in implementing the automatic waste segregation system in real time: 1. High-speed servers and networks are required to manage and process the information from data acquisition unit and sensors.

60

M. Sowmiya et al.

2. Deployment of the designed model requires high computational resources like memory, power, and cost. 3. Response time of segregating the waste must be minimum, to make use of the waste at an early stage.

3  Diabetic Retinopathy Diabetic retinopathy (DR) is a threatening complication of diabetics, in which the retina is affected due to an increase in blood sugar level. DR leads to vision loss if not detected and treated early. The primary challenge lies in the early detection of DR because there may be no symptoms that often occur in early-stage and untreatable at advanced stages. Early screening of diabetic patients requires well-trained clinical experts, patient attention, and laborious process. Recent researches are highlighting the need for an automated diagnosis system in healthcare. Retinal image analysis is evolving for the early detection of systemic diseases because of the noninvasive visualization of the retinal vessel structure. Ophthalmologists can detect the pathological and microvascular abnormalities by examining the retinal image of the patients. An automated diagnosis system is emerging to assist experts in early DR diagnosis and ocular diseases. The objectives of this review for DR screening: 1 . To summarize the theoretical background in DR 2. To review the existing works for the detection of retinal lesions using deep learning methods for early diagnosis From the study, it is found that deep learning is emerging for the detection of retinal lesions and quite a number of researches started using deep learning vigorously.

3.1  Retinal Imaging Modalities Imaging modalities like Fundus photography and optical coherence tomography (OCT) have been developed for visualizing retinal structures given in Fig. 9. Fundus photography provides a 2D representation, which uses a low-power microscope with an attached camera for imaging the interior surface of the eye which includes the retina, optic disk (OD), retinal vasculature, and macula. Fundus photography is widely used for early screening and diagnosis of diseases such as AMD, glaucoma, and DR [24]. OCT is an advanced imaging modality that uses low coherence interferometry to produce cross-sectional images of the retina. Recently, OCT has been used for the detection of diabetic macular edema [24]. Fundus photography is primarily useful for monitoring the progression of diabetic retinopathy over time. DR is diagnosed by the abnormalities and lesions developed over the retina. This work discusses the deep learning approaches developed for early diagnosis system using the retinal image  and reviews the  retinal

Deep Learning in Smart Applications: Approaches and Challenges

61

Fig. 9 (a) Normal Fundus image. (b) Cross-sectional view of retinal layers [Source: Google]

Fig. 10  Fundus photograph showing abnormal lesions and morphologies [24]

abnormalities, various  stages  of DR, methodology established by the current researches, and challenges faced in DR detection.

3.2  Retinal Lesions DR may cause several pathological lesions in the retina, which are shown in Fig. 10: Microaneurysms (MAs)—Early symptom of DR, due to small bulges that cause leakage of the retinal vessel. MAs are small circular red spots which are usually less than 125 microns in diameter with sharp margins [24]. DR severity depends on the number of MAs.

62

M. Sowmiya et al.

Hemorrhages (HM)—Bleeding due to leakage in weak capillaries causes HM. Larger than MA with irregular margin. They appear in dot and blot shape. Hard Exudates (HE)—Lipoproteins and other fat proteins leaking through abnormal retinal vessels. It appears as white or yellowish-white deposits [24]. Cotton Wool Spots (CWS)—Soft exudates, occur due to occlusion of arteriole. It looks like fluffy bright white lesions. Neovascularization (NV)—The growth of new abnormal blood vessels in the retina and the optic disk. The early signs of DR include microaneurysms, hemorrhages, and future developing bright lesions such as exudates and cotton wool spots. Detecting these lesions is a challenging one because of variation in size, color, texture, and shape. However, manual inspection of retinal abnormalities diagnosis of early DR is quite an exciting task in medical image analysis. As a result, many kinds of research have been conducted on automatic identification of these abnormalities to reduce the burden on the ophthalmologists.

3.3  Stages of DR There are two main stages of DR: non-proliferative diabetic retinopathy (NPDR) and proliferative diabetic retinopathy (PDR). NPDR occurs in early stage where the blood vessels start to bulge, outflow the blood makes the retina to get pathological lesions. NPDR has multiple levels which are explained in Table  4. PDR is an advanced severe stage that causes the growth of abnormal new blood vessels called neovascularization. New blood vessels bleed into the vitreous chamber and detach the retina by scar tissue formation which produces severe vision loss. Figure  11 shows the lesions for various stages.

Table 4  Clinical assessment of abnormalities for various DR stages Stages of DR No DR Mild NPDR Moderate NPDR

Severe NPDR

Proliferative DR

Retinal anomalies No abnormal signs Few microaneurysms Microaneurysms and at least any one of the following:  Hemorrhages  Hard exudates  Cotton wool spots Any of the following:  Intraretinal hemorrhages  Venous beading or intraretinal  Microvascular abnormalities One or more of the following:  Neovascularization  Vitreous or preretinal hemorrhages

Deep Learning in Smart Applications: Approaches and Challenges

63

Fig. 11 (a) Mild NPDR, (b) moderate NPDR, (c) PDR, (d) macular edema [Source: Nature]

3.4  Datasets The extensively used datasets for DR diagnosis are, namely, Kaggle, DIARETDB0, MESSIDOR, E-Ophtha, IDRID, STARE, DRIVE, and CHASEDB. Table 5 summarizes some of the datasets.

3.5  Early Prognosis of DR Using Retinal Images Conventionally, researchers have used feature extractors and fed the features to classifiers like SVM and MLP.  But the optimal selection of features is an essential requirement. DL has an excellent capacity to extract the features. This section presents the approaches used in diagnosing DR using DL. DR detection is characterized by lesion-level and image-level methods [25]. • Lesion-based detection—Number of lesions and their locations are detected in assessing the severity level of DR. It involves two steps, namely, lesion detection/segmentation and classification. Retinal lesions are extracted and the exact location of the lesion is confined. This is a challenging task because retinal

64

M. Sowmiya et al.

Table 5  Dataset available for DR research Dataset name Kaggle/ EyePACS

IDRiD MESSIDOR

No. of images Disease category/purpose 35,000 images DR No DR – Mild – Moderate – Severe – PDR 516 images DR, DME 1200 images DR

HRF

45 images

DIARETDB0

130 images

E-Ophtha

381 for MAs, 82 for exudates 402 images

STARE

DRIVE

33 normal, 7 mild DR stage

CHASEDB

28 images

15 healthy 15 DR 15 glaucoma 20 normal 110 DR 148 MAs or small HA and 233 with no lesion, 47 with exudates and 35 with no lesion Retinal diseases—vessels extraction Optic nerve Vessel segmentation

Vessels extraction

Camera specifications

Color video 3CCD camera with 45° FOV Canon CR-1 fundus camera with 45° FOV Fundus camera with 50° FOV

Canon CR5 non-­ mydriatic 3CCD camera with 45° FOV 25° FOV

fundus images contain other objects with similar appearances, such as red dots and blood vessels. • Image-based detection—Evaluates the image to detect whether there are signs of DR or not. Initial medical diagnosis starts with this image level using DL. The presence of blood vessels and optic disc makes the lesion detection process more challenging because of DR lesions, and these two have similar attributes [25]. The general methodology for DR detection and classification includes the steps of pre-processing, segmentation, feature extraction, and appropriate classification method to detect the severity level of DR as shown in Fig.  12. Unlike manual feature-­based approaches, DL methods integrate all of the phases into an integrated context and automatically learn the features and train the model. Deep architectures are able to learn hierarchical features and achieve excellence in medical diagnosis. The complex features are recognized by having a greater number of hidden layers in the network. Automatic feature extraction and limited pre-processing are the main advantages of DL. Classification learning algorithms can be mainly grouped into two categories, namely, supervised and unsupervised learning. Mostly, supervised methods are used in existing studies for screening DR diagnosis [24].

Deep Learning in Smart Applications: Approaches and Challenges

65

Fig. 12  Stages in DR diagnosis

3.5.1  Pre-processing of Retinal Images The reliable screening system depends on the quality of the fundus image. However, clinical studies emphasized 3–12% of retinal images having low quality [26]. Retinal images were captured in different conditions like dissimilar cameras, nonuniform illumination, variations in light reflection, poor contrast, and insufficient pupil dilation. These may affect the accuracy of the diagnosis methods. Image pre-­ processing techniques improve the diagnostic probability for visual and automatic detection. Pre-processing of retinal image analysis typically covers image resize, noise removal, channel extraction, and contrast enhancement. 3.5.1.1  Extraction of Green Channel The retinal image exhibits low contrast due to the fluctuating nature of imaging environments. Most of the retinal lesions and blood vessels are perceptible in the green channel compared to red and blue channels as shown in Fig. 13. The green channel is extracted and enhanced to improve the contrast.

66

M. Sowmiya et al.

Fig. 13  Individual RGB channels of retinal images

Fig. 14 Contrast-­ enhanced image

3.5.1.2  Contrast Enhancement Contrast limited adaptive histogram equalization (CLAHE) is a variation of histogram equalization which helps to enhance the local contrast between background and blood vessels dynamically. CLAHE divides the retinal image into a nonoverlapping region called tiles and applies the histogram distribution for each tile with a clip limit. After applying CLAHE, the segmentation of blood vessels in fundus images has improved for DR disease identification as shown in Fig. 14. Filtering techniques may be used to smooth the background noise.

Deep Learning in Smart Applications: Approaches and Challenges

67

3.5.2  Segmentation of Retinal Pathologies Structures like microaneurysms, hemorrhages, hard exudates, and cotton wool spots in fundus images are essential to the diagnosis and screening of DR.  As already discussed, the attributes of these features have similarity in appearance, and lesion detection is a more challenging part in NPDR and PDR disease identification given in Fig. 15. Recently, many deep architectures like CNN, deep CNN, and transfer learning have been widely used in extracting the features. In this section, we discuss the method adopted to create the training data for learning and DL approaches used for lesion detection for further analysis of DR. Microaneurysm—Microaneurysms are the early pathology developed in the mild stage of diabetic retinopathy (DR), and its precise detection from fundus image is a challenging task. Since the occurrence of MA looks like a microscopic change in pixel level out of total volume of pixels and variation in size of MAs, accurate diagnosis approaches are required. CNN has the ability to extract deep features, and studies have shown that CNN forms the basis of all developed algorithms. The extensively used methods of segmenting MA have been discussed as follows: 1. Image patches are generated from retinal images which contain MAs and non­MA lesion, features are extracted by training a network using MA and non-MA patches, and binary classification is performed. But this method requires the labeled annotated images for training patches [27]. For training, deep CNN architectures have been widely adopted in researches. Also, the autoencoder model is developed to learn high-level features of MA and fed into the Softmax classifier [28]. 2. Every pixel of the image is classified as MA or non-MA using deep neural networks [29]. Also, both the patch-based and image-based methods have been combined, i.e., patches with annotated labels are used for training and trained CNN was used to test the images in pixel-wise and identify the locations of the pathological signs with probability maps [30]. But the challenge lies in labeled data and making the training set balanced in both class MA and non-MA. Additionally, clinical grading has been adopted to improve deep model performance [31]. Moreover,

Fig. 15  Segmentation of retinal abnormalities

68

M. Sowmiya et al.

combining handcrafted features and deep features learned from CNN has worked on the detection of MA. Hemorrhages (HM)—Hemorrhages occur due to leakage of blood from ruptured blood vessels. Both patch- and image-based investigations can be used in detecting hemorrhages. All the process involved in detecting MA is applicable to HA [32]. Hard Exudates (HE)—Hard exudates are formed by the deposition of lipids and lipoproteins located in the outer layer of the retina, with whitish spots or yellowish-­white appearance. The process of detecting HE is similar to MA. Image patches of size 32 × 32 are generated by using the ground truth into two classes, hard exudate and background. Image patches are given for training an eight-layer CNN and achieved 99.4% training accuracy and 98.6% testing accuracy [33]. The main challenge lies in predicting edge pixels. The image processing techniques such as green channel extraction and contrast enhancement can improve the accuracy along with supervised algorithms to detect exudates [34]. Future research can be extended in two approaches: The first approach is to vary the size of the image patch. Then we can find the effect of image patch on the accuracy of prediction. The second approach is to use an ensemble of CNN architectures. 3.5.3  Deep Learning Architectures to Diagnose DR Deep learning has shown capable improvements in retinal image analysis for the detection of eye and systemic diseases. Though such techniques require high training time and extensive data, the development of transfer learning has been chosen as a diverse method to optimize the process [35]. In this section, some of the researches which used CNN in DR detection are discussed. Convolutional Neural Networks (CNN)—With the increasing popularity of CNN in medical image analysis, CNN can identify features such as microaneurysms, exudate, and hemorrhages in the retina and provide a diagnosis automatically. CNN models comprised of the convolutional layer, pooling layer, and dense layer. Activation functions provide nonlinearity to the model, and backpropagation is used to update the weights and learn the model. Batch normalization and dropout are commonly adopted to achieve faster convergence. CNN requires large datasets for training to avoid overfitting, but the ophthalmic diagnosis has smaller datasets. Transfer learning and data augmentation help to train the model using large datasets, followed by fine-tuning on the test dataset. Also, ensemble learning involves training many CNN models independently and predicts the result. The paper [36] implemented a CNN architecture to identify the features in the retina and achieved an accuracy of 75% on 5000 validation images. Transfer learning method with InceptionNet V3 network achieved an accuracy of 90.9% [37]. Pre-trained models like ResNet50, Xception Nets, DenseNets, and VGG are used and achieved accuracy of 81.3% [38]. Transfer learning method with three tasks, namely, classification, regression, and ordinal regression, was performed. Each task is trained based on features extracted with CNN that produced a sensitivity of 0.99 [39].

Deep Learning in Smart Applications: Approaches and Challenges

69

Table 6  Evaluation metrics for DR Metrics Accuracy

Description Ratio of the correctly classified instances over the total number of instances

Sensitivity

Measures the fraction of correctly classified positive diagnosis

Specificity

Measures the fraction of correctly classified negative diagnosis

Area under the Degree of separability of different classes curve (AUC)

Formula TN + TP TP + TN + FP + FN TP TP + FN TN TN + FP –

The ensembles of deep convolution neural network (CNN) models such as Resnet50, Inceptionv3, Xception, Dense121, and Dense169 are combined to improve the classification for different stages of DR on Kaggle dataset. The results were better compared to the individual model [40]. Nowadays, researchers are developing a smartphone-based early DR detection system. Early detection of DR generally involves eyes to be dilated. Retinal images were acquired using the smartphone “Remidio camera” and diagnosed using artificial intelligence techniques [3] (Table 6). 3.5.4  Research Ideas Even though deep learning has grown over state-of-the-art methods, training the network and fine-tuning the parameters still is a challenging area. Problems such as poor image quality, differences in the size of lesions, and the closeness of some lesions to the vessels have caused many detection algorithms to provide low-­ accuracy results. Moreover, MA which is very close to blood vessels is very difficult to be segmented. Blood vessel segmentation requires precise consideration because thin vessels also have a contribution in disease screening. The automated diagnosis methods can be used as a screening tool to support the clinicians. Non-mydriatic detection system may be developed in the future research. Clinicians can also diagnose in an accurate way after performing preliminary steps using an automated system.

4  Conclusion Methods of efficient waste segregation for developing smart bins have been reviewed. The data acquisition system might be useful in capturing real-time images. This paper discussed the identification of waste from images using deep learning. Mostly,

70

M. Sowmiya et al.

CNN has been widely adopted by the researchers to classify the waste items. The concept of transfer learning is involved in a small number of datasets to improve performance accuracy. Since the real-time images have multiple objects, the idea of identifying multiple objects in a single image is yet to be recognized and classified. The categories of the dataset used may be extended in the future. Research openings can reduce the response time in determining the waste after being thrown away. Retinal image analysis through deep learning is evolving to make the screening process automated. Deep learning techniques can be applied for the segmentation of retinal lesions to detect the stages of DR. Deep learning has the capability of extracting more complex features than those extracted by traditional methods. Deep neural networks (DNN) based  automated diagnosis model  can be used in the  clinical screening process which replaces the skilled ophthalmologists and reduces the time. Although much researches have been conducted to implement the same in real time, the epitome of the method is yet to be observed.

References 1. D. Luckey, H. Fritz, D. Legatiuk, K. Dragos, K. Smarsly, Artificial intelligence techniques for smart city applications, in 18th International Conference on Computing in Civil and Building Engineering, vol. 98 (Springer, 2020), pp. 3–15 2. D.H. Keum, S.-K. Kim, J. Koo, G.H. Lee, C. Jeon, J.W. Mok, B.H. Mun, K.J. Lee, E. Kamrani, C.-K. Joo, S. Shin, J.Y. Sim, D. Myung, S.H. Yun, Z. Bao, S.K. Hahn, Wireless smart contact lens for diabetic diagnosis and therapy. Applied sciences and engineering. Sci. Adv. 6(17), eaba3252 (2020) 3. B. Sosale, S.R. Aravind, H. Murthy, S. Narayana, U. Sharma, S.G.V. Gowda, M. Naveenam, Simple, mobile based artificial intelligence algorithm in the detection of diabetic retinopathy (SMART) study. BMJ Open Diabetes Res. Care 8(1), e000892 (2020) 4. A.H.  Chowdhury, N.  Mohammad, R.U.  Haque, T.  Hossain, Developing 3Rs (reduce, reuse and recycle) strategy for waste management in the urban areas of bangladesh: socioeconomic and climate adoption mitigation option. J. Environ. Sci. Toxicol. Food Technol. 8(5), 09–18 (2014). IOSR 5. M. Jaganmohan, Waste management in India—statistics & facts. Energy Environ. Serv. (2020) 6. D. Gyawali, A. Regmi, A. Shakya, A. Gautam, S. Shrestha, Comparative analysis of multiple deep CNN models for waste. Comput. Vis. Pattern Recognit. arXiv (2020) 7. L. Sharan, R. Rosenholtz, E. Adelson, Material perception: what can you see in a brief glance? J. Vis. 9(784), (2009) 8. S. Bell, P. Upchurch, N. Snavely, K. Bala, Material recognition in the wild with the materials in context database, in IEEE Conference on Computer Vision and Pattern Recognition, 2015, pp. 3479–3487 9. G.E. Sakr, M. Mokbel, A. Darwich, M.N. Khneisser, A. Hadi, Comparing deep learning and support vector machines for autonomous waste sorting, in IEEE International Multidisciplinary Conference on Engineering Technology (IMCET), Beirut, 2016, pp. 207–212 10. O.  Awe, R.  Mengistu, V.  Sreedhar, Smart trash net: waste localization and classification. arXiv (2017) 11. P. Nowakowski, T. Pamuła, Application of deep learning object classifier to improve e-waste collection planning. Waste Manag. 109, 1–9 (2020). Elsevier

Deep Learning in Smart Applications: Approaches and Challenges

71

12. Y.  Chu, C.  Huang, X.  Xie, B.  Tan, S.  Kamal, X.  Xiong, Multilayer hybrid deep-learning method for waste classification and recycling. Comput. Intell. Neurosci. 2018, 5060857 (2018). Hindawi 13. A.  Krizhevsky, I.  Sutskever, G.E.  Hinton, ImageNet classification with deep convolutional neural networks. Commun. ACM 60(6), (2017). ACM Digital Library 14. U. Ozkaya, L. Seyfi, Fine-tuning models comparisons on garbage classification for recyclability. Comput. Vis. Pattern Recognit. arXiv (2018) 15. V.  Ruiz, A.  Sanchez, J.F.  Vélez, B.  Raducanu, Automatic image-based waste classification, from bioinspired systems and biomedical applications to machine learning. Lecture Notes in Computer Science, vol. 11487 (Springer, 2019) 16. C. Bircanoglu, M. Atay, F. Beşer, O. Genç, M.A. Kızrak, RecycleNet: intelligent waste sorting using deep neural networks, in International Symposium on Innovations in Intelligent Systems and Applications (INISTA), IEEE, 2018, pp. 1–7 17. G.  Mittal, K.B.  Yagnik, M.  Garg, N.C.  Krishnan, SpotGarbage: smartphone app to detect garbage using deep learning, in 2016 ACM International Joint Conference on Pervasive and Ubiquitous Computing, 2016, pp. 940–945 18. R.A.  Aral, Ş.R.  Keskin, M.  Kaya, M.  Haciomeroglu, Classification of TrashNet dataset based on deep learning models, in IEEE International Conference on Big Data, USA, 2018, pp. 2058–2062 19. J. Sousa, A. Rebelo, J.S. Cardoso, Automation of waste sorting with deep learning, in 2019 XV Workshop de Visao Computacional (WVC), Sao Bernardo do Campo, Brazil, 2019, pp. 43–48 20. C. Shorten, T.M. Khoshgoftaar, A survey on image data augmentation for deep learning. J. Big Data 6, 60 (2019). Springer 21. X. Xu, X. Qi, X. Diao, Reach on waste classification and identification by transfer learning and lightweight neural network, Preprints (2020) 22. K.  Ahmad, K.  Khan, A.  Al-Fuqaha, Intelligent fusion of deep features for improved waste classification. IEEE Access 8, 96495–96504 (2020) 23. H.  Erdal, I.  Karahanoglu, Bagging ensemble models for bank profitability: an empirical research on Turkish development and investment banks. Appl. Soft Comput. 49, 861–867 (2016). Elsevier 24. M. Badara, M. Harisa, A. Fatima, Application of deep learning for retinal image analysis: a review. Comput. Sci. Rev. 35, 100203 (2020). Elsevier 25. L.  Seoud, T.  Hurtut, J.  Chelbi, F.  Cheriet, J.M.P.  Langlois, Red lesion detection using dynamic shape features for diabetic retinopathy screening. IEEE Trans. Med. Imaging 35(4), 1116–1126 (2016) 26. M.D.  Abràmoff, M.K.  Garvin, M.  Sonka, Retinal imaging and image analysis. IEEE Rev. Biomed. Eng. 3, 169–208 (2010) 27. G. Indumathi, V. Sathananthavathi, Microaneurysms detection for early diagnosis of diabetic retinopathy using shape and steerable Gaussian features, in Telemedicine Technologies, Big Data, Deep Learning, Robotics, Mobile and Remote Applications for Global Healthcare, 2019, pp. 57–69 28. J.  Shan, L.  Li, A deep learning method for microaneurysm detection in fundus images, in 2016 IEEE First International Conference on Connected Health: Applications, Systems and Engineering Technologies (CHASE), Washington, 2016, pp. 357–358 29. M. Haloi, Improved Microaneurysm Detection Using Deep Neural Networks, Computer Vision and Pattern Recognition (Cornell University, Ithaca, 2016) 30. P. Khojasteh, B. Aliahmad, D.K. Kumar, Fundus images analysis using deep features for detection of exudates, hemorrhages and microaneurysms. BMC Ophthalmol. 18(1), 288 (2018) 31. X.  Sui, Y.  Jiang, Y.  Ding, Y.  Peng, W.  Jiao, B.  Zhao, Y.  Zheng, Human grading of diabetic retinopathy improves deep learning based automatic segmentation of microaneurysms from fundus image. Invest. Ophthalmol. Vis. Sci. 61(7), 2037 (2020)

72

M. Sowmiya et al.

32. R.S. Biyani, B.M. Patre, Algorithms for red lesion detection in Diabetic Retinopathy: a review. Biomed. Pharmacother. 107, 681–688 (2018) 33. A. Benzamin, C. Chakraborty, Detection of Hard Exudates in Retinal Fundus Images Using Deep Learning, Image and Video Processing (Cornell University, Ithaca, 2018) 34. N. Theera-Umpon, I. Poonkasem, S. Auephanwiriyakul, et al., Hard exudate detection in retinal fundus images using supervised learning. Neural Comput. & Appl. 32, 13079–13096 (2020) 35. S. Wana, Y. Lianga, Y. Zhang, Deep convolutional neural networks for diabetic retinopathy detection by image classification. Comput. Electr. Eng. 72, 274–282 (2018). Elsevier 36. H. Pratt, F. Coenen, D.M. Broadbent, S.P. Harding, Y. Zheng, Convolutional neural networks for diabetic retinopathy. Proc. Comput. Sci. 90, 200–205 (2016). Elsevier 37. M.T. Hagos, S. Kant, Transfer Learning Based Detection of Diabetic Retinopathy from Small Dataset, Computer Vision and Pattern Recognition (Cornell University, Ithaca, 2019) 38. R. Sarki, S. Michalska, K. Ahmed, H. Wang, Y. Zhang, Convolutional neural networks for mild diabetic retinopathy detection: an experimental study. bioRxiv (2019) 39. B. Tymchenko, P. Marchenko, D. Spodarets, Deep Learning Approach to Diabetic Retinopathy Detection, Machine Learning (Cornell University, Ithaca, 2020) 40. S. Qummar et al., A deep learning ensemble approach for diabetic retinopathy detection. IEEE Access 7, 150530–150539 (2019) Ms. Sowmiya Muruganantham  completed her Bachelor’s in Electronics and Communication Engineering at PSG College of Technology, Coimbatore (2013). She obtained her master’s degree from Coimbatore Institute of Technology, Coimbatore (2015), and is currently pursuing PhD from Anna University, Chennai. She is working as an assistant professor, in PSG Institute of Technology and Applied Research, Coimbatore. Her major interest areas include communication systems and medical image analysis and deep learning. She is a life member of the Indian Society of Technical Education.

Dr. B.  Banu Rekha  is currently an associate professor, Biomedical Engineering, PSG College of Technology, Coimbatore, India. She has done her B.E. degree in Electronics and Instrumentation Engineering from Government College of Technology, Coimbatore; M.E. degree in Medical Electronics from Anna University, Chennai; and her doctoral degree from PSG College of Technology, Coimbatore, in the area of sleep research. Her areas of interest are biosignal and image processing, medical informatics, and device design for healthcare and energy management. She has nearly 20 publications in international and national journals and conferences.

Deep Learning in Smart Applications: Approaches and Challenges

73

Prof. Dr. R.  Kanthavel  is currently working in King Khalid University, KSA, with more than 20  years of teaching and research experience in the field of information and communication engineering. He has the credit of more than 100 research articles in peer-reviewed international journals and many engineering text books. His areas of interest include intelligent modern communication networking, cooperative communication, machine learning, and data science.

Smart Metering Using IoT and ICT for Sustainable Seller Consumer in Smart City L. Sathish Kumar, M. Ramanan, Jafar A. Alzubi, P. Jayarajan, and S. Thenmozhi

1  Introduction Smart networks, the seek power, access management, infrastructure development, renewable storage sources, and EV charging are key aspects of potential electricity systems. The energy demand patterns have shifted, becoming more volatile, due to the impact of implementing renewable energies, which can be used for consumption itself, reducing the network requests, developing smart devices that can be controlled by means of control strategies, by programming, modulating, anticipating, delaying, determining highly flexible requirement, as well as finally penetrating the significant growth of renewable energy. The development of the electric grid introduced a new concept to be defined: the Seller Consumer. The electricity consumer will potentially use electricity while at the same time generate it and determine whether to use some of the energy directly or to export the excess to the smart infrastructure. It has brought in a major shift in the electricity sector, which would have to offset the generation volatility in the short term, stemming also from the arrival of the propositional consumer into the infrastructure. The order for electricity from loads, which can be handled and monitored centrally, has

L. Sathish Kumar (*) School of Computing Science and Engineering, VIT Bhopal University, Sehore, India M. Ramanan Department of Physical Sciences and IT, Agricultural Engineering College and Research Institute, TNAU, Coimbatore, India J. A. Alzubi Faculty of Engineering, Al-Balqa Applied University, Salt, Jordan e-mail: [email protected] P. Jayarajan · S. Thenmozhi Department of ECE, Sri Krishna College of Technology, Kovaipudur, Coimbatore, India © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 R. Maheswar et al. (eds.), Challenges and Solutions for Sustainable Smart City Development, EAI/Springer Innovations in Communication and Computing, https://doi.org/10.1007/978-3-030-70183-3_4

75

76

L. Sathish Kumar et al.

the primary purpose of reducing electricity consumed costs by means of trade decisions centered on demand variations over period. Application and regulation and flow of energy have larger implications for domestic consumers, and actions can be more readily replicated and repetitive, whereas in a building used for business operations, the factors are numerous and nuanced in regard to the nature of operation function, employees, and schedules. The smart meters are the core of integrated networks allowing greater convergence of electricity and ICT.  The smart meters and related technology allow real-time, two-way contact among suppliers and consumers—thereby allowing more efficient energy flow management and engagement. For this cause, we evaluated the activities of the residential consumers, Seller Consumer, with this chapter to arrive at the description of a new paradigm of the electricity grid with blockchain integration.

2  S  ustainable Seller Consumer and Requirement Side Maintenance A core factor in the implementation of smart networks is the reliability of the electricity provided and needed by customers, which can be accomplished by ICT, information and communications technologies, IoT, and smart meters connected with smart networks. Consumers of the smart city should not be mere consumers but Seller Consumers. After production they will have a constructive role: micro combined heat and photovoltaic panels (PVs) providing heat or electricity for households and individuals or communities [1]. This module includes the alternative current and direct current interface, RES multidirectional converters, the EVs bidirectional converter, control circuit, as well as other related electronics for electricity. Figure 1 represented the smart home network. This should enable flow velocity and smart infrastructure to be handled in a bigger way adaptable, effective approach [2]. Throughout the demand side response (DSR) concept, the requirement side of the most adaptable electrical grid is addressed in its consumption. The definition then became demand side management (DSM) and involves all of the energy system’s requirement side implementations [3]. Edge of request answer is seen as a crucial element in the future energy market [4]. Requirement side control and constructive competition in prospective decentralized power networks are anticipated to promote the capacity to play a major competitive position in managing energy. Smart meters form a significant possible connection between apartments and the electrical infrastructure. Energy users now have the potential to see how much they’re paying in near real time for their energy consumption [5]. This understanding means that customers will determine at what cost to spend for electricity, merely by altering their activity. Currently though, price variations are too limited for different consumers to produce sufficient incentives for customers to change patterns of consumption. Limited demand shifts and higher energy costs aren’t enough to inspire DSR investments.

Smart Metering Using IoT and ICT for Sustainable Seller Consumer in Smart City

Smart Network

e

om

H rt

a

Sm

77

Smart meter

Device Load

Power Resource

Fig. 1  Smart network for the n-number of smart home

Improved demand variance could render requirement side answer more efficient. In order to achieve a DSM, elements must be implemented that make the smart infrastructure and the electricity system efficient in order to maximize the wholesale sector’s electricity pricing and grid management capacity [6]. Advanced analytics model agents trading energy mostly with wholesale sector would have to adopt the optimization approach enforced for the better interests of Seller Consumers, e.g., one that maximizes income, recommends the smart agent’s commercial conduct, and defines the general guidelines, while selling electricity to the market is more beneficial. The handler shall take care of the following while optimizing earn. Therefore, the exchange advisor obtains a good profit by maximizing the proceeds it receives from the selling of electricity at each and every minute and the costs resulting from the price of its consumption itself, at a given period of every hour add the price of mortification of the micro systems, PV, and wind. As a consequence, network integration is a crucial factor, and smart infrastructure plays a critical position in the operation of the electricity sector, where various traders interact, serving energy producers, energy distributors, electric providers, and electric customers, and therefore the needs of all players in the energy system. Electric prices [7] should be calculated continuously, depending on production and consumption. Figure 2 shows dynamic energy trading the competitive energy sector in which brokers are no longer involved, mediators of power distribution providers, on both the wholesale and retail marketplaces, and design the activities of single consumers, their evaluation of energy thresholds, and risk attitudes. In reality, professional customers, via smart meters, identify various trends of use (e.g., climate-dependent)

78

L. Sathish Kumar et al.

Flow of Energy

Energy Trading Intermediate Software Fig. 2  Dynamic energy trading

and desires, which require specific variable or fixed pricing structures, In addition to efficiency, the sector will lead to the introduction of RES power at competitive electricity cost. That often refers to the Seller Consumers, through which the intelligent trader receives gains for the Seller Consumers (e.g., by means of RES or by means of electric vehicles which charge during financial hours and supply the network with electricity). The nature of the issue resides [7] in the price system: when supplying and consuming some certain volume of electricity, the agents impact the quality of electricity reimbursement, which in effect affects the optimum sum of sale and purchasing, and the consumer is unable to resolve the gap without sufficient capacity.

3  Unstable Smart Home Load and Smart Metering In this chapter, we would like to demonstrate how it is important, in the shift to complex centrally controlled structures, to incorporate smart metering, the monitoring of which, in the preferred interval of time, the domestic demand, gives indicators of delay or reduction of the need. The balance of supply and consumption within the intelligent power environment is practical and effective because there is a reasonably diverse and varied group of customers. The smart cities are created [8], smart communities of sufficient scale, in which effective demand control only at distribution stage is required to accommodate erratic production and requirement for electricity. That would lead to economic gains for the multiple entities at risk. The smart city will consist of thousands of families with various kinds of power loads, a collection

Smart Metering Using IoT and ICT for Sustainable Seller Consumer in Smart City

79

of buildings and homes with different modes of sustainable energy generation and associated energy use for electricity, domestic loads, EV fleets, and building control systems. Thus, by way of the Marketing Intelligent Agent Application, the monitoring and management mechanism among requirement and supply of energy, production itself, and on-site use is highly unpredictable; modeling should also take into account the incorporation of intelligent devices with real-time dynamic monitoring, consumer actions, energy conservation in the SG, and safe, reliable electric power generated as far as practicable through renewable energy sources (RES) and maintained to minimize the costs. The domestic energy consumption modeling is known [9–11]: this should begin from a certain amount of residences, taking into account the number and type of loads probability of having in each residence. The control scheme, in conjunction with both the defined optimization method, has first, will take into consideration the following goals: (a) Changing cost: the use of such products should be delayed over time, reducing the consumer’s satisfaction to the slightest disturbances. (b) Capacity reduction: carried out even under exceptional or emergency circumstances, because such are cases under which the implementation has adverse consequences on customers. Household loads such as a form of heat or electricity storage were also optimal for changing the power over time, like fridges that are operated discontinuously, usually with such a 20-min switch-on and a 40-min turn-off. This is necessary to change the running time under different situations, because it has little significance in the again activation. In fact, conscious habits can be triggered, recommending that it not opened constantly over this time period. Many loads such as vessels cleaner, washing appliances, and driers all are, of course, often ideal for starting adjustments, and it is important to know the form of load, the average running period, the Seller Consumers’ expectations, and how to adjust the begin and end loads to maximize electricity usage. By means of smart metering, it’s indeed to know whenever the high energy are starting to work, and by means of a channel, with smart agent controller design, the household loads can be managed and distributed over time, also in regard to the priorities laid by the individual Seller Consumer. From Table 1, remember that household loads like lamps and television are deemed necessary, and they continue to function during an evacuation. The use of a mobile and screen in a household is determined while the percentage of individuals who may be at household and involved represents the common behavior of people who live by adopting their respective lifestyle patterns. By means of a smart meter [12], it’s really necessary to monitor the energy usage and usage patterns of different electrical devices, whereby the information could be generalized as important to evaluate the managerial method of household devices with suitable consolidated weekly statuses in co-relation to their standard usages (low consumption at morning, increased consumption at night). It has been represented by Figs. 3, 4, 5, and 6. The numbers are focused on calculating a significant data collection emerging from the houses; via the smart meter, this occurs in each house in a timely manner.

80

L. Sathish Kumar et al.

Table 1  Priority of smart home load Smart home loads Night lightning Play store Refrigerator Emergency lightning Day lightning Shades Steam oven A/C Washing machine Coffee maker Micro oven Washroom devices Home security Vacuum cleaner Kitchen appliances Dryers

Time (min/h) 8 h 1–4 h 20 min 10 min 1 min 1 min 1–30 min 6–12 h 20–120 min 2 min 20–60 min 5–10 min 4–8 h 30 min 5 min 5–15 min

Smart network priority 0 1 0 0 1 1 1 1 1 1 2 2 0 2 1 2

Seller consumer priority 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

70

Load (W)

60 50 40 30 20 0

0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00 13.00 14.00 15.00 16.00 17.00 18.00 19.00 20.00 21.00 22.00 23.00

10

Time (Hr) Fig. 3  Heater daily load in April to May

Such regular patterns contribute to the classification of devices by use, including behavior that according to the study [11, 13] is also fixed activities or stochastic activities or smooth activities or of the sort as seen in Fig. 7 and Table 2. For a stochastic simulation, the vector likelihood of using the system during the day could be taken into account by assigning a behavior profile to every unit in the model [14, 15]. A set of devices is allocated model to the consumption profile, the smart agent of each “n” building, which is associated with one of the regular activity profiles, which quantifies the likelihood of the defined operation which must be calculated as a feature of the day’s selected time periods. The power consumed by the common “n” home automation will take into consideration the entirety of the

Smart Metering Using IoT and ICT for Sustainable Seller Consumer in Smart City

81

180 160

Load (W)

140 120 100 80 60 40 0

0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00 13.00 14.00 15.00 16.00 17.00 18.00 19.00 20.00 21.00 22.00 23.00

20

Time (Hr) Fig. 4  Fan daily load in April to May

120

Load (W)

100 80 60 40

0

0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00 13.00 14.00 15.00 16.00 17.00 18.00 19.00 20.00 21.00 22.00 23.00

20

Time (Hr) Fig. 5  Day time lighting in April to May

modules in consideration of the data received by the actual loads, not operating 7 days, and the loads themselves being run concurrently. Such elements lead to the determination of a predictive intelligent device usage model, factors that constitute the control method the intelligent operator framework must execute for the “n” home automation. Such factors ultimately will decide the optimum choice between the change and decrease power, as in Fig. 8, with particular regard to condition. Particularly throughout the situation of load mitigation, Seller Consumers will have a non-smooth scenario causing an annoyance, while at the same period would have a benefit for network reliability that must be directly measured and/or commercialized, by stable and optimized flow of energy contracts.

82

L. Sathish Kumar et al.

80 70

Load (W)

60 50 40 30 20 0

0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00 13.00 14.00 15.00 16.00 17.00 18.00 19.00 20.00 21.00 22.00 23.00

10

Time (Hr) Fig. 6  Load of washing machine in April to May

200

Smart Home Load (%)

180 160 140 120 100 80 60 40 20 0 0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 18.00 20.00 22.00 24.00 Time (Hr.) Fig. 7  Types of activities of smart home load

4  Sustainable Seller Consumers The electricity market faces enormous uncertainty and needs to handle a plurality of details, and this includes, in real time, the need to build a modern electrical network, versatile and stable and efficient and, efficiently, providing possibilities for new technologies and increased financial benefit. Through new applications and technology related with the Internet of Things (IoT), we could inculcate wisdom in previous infrastructures, including in a smart network, with connected sensors capable of connecting with each other even though over large ranges, but all this would have to

Smart Metering Using IoT and ICT for Sustainable Seller Consumer in Smart City

83

Table 2  Types of activities of smart home load Time (h) 0 2 4 6 8 10 12 14 16 18 20 22 24

Smooth activities 18 16 18 15 19 18 19 18 17 18 18 17 18

Fig. 8  Architecture of intelligent agent

Stochastic activities 30 28 29 45 60 58 60 55 42 60 35 33 35

Fixed activities 39 37 49 53 76 87 85 86 80 94 80 60 40

Smart Meter

Device Load

Priority smart home load

Smart Meter

Load change

Intelligent Contract

Intelligent intermediate Software

be achieved with optimum protection. In addition, the protection of networks associated with the energy sector, IoT, SG, smart networks, and smart metering should be the focus and among the most significant improvements not just for Seller Consumers as well as for distributors and decision experts. Intruders can change the communication and accessing data, too. This problem needs to be debated not only because of intentional attacks as well as for human mistakes or inadequacies in equipment [16]. The electricity details of consumers must be protected, because

84

L. Sathish Kumar et al.

consumers may not wish to exchange their knowledge with unauthorized persons or external agencies [17]. The protocols of wireless networking, used for transferring data, are much more fragile because of wind. The blockchain technology is known for its association with cryptocurrency but in turn makes IoT very feasible. And if this form of blockchain software is being studied, it demonstrates the possibility that a frame-link network will operate inside the electricity infrastructure [18]. A big disadvantage of the new SG is that local consumers will have to go into the grid to pay the entire retail cost for the electricity produced by their neighbor, redistributing surplus electricity, often with substantial costs, for example, the customer who would be on the different roadside due to a nearby electric supply. Blockchain had potential to transform the energy grid, by the control of flow of energy, in any financial trade, in sustainable energy credits. The aim would be to achieve a democratic framework, in which each consumer may become Seller Consumer, without needing any further mediators. Blockchain is now a cryptographic ledger, and enhanced capacity is a block that has to be checked by a process that has to create/validate a hash, that is, a code composed of specific letters and numbers. Hash techniques are being used to transform random length information to a set value, producing a hash. Second section encrypted information could not be based on the similar hash integer or letter, nor would hash parameters provide details concerning the information’s content [19]. If the checked and accurate version of a node, among other channel members, is accessed from many other frames, the BC may expand. As electronic device can interact via IoT, users can send and back up data and details from peer-to-peer the BC [20]. The electricity market varies from cost market as it must still provide the actual commodity itself. It has been mentioned clearly in the following figure (Fig. 9). Exchanges are indeed not only about parameter and knowledge, but it’s also about the exchange in electricity provided across the grid infrastructure. At the other hand, renewable energy may be used if energy generation becomes inadequate. Thus, blockchain software could monitor connectivity flows and backup networks. Intelligent contracts can be used to handle the surplus and virtual electricity stations Fig. 9  Blockchain power interface

Energy Trade

Smart Grid

Smart Metering

IoT

Intelligent Applications

Intelligent Contract Blockchain

Smart Metering Using IoT and ICT for Sustainable Seller Consumer in Smart City Fig. 10  Process model of blockchain

85

SSC Intelligent Contracters

Near by Consumer

Power Trade

Smart Grid

SSC

activities. Blockchain can control and monitor electric networks via the smart meters and IoT [21]. Preferably, all sharings are rendered on the basis of blockchain software contracts, e.g., on the basis of predetermined specific guidelines on consistency, size, quantities, etc., a distributed distribution mechanism without such third-­ party mediators, without brokers. The trade is focused on Renewable Energy Sources Coin (RESCoin), which we can see in Fig.  10, extracted from Seller Consumers’ output, and will therefore share electricity with nearby consumers in order to minimize loss in energy transport. It would be important to incorporate blockchain software in the field of smart homes that connect with each other and with other homes beyond the smart home and provide a means of interface capable of sharing and processing similar data. Seller Consumers may utilize the blockchain platform to collect meter measurements and billings with connection to the smart meters. Another of the key benefits of a marketing strategy focused on blockchain is that all energy supplied to networks could be allocated to specific consumers in limited periods. This implies that all of the energy generated and used could be tuned to delicate parameters with great precision. The physical electricity such will proceed to flow directly first from nearby electricity source to the end consumer, without the requirement of intermediaries and even lack of power control.

5  Conclusion Seller Consumers are important method for improving a sensible future and using the smart network more effectively. Through minimizing transmission costs, they will effectively leverage wind power and solar power resources and drastically decrease the volume of electricity drawn from the grid. Additionally, Seller

86

L. Sathish Kumar et al.

Consumers can offset the market volatility of wind power and solar energy generation only via their electric vehicle batteries. So, there would be large peaks in excessive demand when controlled by complex management algorithms. The electricity system will not be competitive in this manner. So the form of network on which the electricity market structure should calculate has been defined through this article, and would be able to control the requirements effectively by smart metering, either delays or significant reduction of the whole load itself through causing conscious actions on Seller Consumers, in order to maintain the actual objective purpose of full commercial benefit. Excluding distributors but exclusively by Seller Consumers, such transfers will be done efficiently through only blockchain. By comparison, accumulated power may be used if there is inadequate electricity generation. Thus, blockchain algorithms may monitor the network transfers and backup systems directly. Intelligent contracts can be used to handle the trying to balance and virtual power stations tasks. Blockchain will handle and maintain SG and the electricity market through the IoT and smart meters.

References 1. N. Leskinen, J. Vimpari, S. Junnila, Using real estate market fundamentals to determine the correct discount rate for decentralized energy investments. Sustain. Cities Soc. 53, 101953 (2020) 2. K. Abdollah, W. Su, T. Jin, A machine learning based cyber attack detection model for wireless sensor networks in microgrids. IEEE Trans. Ind. Inf. 17(1), 650–658 (2020) 3. M. Mier, C. Weissbart, Power markets in transition: decarburization, energy efficiency, and short-term demand response. Energy Econ. 86, 0140–9883 (2020) 4. N.J.  Hewitt et  al., Domestic Demand-Side Response: The Challenge for Heat Pumps in a Future UK—Decarbonised Heating Market, Renewable Energy and Sustainable Buildings. Innovative Renewable Energy (Springer, Cham, 2020) 5. F. Teng, Z. Ding, Z. Hu, P. Sarikprueck, Technical review on advanced approaches for electric vehicle charging demand management, part I: applications in electric power market and renewable energy integration. IEEE Trans. Ind. Appl. 56(5), 5684–5694. https://doi.org/10.1109/ TIA.2020.2993991 6. H. Park, M.K. Sim, D.G. Choi, An intelligent financial portfolio trading strategy using deep Q-learning. Expert Syst. Appl. 158, 113573 (2020) 7. A.Y. Rodríguez González, M.P. Alonso, F. Lezama, L. Rodríguez, E.M. de Cote, E.F. Morales, L.E. Sucar, D.D. Crockett, A competitive and profitable multi-agent autonomous broker for energy markets. Sustain. Cities Soc. 49, 101590 (2019). ISSN 2210-6707 8. A.K. Erenoğlu, İ. Şengör, O. Erdinç, J.P.S. Catalão, Chapter 6: Blockchain and its application fields in both power economy and demand side management, in Blockchain-based smart grids, ed. by M. Shafie-khah (Academic, pp. 103–130, 2020). ISBN 9780128178621 9. S. Yilmaz, A. Rinaldi, M.K. Patel, DSM interactions: what is the impact of appliance energy efficiency measures on the demand response (peak load management)? Energy Policy 139, 111323 (2020). ISSN 0301-4215 10. R.J. Hafner, S. Pahl, R.V. Jones, A. Fuertes, Energy use in social housing residents in the UK and recommendations for developing energy behaviour change interventions. J. Clean. Prod. 251, 119643 (2020). ISSN 0959-6526

Smart Metering Using IoT and ICT for Sustainable Seller Consumer in Smart City

87

11. P.M. Mammen, S. Mehrotra, H. Kumar, K. Ramamritham, Want to reduce energy consumption, which floor should I prefer?, in Proceedings of the Eleventh ACM International Conference on Future Energy Systems (Association for Computing Machinery, New York, NY, USA, 2020), pp. 534–538 12. J. Barton, M. Thomson, P. Sandwell, A. Mellor, A domestic demand model for India. Advances in energy research, vol. 1, in Springer Proceedings in Energy (Springer, Singapore, 08 May 2020). ISBN: 978-981-15-2666-4 13. http://www.eerg.it/resource/pages/it/Progetti_-_MICENE/compendio_misure consumi_ elettrici.pdf 14. L. Yang, P. Xie, C. Bi, R. Zhang, B. Cai, X. Shao, R. Wang, Household power consumption pattern modeling through a single power sensor. Renew. Energy 155, 121–133 (2020). ISSN 0960-1481 15. D.N. Molokomme, C.S. Chabalala, P.N. Bokoro, A review of cognitive radio smart grid communication infrastructure systems. Energies 13, 3245 (2020) 16. J.  Torriti, Temporal aggregation: time use methodologies applied to residential electricity demand. Util. Policy 64, 101039 (2020). ISSN 0957-1787 17. G. Dileep, A survey on smart grid technologies and applications. Renew. Energy 146, 2589– 2625 (2020). ISSN 0960-1481 18. Z. Yan, J.L. Hitt, J.A. Turner, T.E. Mallouk, Renewable electricity storage using electrolysis. Proc. Natl. Acad. Sci. U. S. A. 117(23), 12558–12563 (2020) 19. L. Wan, Z. Zhang, J. Wang, Demonstrability of Narrowband Internet of Things technology in advanced metering infrastructure. J Wireless Commun. Netw. 2019, 2 (2019) 20. M.  Ma, Z.  He, Q.  Xu, X.J.  Li, Design and development of smart home sensing supported by blockchain technology, in Proceedings of the 2019 7th International Conference on Information Technology: IoT and Smart City (ICIT 2019) (Association for Computing Machinery, New York, NY, USA, 2019), pp. 525–530 21. H.  Jaakkola, J.  Henno, J.  Mäkelä, B.  Thalheim, Artificial intelligence yesterday, today and tomorrow, in 2019 42nd International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO), Opatija, Croatia, 2019, pp. 860–867 Dr. L. Sathish Kumar  is an assistant professor in the School of Computer Science and Engineering at VIT Bhopal University, Bhopal, Madhya Pradesh. He holds a doctorate in Computer Science Engineering and master degrees in Computer Science from Alagappa University, Karaikudi, and bachelor’s degree in Information Technology from APSA College, Tiruppattur. He has 6 years of teaching experience from various institutions. He has 27 publications in highly reputed international journals and conferences like SCI, Scopus, Springer, and Web-of-Science. He has published five textbooks entitled Desktop Publishing, Desktop Publishing Second Edition, Artificial Intelligence, Data Communication and Networking, and Java Programming. He has received the Best Young Faculty Award from Novel Academy, Pondicherry. He is an active researcher in the field of deep learning, machine learning, medical image processing, and data science.

88

L. Sathish Kumar et al. Dr. M. Ramanan  received his B.Tech (Information Technology) and M.E. (Computer Science and Engineering) from Anna University, Chennai, and Ph.D. (Computer Science and Engineering) from Anna University, Chennai. He has 8 years of teaching experience. He is currently working as teaching assistant in the Agricultural Engineering College and Research Institute, TNAU, Coimbatore, India. He has published many research papers in national/international conferences and journals. His research interest is mainly focused on cloud and IoT.

Jafar A.  Alzubi  is an associate professor at Al-Balqa Applied University, School of Engineering, Jordan. He received his Ph.D. degree in Advanced Telecommunications from Swansea University, Swansea, UK (2012); Master of Science degree (Hons.) in Electrical and Computer Engineering from New York Institute of Technology, New York, USA (2005); and Bachelor of Science degree (Hons.) in Electrical Engineering, majoring in Electronics and Communications, from the University of Engineering and Technology, Lahore, Pakistan (2001). Jafar works and researches in multi- and interdisciplinary environment involving machine learning, classifications and detection of Web scams, Internet of Things, wireless sensor networks and its security, and using algebraic-geometric theory in channel coding for wireless networks. He managed and directed few projects funded by the European Union. A cumulative research experience for over 10 years resulted in publishing more than 50 papers in highly impacted journals. Dr. P. Jayarajan  has completed his B.E. (EEE) from Madurai Kamaraj University in the year 2004 and M.E. (Applied Electronics) from Anna University in the year 2008 and is pursuing Ph.D. in the field of Wireless Sensor Network under Anna University. He has about 12 years of teaching experience and is presently working as an associate professor in the Electronics and Communication Engineering Department, Sri Krishna College of Technology, Coimbatore. He has published 20 papers at international journals and conferences. His research interest includes wireless sensor network, modeling, and simulation.

Smart Metering Using IoT and ICT for Sustainable Seller Consumer in Smart City

89

S.  Thenmozhi  completed her B.E. degree in Electronics and Communication Engineering and M.E. degree in Communication Systems in 2011 and 2013, respectively, from Anna University. She is currently a part-time research scholar and working as an assistant professor at Sri Krishna College of Technology, India. Her research areas include mobile ad hoc networks, wireless sensor networks, and security and privacy issues in IoT.

Deep Learning-Based Activity Monitoring for Smart Environment Using Radar N. Susithra, G. Santhanamari, M. Deepa, P. Reba, K. C. Ramya, and Lalit Garg

1  Introduction It is evident from the advent of various technological advancements including IoT that smart cities determine the nation’s future in terms of economic salvation, resource management, connectivity across the nation, optimum reach of all the utilities to the nook and corner of the country, environmental safety, and societal safety, and the list is endless. This aspiration has kindled a growing interest among the entrepreneurs, engineers, and investors in taking part in the race to build smart cities. Smart cities demand development of sustainable monitoring systems for all kinds of environments. Urbanization, sustainable development, and inclusive growth necessitate the efficient and intelligent utilization of natural resources for a better living [1]. The development of a smart city relies on strongly utilizing information and communications technology (ICT) to gather data on a large scale, to facilitate in planning and development of cities, and to arrive at optimal engineering decisions. ICT also plays a major role in the integration of the different infrastructure layers of N. Susithra (*) · G. Santhanamari · M. Deepa · P. Reba Department of Electronics and Communication Engineering, PSG Institute of Technology and Applied Research, Coimbatore, India e-mail: [email protected] K. C. Ramya Department of EEE, Sri Krishna College of Engineering and Technology, Coimbatore, India e-mail: [email protected] L. Garg Department Computer Information Systems, Faculty of Information and Communication Technology, University of Malta, Msida, Malta e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 R. Maheswar et al. (eds.), Challenges and Solutions for Sustainable Smart City Development, EAI/Springer Innovations in Communication and Computing, https://doi.org/10.1007/978-3-030-70183-3_5

91

92

N. Susithra et al.

Fig. 1  Characteristics of a smart city

a smart city and paves way for interoperability among different stakeholders to make it more liveable and to safeguard the environment [2]. Smart city builds enhanced real environments connected to the virtual information world, with positive impacts on the different aspects of health and quality of life of the citizens. All these different aspects of the smart city framework are categorized into six major characteristics. Figure 1 shows the major characteristics of a smart city [3]. These characteristics help in assessing the performance of the smart city. Some of the main indicators that are linked to these characteristics in the development of smart cities are traffic monitoring for smart transportation [4], advanced driver assistance systems (ADAS) [5], ICT-driven smarter healthcare, elderly care [6, 7], and initiating sustainable methods for human-wildlife coexistence [8]. Traffic Monitoring: Video surveillance systems are used to monitor, track, and extract information from moving vehicles and pedestrians enabling smart traffic monitoring solutions like traffic assessment, pollution avoidance, route optimization, and accident/collision prevention and ensure safety on the road. Unlike the most commonly used vision-based monitoring solutions that are often affected by lighting and weather conditions, radar (radio detection and ranging)-based monitoring systems enable wide-range detection and are least affected by adverse climatic conditions [9]. Advanced Driver Assistance System (ADAS): The increasing rate of accidents can be attributed to traffic congestion, rash driving, drunken driving, carelessness, overriding the traffic rules, etc. Advanced driver assistance system (ADAS) provides intelligent driver assistance using in-vehicle sensors with features including fatigue warning, parking assistance, blind-spot detection, collision/obstacle detection, detection of road defects, accessing different controls through gesture recognition, etc. It also aids in emerging autonomous vehicle systems with its intelligent features. Owing to the high-speed processing of radar-based systems, these intelligent applications are developed using automotive radar sensors. Radar has the

Deep Learning-Based Activity Monitoring for Smart Environment Using Radar

93

ability to spot the dynamic and stationary objects in the vicinity of the vehicle while locating the vehicle along with its speed and trajectory information. Undoubtedly, today’s modern vehicles use radar-based sensors to get the most out of echo signals [10]. Elderly Care: One other aspect of the smart city development is paving the way for prolonged independent living and a sense of security for elderly people or patients who have chronic illness or less assistance from caregivers, with the help of uninterrupted monitoring. Continuous monitoring of the physiological vital signs such as heartbeat, breath pattern, oxygen saturation (SpO2), and sleep are very crucial in discriminating between normal daily activities and abnormalities in the patients. Medical equipment with bodily contact causes distress to the patients. Whereas, microwave Doppler radars and ultra-wideband radars act as noncontact biosensors in monitoring the physiological vital signs and are most suitable for assisted living purposes. Moreover, the radars could be able to detect the elderly people even if there are physical obstructions and also at poor ambient lighting conditions [11]. Human-Animal Conflict: The widespread urbanization has led to the depletion of the natural dense forests and has created a negative impact on wildlife. Animals straying into human habitats, animals foraging for food, and animal-vehicle collision are prominent issues in roads/rails or places adjoining the human habitats. Smart and sustainable development ensures conservation of flora and fauna while helping the coexistence of humans and animals. For this purpose, radars are mounted at the animal crossing areas, and warning signals are sent to the vehicles in advance in order to protect both the parties concerned. Micro-Doppler signatures from radars are most widely used in order to distinguish between animals and humans and encourage human-animal cohabitation [12]. Having cited different aspects of smart cities, it is observed that radar (radio detection and ranging) plays a vital role in various domains. Radar is a long-range sensing system used to detect and track the target based on the reflection of electromagnetic waves from the target. Initially, radars were extensively used for military applications such as surveillance, detection and tracking of aircraft, missiles, satellites, etc. Recently radar sensor systems have also been used for civilian purposes. Furthermore, advanced radar systems play a vital role in the sustainable development of smart cities. Due to advancements in integrated circuits technology, a complete radar system can now be integrated into a small chip, thereby reducing the production cost. Moreover, unlike other sensing systems such as vision-based sensors, cameras, and LIDARs, radar systems are more affordable, low-cost monitoring systems that are compact and more reliable, provide information-rich representation about the target’s environment, and suit for large-scale deployment [13]. Having obtained the radar information, it is essential to have an efficient computing system to recognize the data. Machine learning (ML) is the key tool for classifying the echo signals. Conventional ML techniques such as SVM and random forest are used to classify the targets and their activities. However these techniques suffer from several drawbacks. (1) In ML, the features are extracted manually. This is

94

N. Susithra et al.

laborious and also causes human errors. (2) Low-level statistical task-specific parameters such as mean and variance are used in the classification process, and hence the model trained with these parameters will not give accurate results if it is applied to a new scenario [14]. Thus conventional ML methods might often fail to provide a robust system, and deep learning (DL) overcomes these limitations. Using DL, the features are extracted automatically through hierarchical structures. Parallel processing and fast computing using advanced processors and graphic processing units (GPU) help the DL algorithms in providing excellent learning ability. This chapter brings together the concepts of radars, their applications, and the different ML, DL, and hybrid deep learning methods that can be used to classify the radar signals. We examine the role and potential of radars and the learning techniques as one of the key pillars of smart mobility management in smart cities. The remainder of this chapter is organized as follows: Sect. 2 provides the theoretical concepts of different types of radars along with their potential applications. Sect. 3 explains the micro-Doppler signatures and the time-frequency analysis for radar signal processing. Sect. 4 provides a comprehensive view of machine learning and deep learning algorithms used for the classification of targets. This is followed by the future research scope in Sect. 5. A summary of the chapter is provided in Sect. 6.

2  Common Types of Radar Common radar types used for object detection and tracking are continuous-wave radar, pulse radar, moving target indicator radar, pulse Doppler radar, frequency-­ modulated continuous-wave radar, pulse-modulated continuous-wave radar, ultra-­ wideband radar, stepped-frequency continuous-wave radar, synthetic aperture radar, and interferometric SAR. Radar sensor systems at mm-wave ranges are also available for applications such as automotive which provides high resolution in detecting targets. This section explains the fundamental concepts and characteristics of the commonly used radar types [15, 16].

2.1  Continuous-Wave Radar In continuous-wave radar (CW radar), a continuous wave cosω2πfct is radiated by the transmitting antenna, and a separate antenna is used for receiving the echo signals. Based on the shift in frequency of the received signals (Doppler shift), the velocity of the moving target is identified. At the receiver, the mixer helps to extract the Doppler components when the received signals and the transmitted signals are given as input. It is then converted to digital signal for further processing. If there is no moving target, the received signal shows no Doppler shift, and the echo due to

Deep Learning-Based Activity Monitoring for Smart Environment Using Radar

95

Fig. 2  CW radar system

static targets and clutter will be centered at fc. In most of the CW radars, an I/Q (in-­ phase/quadrature) demodulator is present which will separate the I and Q components. Figure 2 shows a CW radar system. At the receiver, after converting to intermediate frequency (IF) and performing low-noise amplification, the received signal is gated to remove the clutters, and they are processed by low-bandwidth narrow-bandpass filters. The Doppler amplifier eliminates the frequencies reflected from the stationary targets, and it is designed to amplify the Doppler frequency from the reflected target only. The lower cutoff frequency of this amplifier is selected such that it filters out the DC components from the stationary objects, while passing the low Doppler frequencies, and the upper cutoff frequency is selected such that it also amplifies the expected highest Doppler frequency. Thus the Doppler amplifier is a wideband amplifier. However, a wideband amplifier may introduce more noise. Hence, instead of wideband amplifier, a bank of narrowband filters are used. The center frequencies of these filters are chosen such that they cover the expected range of Doppler frequencies. The continuous-wave radar cannot measure the distance, since there is no timing mark. Moreover, these radars cannot differentiate more than one target moving at equal velocity. Applications of CW Radars  It is used in police speedometer to monitor the speed of the vehicles. In railways, the conventional tachometers are replaced by continuous-­ wave radar-based speedometers to eliminate the measurement errors caused due to wheel slip or wheel slide when brake is applied. Table  1 lists few commercial continuous-­wave radars available [17].

2.2  Pulse Radar Pulse radar systems transmit and receive short-duration pulses in a half-duplex manner [18]. Figure 3 shows the pulse radar waveform. The parameters that define the pulse radar waveform are the carrier frequency, pulse width t, pulse repetition

K-LC2 RADAR TRANSCEIVER

K-MC5_LP

Model CDM324

Manufacturer Cristek Interconnects RFbeam Microwave GmbH RFbeam Microwave GmbH

Table 1  Commercial continuous-wave radars Range and power Frequency consumption 24 GHz 15 m Low-power consumption 24 GHz Less than 30 mW power consumption 24 GHz >10 m for a moving person >26 m for moving car

Frequency range min/max Application 24.125 GHz/24.25 GHz Noncontact detection, moving object detection 24.125 GHz/24.25 GHz Object speed measurement, industrial applications 24.125 GHz/24.25 GHz Direction-sensitive movement detectors, industrial applications

96 N. Susithra et al.

Deep Learning-Based Activity Monitoring for Smart Environment Using Radar

97

Fig. 3  Pulse radar waveform

interval (PRI), and pulse repetition frequency (PRF). The pulse radars unambiguously estimate the range of the target based on the round-trip time delay τ (the time taken to return back after hitting the target). Also, it measures long target ranges using high-power transmitters. However, the pulse radars cannot measure the Doppler frequency. The range of the target measured by a pulse radar is given by R = ct , where τ 2 is the round-trip time delay. The ambiguities in range measurement may be prevented by assuming that the pulse repetition interval PRI > t = 2R / c . The maximum unambiguous range that can be estimated is Rmax = c ´ PRI , where c is the 2 speed of light. In terms of the pulse repetition frequency (PRF), Rmax = c . 2 PRF ct If there are two targets, then the range of first target is estimated as R1 = 1 , 2 ct 2 and the range of second target is R2 = , where τ1, τ2 are the round-trip time 2 delay from target 1 and target 2, respectively. The radar unambiguously detects the range of these two targets only if R1 - R2 ³ ct , where t is the width of the pulse. 2 This is called range resolution. However, pulse radars suffer from large blind spot. Based on the Nyquist sampling criterion, in order to avoid ambiguities, the maximum frequency that can be measured is ±PRF . Hence, the maximum Doppler 2 frequency that can be unambiguously measured is fdmax = ± PRF . Therefore, the 2 maximum detection range is inversely proportional to PRF, while the maximum Doppler frequency is directly proportional to PRF. The pulse radars are further classified as low PRF, medium PRF, and high PRF radars to measure the range and/or velocity. However, there are no fixed frequency intervals to define these radars. Basically pulse radars are capable of estimating only the range. However, the variants of pulse radars utilize the Doppler frequency shift to identify the small moving targets in the presence of fixed targets/clutters, for example, moving target indicator (MTI) and pulse Doppler radars. Low PRF (LPRF)

•  Accurate long unambiguous range measurement. But exerts ambiguities in Doppler measurement • The moving targets in the presence of heavy background clutters (e.g., moving targets in the ground) are difficult to identify •  Moving target indicator (MTI) radar is a low PRF radar

98

N. Susithra et al.

High PRF (HPRF)

•  Provides unambiguous Doppler measurement over a large velocity range but shows some range ambiguities • In HPRF radar, many pulses are transmitted within a particular interval of time, and it leads to high average transmit power. This in turn increases the detection range •  Pulse Doppler radar is a HPRF radar Medium PRF •  Ambiguous range and Doppler measurements (MPRF) •  MPRF radar shows better performance, when the targets move at low speeds • Other merits: Adequate average transmit power, e.g., airborne fire control radars

2.3  Moving Target Indicator Radar Moving target indicator (MTI) radars distinguish the moving targets from the clutters and provide unambiguous range information but ambiguous velocity information. These radars come under the category of LPRF radars. Figure  4 shows the block diagram of MTI radars. Let the reference signal generated by the stable oscillator be Vref = A1 sin 2πf0t. The received echo signal is given by Vreceived = A2 sin ( 2p ( f0 + fIF ± fd ) t - 4p fo Ro / c ), where Ro is the range of the target. At the output of the mixer, the difference signal is Vdifference = A3 sin ( 2p ( fIF + fd ) t - 4p fo Ro / c ). For fixed objects, the return signal shows zero Doppler shift, and therefore Vdifference won’t vary with time. But, if the target object is in motion, the Doppler shift is nonzero, and therefore Vdifference varies with respect to time. Moreover, if fd < 1 , PRI then the information about fd can be easily obtained from one pulse. Instead, if fd > 1 , then the information about fd is extracted from many pulses. PRI If Vdifference is displayed on an A-scope (amplitude vs time), the amplitude of the signals reflected from the fixed targets does not change in the successive pulses, whereas the reflected signals from the moving target show variation in amplitude. In order to extract the Doppler information from the reflected signal, delay line cancellers are used in MTI radars. The delay line canceller delays the return echo signal by one pulse time. The present signal is then subtracted from the delay line output. Thus, the echo signals from the stationary targets are cancelled out, whereas the difference in two consecutive signals from the moving targets remains. Applications of MTI Radars  MTI radars are used as air surveillance radars for tracking the target accurately in the presence of clutter.

Deep Learning-Based Activity Monitoring for Smart Environment Using Radar

99

Fig. 4  MTI radar

2.4  Pulse Doppler Radar The pulse Doppler radars fall under the category of high PRF radars. The average transmit power of pulse Doppler radars is quiet high. The pulse Doppler radars show ambiguities in detecting long ranges. Generally, pulse Doppler radars provide unambiguous velocity information and ambiguous/unambiguous range information. The Doppler radars use range gate and Doppler filters to measure the Doppler frequency accurately. The pulse Doppler radar exploits the Doppler shift information to estimate the velocity of the target. Analog or digital signal processing may be employed at the receiver of the pulse Doppler radar as shown in Figs. 5 and 6, respectively. For the analog signal processing, the output of the IF amplifier is gated into different time ranges and then filtered by a narrowband filter. The envelopes of all filter output are collected by a commutator. Based on the commutator output, the range and velocity information of the targets are computed. These narrowband filters can perform best, only if they possess the matched filter characteristics. The signal transmitted by a pulse Doppler radar is given as:

Vtx ( t ) = A1 cos ( 2p ft + q ( t ) )

(1)

where A1 is the amplitude of the transmitted signal, fc is the frequency of the transmitted signal, and θ(t) contains the phase information. The reflected signal from the target for a pulse Doppler radar is:



æ 4p Ro 4p x ( t ) æ 2R Vref ( t ) = A2 cos çç 2p fc t +q çt - o c l l è è

öö ÷ ÷÷ ø ø

(2)

100 Fig. 5  Analog processing steps of pulse Doppler radar

Fig. 6  Digital processing steps of pulse Doppler radar

N. Susithra et al.

Deep Learning-Based Activity Monitoring for Smart Environment Using Radar

101

where A2, fc, and λ are the amplitude, frequency, and wavelength of the return signal from the target. c is the speed of light. Ro is the nominal range between the radar and the target, and x(t) represents the relative displacement of the target. In the digital signal processing approach, the IF amplifier output is split into I and Q components. The I and Q waveforms are then converted into digital and then transformed into frequency domain using N-point FFT. N corresponds to the number of data samples considered at each range cell. The outputs are then envelope detected, and the range and velocity information of the targets are computed. Table 2 lists few commercial pulse Doppler radars available.

2.5  Frequency-Modulated Continuous-Wave Radar Frequency-modulated continuous-wave (FMCW) radar is a class of continuous-­ wave radar in which a continuous carrier wave, after frequency modulation by a sinusoid/triangular/sawtooth wave, is transmitted from the transmitting antenna. Figure 7 shows the block diagram of FMCW radar. The frequency of the continuous-­ wave signal changes linearly over a period of time, and it is called the chirp signal (Fig. 8). FMCW radar is helpful in estimating the range of the target at high accuracy. The parameters that define chirp signal are: Table 2  Commercial pulse Doppler radars [19, 20] Model RCR-50 OPS 241A

Manufacturer GE-Security OmniPreSense

BumbleBee radar system

The Samraksha Company

Fig. 7  FMCW radar block diagram

Range of Frequency sensing 5.8 GHz 15 m 24 GHz 1–12 m

5.8 GHz

10 m

Application To detect human-sized objects Traffic monitoring, speed/ direction reporting, presence detection, collision avoidance Robust intrusion detection, tracking by estimation of velocity and vibration monitoring

102

N. Susithra et al.

Fig. 8  Chirp signal

• fc starting frequency of the chirp. • fmax maximum frequency of the chirp. • Bandwidth BW is the difference between the starting and ending frequency of the chirp (fmax − fc). • Tc duration of 1 chirp and slope of the chirp S = BW . Tc At the receiver, the receiving antenna picks the delayed version of the transmitted chirp signal and is mixed with the transmitted signal resulting in the generation of an IF signal. The frequency of this IF signal is the beat frequency fbeat. IF signal is then converted into digital signal for processing. The range of the target can be calculated from the beat frequency given as fbeat = S t = S ´ 2 ´ Range / c, where τ is the round-trip time delay between the transmitted and the received chirp. Therefore the range of the target is given as:



Range =

fbeat ´ t S´2

(3)

If multiple objects are present in front of radar, then the radar receives multiple return signals. Each return signal, if multiplied with the transmitted signal, will result in a different IF tone (beat frequencies). Each IF is proportioned to the range corresponding to the respective object. The range resolution is given by Rres = c / 2 ´ BW and it depends on the sweep bandwidth. The maximum range that could be detected depends on the sampling frequency (fs), and it is given by Rmax = fs c / 2 S . If multiple objects are moving with different relative velocities and are at the same distance from the radar, then it is difficult to estimate the range by the above said method. The reason is all these objects show only a single peak in the spectrum. Hence, in order to find the velocities of these targets, the phase of the IF signal is considered.

Deep Learning-Based Activity Monitoring for Smart Environment Using Radar

103

In order to find the relative velocity of the target objects, the phase of the IF signal (Φo) is considered. The phase difference between the transmit and receive signal is:



DF = 2p fc Dt = 2p fc

2 R 4p R = c l

(4)

Consider two consecutive chirps separated by Tc. The objects present in the consecutive chirps show change in range due to their movements. Let ΔR be the change in the range of the objects in the two consecutive chirps. Then, the corresponding phase difference in these two consecutive chirps is: DF =



4pDR l

(5)

4pn Tc (6) l lDF where ν is the velocity of the object. Therefore, n = . Hence, in multiple object 4p Tc scenario, in order to estimate their velocities, first the range FFT is computed for each chirp, and this results in a single IF for each chirp. Then, Doppler FFT is computed for all the range FFTs. This results in N frequency peaks in the resultant spectrum. N represents the number of objects. These peaks correspond to the phase difference of the individual targets. Therefore, the relative velocities of the targets are given by: DR = n Tc Þ DF =



n1 =

lDF1 lDF 2 n2 = 4p Tc 4p Tc

(7)

These phase differences in the Doppler FFT can be distinguished only if they are separated by 2p radians: N 2p DF ³ (8) N

n>

l l 2p = = n ressolution 4p NTc 2 NTc

(9)

Velocity resolution (νressolution) is the minimum velocity difference between the objects that could be detected. Similarly if ΔΦ is less than π radians, velocity can be measured without any ambiguity: DF < p



n=0 and x30 and x 60 and x90 and x120 and x250: return (((500-400) * (x-250)) / (380-250)) +400

Script for Calculating PM2.5 Sub-index Class Expression index(!SubIndex!) Code block def index(x): if x >=0 and x 50 and x 100 and x200 and x300 and x400 and x0 and x50 and x 100 and x250 and x350 and x430: return (((500-400) * (x-430)) / (510-430))

Script for Calculating PM10 Sub-index Class Expression index(!SubIndex!) Code block def index(x): if x >=0 and x 50 and x 100 and x200 and x300 and x400 and x=0 and x40 and x 80 and x380 and x800 and x1600: return (((500-400) * (x-1600)) / (3200-1600)) +400

Script for Calculating NO2 Sub-index Values Expression subindex( !Value!) Code block def subindex(x): if x>=0 and x40 and x 80 and x180 and x280 and x400: return (((500-400) * (x-400)) / (800-400)) +400

165

166

N. Munoth and N. Sharma

References 1. N. Munoth, R. Jain, G. Raheja, T. Brar, Issues of sustainable redevelopment of city core: a study of developed and developing countries. Inst. Eng. (India) 92(2), 117–122 (2013). https:// doi.org/10.1007/s40030-­013-­0045-­8 2. India: Air Quality Standards (2020 August), Retrieved from Transport Policy: https://www. transportpolicy.net/standard/india-­air-­quality-­standards/ 3. CPCB, in National Air Quality Index (PR Division on behalf of Dr. A.B. Akolkar, Member Secretary, 2020) 4. N.  Sharma, Urban Air Quality Analysis for Ahmedabad—A Spatio Temporal Assessment (Ahmedabad, Unpublished Thesis, 2020) 5. Air Pollution (2020), Retrieved from Law and Your Environment: http://www.environmentlaw. org.uk/rte.asp?id=2#:~:text=The%20World%20Health%20Organisation%20defines,with%20 the%20enjoyment%20of%20property.%22 6. J.H.  Bethel Afework, in Secondary Pollutant (2019 January), Retrieved from University of Calgary: https://energyeducation.ca/encyclopedia/Secondary_pollutant 7. C.P. Board, Air Quality Trends and Action Plan for Control of Air Pollution from Seventeen Cities (2006) 8. About SAFAR (2020 July), Retrieved from SAFAR, India: http://safar.tropmet.res.in/ ABOUT%20SAFAR-­1-­2-­Details 9. F. Ministry of Environment, in National Clean Air Action Plan (NCAP), Delhi, 2018 10. Spatial Modelling (2020 July), Retrieved from Techopedia: https://www.techopedia.com/ definition/1940/spatial-­modeling 11. ESRI, in ArcMap (2020 July), Retrieved from ArcGIS for Desktop: https://desktop.arcgis. com/en/arcmap/10.3/tools/coverage-­toolbox/project.htm 12. ArcGIS Model Builder: How to Create a Custom Toolbox and Export as a Python Script (2020 August), Retrieved from GIS Geography: https://gisgeography.com/ arcgis-­model-­builder-­custom-­toolbox-­python/ 13. Ahmedabad, Gujarat, India (2020 August), Retrieved from Latlong.net: https://www.latlong. net/place/ahmedabad-­gujarat-­india-­1187.html 14. Ahmadabad (Ahmedabad) District: Census 2011-2020 Data—Corona Virus | Covid 19 Data (2020), Retrieved from Census 2011: https://www.census2011.co.in/census/district/188-­ ahmadabad.html 15. Historic City of Ahmadabad (2020 August), Retrieved from UNESCO: https://whc.unesco. org/en/list/1551/ 16. City—Ahmedabad (Gujarat, India) (2020), Retrieved from urbanemissions.info: https:// urbanemissions.info/india-­a pna/ahmedabad-­i ndia/#:~:text=In%20Ahmedabad%2C%20 there%20is%201,PM10%2C%20SO2%2C%20and%20NO2 17. Average Humidity in Ahmedabad (Gujarat) (2020 August), Retrieved from Weather and Climate: https://weather-­and-­climate.com/average-­monthly-­Humidity-­perc,ahmedabad,India 18. Ahmedabad Climate (2020), Retrieved from CLIMATE-­DATA.ORG: https://en.climate-­data. org/asia/india/gujarat/ahmedabad-­2828/ 19. Climate Ahmedabad (2020 August), Retrieved from Meteoblue.com: https://www.meteoblue. com/en/weather/historyclimate/climatemodelled/ahmedabad_india_1279233 20. I.I. AMC, Protecting Health from Increasing Air Pollution in Ahmedabad (2017) 21. Annual Report (2020), Retrieved from GPCB: https://gpcb.gujarat.gov.in/webcontroller/page/ annual-­report 22. ArcMap (2020 July), Retrieved from ArcGIS for Desktop: https://desktop.arcgis.com/en/arcmap/10.3/manage-­data/creating-­new-­features/creating-­a-­buffer-­around-­a-­feature.htm 23. ArcMap (2020 July), Retrieved from ArcGIS for Desktop: https://desktop.arcgis.com/en/arcmap/10.3/tools/spatial-­analyst-­toolbox/understanding-­interpolation-­analysis.htm

GIS-Based Air Quality Index Spatial Model for Indian Cities

167

24. Types of Interpolation Methods (2020 July), Retrieved from GIS Resources: http://www. gisresources.com/types-­interpolation-­methods_3/#:~:text=Interpolation%20is%20the%20 process%20of,noise%20levels%2C%20and%20so%20on 25. ESRI, How Kriging Works (2020), Retrieved from ArcGIS for Desktop: https://desktop.arcgis. com/en/arcmap/10.3/tools/3d-­analyst-­toolbox/how-­kriging-­works.htm 26. S.H. Pant, What Does the Air Quality Index Really Show Us? (2020 August), Retrieved from FIT: https://fit.thequint.com/health-­news/explaining-­air-­quality-­index Dr. Navneet Munoth  is an Architect cum Urban Planner serving the role of an academician at the Maulana Azad National Institute of Technology Bhopal, an Institute of National Importance established by the Government of India. Dr. Munoth pursued his Bachelors in Architecture from the National Institute of Technology Hamirpur and the subsequent postgraduation as well as doctorate from the Indian Institute of Technology Roorkee, one of the premier institutions of the country. His doctoral research was in the domain of urban planning and management, and its topic was ‘Strategy for Sustainable Redevelopment of City Core: A Case Study of Ajmer City’. Dr. Munoth has been serving on his current position as an Assistant Professor in the Department of Architecture and Planning, MANIT Bhopal, for the past 10 years, and has been consistently improvising himself not only as a teacher but also as an interdisciplinary researcher. A lot of honours and awards which he has received as an academician both for his teaching and research pursuits stand as a proud testimony to his excellence in all his endeavours. Dr. Munoth has authored a total of 20 research papers that have been published in several reputed international and national journals and has also authored 4 chapters in different books of urban planning and intelligence. Ar. Neha Sharma  is a Geomatics Engineer as well as an Architect, presently working as a PGET at L&T-Nxt, Chennai HQ, and Geomatics Program Ambassador at CEPT University. She completed her M.  Tech in Geomatics (2018–2020) from CEPT University, Ahmedabad, as one of the best students. Her master’s thesis topic was ‘Urban Air Quality Analysis for Ahmedabad—A Spatio Temporal Assessment’. She has pursued her Bachelors in Architecture (2013–2018) from Maulana Azad National Institute of Technology Bhopal, an Institute of National Importance established by the Government of India. She has also participated as a student volunteer in CEPT in organizing various international and national training, conferences and symposiums. Her interests are remote sensing, GIS, atmospheric remote sensing, artificial intelligence, machine learning, deep learning, spatial analysis, programming, data analysis, WebGIS and 3D modelling. Neha is passionate about exploring new software, painting, sketching and crafting. She has an interest in reading about astrology and palmistry too.

Intelligent Wearable Electronics: A New Paradigm in Smart Electronics Ribu Matthew, Jyotirmoy Dutta, R. Maheswar, and Kawsar Ahmed

1  Introduction The quintessential question in the world of technology is always “What next?” Well, there could be many answers to this question, as today we strive to achieve the next technology revolution “Industry 4.0” propelled by the rise of new technologies. It could be 5G communication with lightening speeds or artificial intelligence (AI) intruding our lives like never before, integration of billions of transistors and multifunction units as system on a chip (SoC), a plethora of connected devices talking to each other with Internet of Things (IoT), maybe a foldable phone in pocket, or augmented reality (AR) revolutionizing the teaching-learning process, to mention a few. Among various technologies, one technology which perhaps best amalgamates all these emerging technologies integrating the knowledge of multi-domains on a single platform is “wearable technology.” The term wearable technology is also known as “wearable electronics (WE)” or “wearable computing,” colloquially called “wearables” or “intelligent/smart wearable electronics.” Wearable electronics—this new paradigm in smart electronics integrates areas of material science, communication, cloud computing, IoT, big data, SoC, and AI, to cite a few. Wearable technology represents all those devices which could be worn

R. Matthew (*) · R. Maheswar School of Electrical and Electronics Engineering (SEEE), VIT Bhopal University, Bhopal, India J. Dutta Centre for Digital Innovation, Christ (Deemed to be University), Bengaluru, India K. Ahmed Department of ICT, Mawlana Bhashani Science and Technology University (MBSTU), Tangail, Bangladesh e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 R. Maheswar et al. (eds.), Challenges and Solutions for Sustainable Smart City Development, EAI/Springer Innovations in Communication and Computing, https://doi.org/10.1007/978-3-030-70183-3_7

169

170

R. Matthew et al.

directly as accessories, integrated in clothing, implanted in a user’s body, loosely attached to a person, or even tattooed on skin. These devices are mostly hands-free gadgets and get their intelligence from their capability of data transfer via the communication protocols or Internet. The “loosely attached devices” to wearable technology mainly comprise of smartphones as an integral constituent [1]. Whether smartphones could be considered a wearable technology or not is debatable, but with a rapid increase in the mobile app ecosystem connected to wearables like wristbands and head gears, etc., smartphones stay as an integral part of wearable technology. The broad classification of wearable devices is devices which are worn on wrists and head mounted or strapped (to the torso, for instance) as shown in Fig. 1. Another class of wearables is those devices which can be put as garments through electronic textiles. Further, the advanced class of wearables are used as sensor patches on the body or in the form of a tattoo known as epidermal electronics which is an emerging field of wearable technology [2]. In recent years, WE technology has found wide range of applications from consumer electronics and sports to education as shown in Fig.  2. One of the major applications of WE is in the area of healthcare especially in health monitoring and prognosis, activity recognition, and assisted living. Similarly, in the area of entertainment, wearable devices assist consumers to have an immersive experience while playing virtual reality games and watching augmented reality videos and movies using smart glasses and wearable gaming devices. Wearable technology also has found applications in professional sports, especially in obtaining feedback about a player’s performance. Vital information/data is gauged from athlete’s body and utilized for performance analysis and enhancement. Wearables are also playing a critical role in providing intelligent inputs to professional coaches investigating sports kinematic and kinetics because the information obtained by wearable electronics are beyond the restricted lab environments [3]. Some of the parameters which can be monitored through hands-free wearable gadgets include indicators of performance or injury risk in real time, heart rate, global positioning data, accelerometer

Fig. 1  Classification of wearable devices (source from [2])

Intelligent Wearable Electronics: A New Paradigm in Smart Electronics

171

Fig. 2  Application areas of wearable electronics

data, and activity data, to cite a few. In workplace, wearables are being used in monitoring physiological responses of employees such as heart rate variability or electrothermal activity and for measuring the employees’ stress responses and/or fatigue [4]. Such monitoring has become essential, since according to the International Labour Organization (ILO), in every 15 s, a worker dies due to work-­ related accident or disease [5]. In this regard, wearable devices can help to improve safety and enhance productivity. The inclusion of wearables in oil and natural gas industry, mining, warehousing, and other labor-intensive industry has just started to be explored and has been found to be effective. Smart jewelry is another domain in wearable technology that has enormous potential to grow in coming years. Additionally, fashion designers have also begun to experiment smart textiles. For instance, eco-conscious textiles which monitor ambient parameters like air quality level have been reported [2]. As the number of mobile subscribers and Internet penetration increases simultaneously with decreasing data costs in developed and emerging economies, new trend in the WE domain is evolving [6]. Application domains of WE are moving beyond the regular health trackers and smart watches with more focus on wearable user interface and analytics. Emergence of WE as a service and merging of wearable cameras and virtual reality (VR) are a few among the many trends being witnessed [7]. Today, the WE market is projected to reach USD 51.60 billion by the year 2022 [8]. Thus, it is prudent to understand the applications, opportunities, constraints, and challenges of WE. This chapter provides a concise overview of the development in WE, its evolution, and present market size. Further, we elucidate the typical architecture and operation of WE.  In the subsequent section, a few examples of popular wearable devices like WE in healthcare, WE as smart textile, and WE in education are elaborated. Further, the specifics of power/energy unit of WE along with their interface with cloud computing is summarized in later sections. The chapter concludes by briefly discussing on challenges and future trends in WE technology.

172

R. Matthew et al.

2  A Brief History Although the concept of WE looks like another latest trend sweeping the digital world and pushing the limits of human imagination through multiple technologies, yet the concept itself isn’t as new as one might think. A timeline of developments in WE is depicted in Fig. 3. Treatise encompasses examples which prove that back in the year 1672, one pioneering designer in China created a functioning smart ring

Fig. 3  A brief timeline of wearables

Intelligent Wearable Electronics: A New Paradigm in Smart Electronics

173

during the era of the Qing dynasty (1644–1911). The smart ring specifics were 1.2 cm length and 0.7 cm width abacus that sits on the finger. The smart ring functioned as a counting tool and helped the traders to do quick calculations. During the 1700s, many accidents on sea prompted the British government to award £20,000 to the first man to solve the navigational problem to develop a means with which voyage to the West Indies was completed within half a degree (longitude). Later, John Harrison completed his first chronometer in the year 1735. Similarly, in 1762 Harrison’s famous No. 4 marine with chronometer depicted an error of only 5 s. The first lightweight camera was invented by Julius Neubronner in the year 1907. He used it for pigeon photography where the camera would be tied to a pigeon. This reported miniaturized camera had an integrated pneumatic timing mechanism which resulted in the activation of the shutter at set intervals [9]. Many historians consider the Sony TR-55 as a benchmark for portable/electronics gadgets we use today. A milestone in electronics industry was the invention of transistor. This breakthrough permitted devices to be much compact than earlier version with vacuum tubes resulting in radios with dimensions of a few inches. A major breakthrough in the development of compact electronics was in 1961, when Edward Thorp and Claude Shannon developed a computer that could fit into a shoe. In the year 1975, wearable technology became mainstream when the first calculator wristwatch was released—an icon of the 1970s and 1980s. The Walkman, launched in 1979 by Sony, became a rage among the youngsters and a go-to music device throughout the 1980s. In the year 1987, digital hearing aids were reported revolutionizing the healthcare industry. These could be termed as the initial steps wearable electronics was taking in what later became one of the most potent WE applications. Later, in the year 1994, Steve Mann reported wireless wearable webcam leading the way for future IoT technologies. Subsequently, the 1990s saw a number of wearable technology events which indicated a rise in popularity of WE and things to come. In the year 2000, the Bluetooth gear was invented. Ericsson T36 was the first Bluetooth mobile device developed. Updated T39 model of the device made it to stores in 2001. The operating frequency of the hands-free wireless device was between 2402 MHz and 2480 MHz or 2400 MHz and 2483.5 MHz. Later, Apple iPods hit the stores in the year 2001. The iPod is a series of compact, portable media players and multipurpose pocket computers. In 2004, Smart Personal Objects Technology (SPOT) watches were released by Microsoft. With this, Microsoft integrated Internet-connected devices to a one single gadget, but being a chargeable service the SPOT watches never took off [10]. In 2008, Fitbit was launched. Though the work on the Oculus Rift headset started in 2012, it hit the market in 2016. Another important milestone was in the year 2013 when Google Glass was launched in the market followed by Apple Watch in 2015. Like wearable gadgets, smart clothing/textile is also not faraway with smart textile designers focusing on solutions to integrate fabrics and technology indicating that soon smart clothing will be a part of the mainstream. Some solutions like the Nadi X yoga pants are already available.

174

R. Matthew et al.

3  Market Size for Wearable Electronics The market for WE has been around for around 20 years, but in recent times it has seen a sudden spurt in growth. In the year 2015, the overall market value of wearable technology was valued at USD 15.74 billion. By 2022, this is expected to reach USD 51.60 billion at a CAGR of 15.51% between the years 2016 and 2022 [8]. Many reasons could be attributed to this rise. One reason is the progress in technologies related to IoT in which not only the computers are interconnected as in traditional Internet, but also those devices with lower computing capacity are connected to the Internet along with other peripheral devices. Thus, a connected world is being envisioned which will have connectivity among various devices like automated cars, drones, sensors, and smart homes, to cite a few. For consumers, it makes their lives and jobs easier leading to an increased quality of life. Another factor that aids to the growth of WE is that people are getting more health conscious. Only till a few years back most people would pay less attention to any change in the pulse rate, sleep patterns, steps walked, calorie counts, etc. With more health apps being created, there has been a renewed interest in healthcare through WE. In the year 2016, the Global Fitness Trackers Market size was valued at $17,907 million, and by 2023 it is expected to reach around $62,128 million [11]. Another reason is industry’s renewed interest in upgrading safety parameters, especially in the oil-natural industry, manufacturing industry, and an increase in the use of technology in sports and education sectors. With data being the new gold and AI making new strides by each passing day, we are at a junction where the emerging technologies like big data, AI, IoT, etc. are overlapping each other. With data costs coming down and mobile being in reach of common people, this may be the right time for wearables to take off. Thus, WE is one platform which can integrate and harness the best from each of the aforementioned emerging technologies giving consumers an improved quality of life.

4  WE Architecture and Operation In principle, a wearable technology device, i.e., wearable electronics, interfaces with the skin or epidermis to gather critical and meaningful data. A wearable sensor is the most critical module of wearable electronics in addition to signal processing/ conditioning circuitry and communication modules. Various levels of integration of WE are shown in Fig. 4. Wearable sensors are mainly electrical, mechanical, optical, chemical, magnetic, etc. These sensors are typically realized with nano-electro-­ mechanical systems (NEMS) technology. A few reported examples include chemical/biological sensors [12–14], accelerometers [15–19], pressure sensors [20, 21], and thermal sensors [22] along with actuators [23]. These sensors convert

Intelligent Wearable Electronics: A New Paradigm in Smart Electronics

175

Fig. 4  Types and integration levels of smart wearables (source from [24])

signals measured from the body as an analyte into an equivalent electrical signal. The equivalent electrical signal may be displayed in the electronic device after signal conditioning or transmitted for further processing/interpretation. Various stages of conversion of analyte concentration into an equivalent electrical signal may include pre-concentration chambers, fluid mixers, samplers, transduction mechanism (readout), and display. The basic working principle of a typical WE is elaborated in the subsequent subsections.

4.1  Epidermis as the Information/Data Site The epidermis the largest organ of human body is a multilayered structure with its varying mechanical, optical, and electrical properties. Electrically it is resistive in nature. The epidermis is a complex structure; therefore as an information site, it is challenging to extract data from its surface or within.

176

R. Matthew et al.

4.2  Sensor Modules Typical sensor modules that are used in WE include mechanical, electrical, and optical sensors. In this section we elucidate the working principle and features of mechanical and electrical sensors in detail. 4.2.1  Mechanical Wearable Sensors Important classes of mechanical wearable sensors based on sensing technology include the following: • Piezoresistive • Capacitive • Piezoelectric 4.2.1.1  Piezoresistive Mechanical Wearable Sensors The phenomenon of change in electrical resistance of a conductor element when applied with mechanical pressure is known as piezoresistive effect. For a conductor the electric resistance (R) is a function of both intrinsic parameter and geometric constants given by Eq. (1):



R=

rl wt

(1)

where the symbol ρ represents the intrinsic property, i.e., electrical resistivity of material, whereas the notations l, w, and t are the length, width, and thickness of the conductor element, respectively. It may be noted that the conductor element can be metals, semiconductors, or doped polymers [25]. When applied with an external pressure, there is a relative change in the initial resistance of the conductor (R0) due to change in intrinsic and/or geometric parameters of the conductor represented by Eq. (2):



DR Dr * Dl = R0 Dw * Dt

(2)

In case of semiconductor piezoresistors, the change in intrinsic parameter is higher, whereas in metals the change in geometrical parameters under pressureinduced resistance change is dominant. Piezoresistive sensor-based WE have been extensively used to collect data from human body. These sensors are typically mounted on body parts with mechanical motion to measure various body parameters like bending with strain sensors [26, 27], touch using pressure sensor (multielectrode system) [28–30], etc. The active

Intelligent Wearable Electronics: A New Paradigm in Smart Electronics

177

material for piezoresistive readout-based wearable sensors is polymeric composite doped with conductive fillers such as carbon nanotube [31–35], graphene [36, 37], carbon black [38], nanoparticles (gold [39], silver [40]), and nanofibers/nanowires [41, 42]. In recent times, apart from conventional plastic/polymeric materials like PDMS, SU-8, etc., materials such as paper [43], plant, and plant-derived materials [44–46] have been explored for developing piezoresistive sensor substrate. Such sensors depict characteristics like high sensitivity, large dynamic range of measurement, compactness, high stretch ability, and low hysteresis. However, such sensors have certain limitations like thermal drift that may be nullified by carful design strategies [47–53]. 4.2.1.2  Capacitive Mechanical Wearable Sensors Capacitive sensing technique has been a popular method mainly due to high sensitivity, immunity toward electromagnetic (EM) wave radiations, low thermal drift, low power consumption, and low detection limits, to mention a few. The most widely used capacitive sensor configuration is a parallel plate configuration, in which the capacitance (C) is governed by the mathematical equation (Eq. 3):



C=

eA d

(3)

where the symbol ε represents the permittivity of dielectric medium between the parallel plates, whereas the notations A and d depict the area of a parallel plate capacitor and distance between the two plates, respectively. Upon application of an external stimulus (mechanical input), there is change in the nominal capacitance value (ΔC) given by Eq. (4):



DC e * DA = C0 Dd

(4)

where the symbol ε represents permittivity of the medium, and the variables A and d depict area of parallel plate capacitors and distance between the two capacitor plates, respectively. In addition to the aforementioned two-plate configuration, capacitive sensors have been also realized with differential mode configuration. The relative change in initial capacitance value can be due to the following two factors: (1) change in geometrical parameter mainly “d” under external mechanical stimuli and (2) change in the parameter value “ε” when the dielectric of the capacitor is modified due to the external measurand entity. This principle of sensing has been used to realize WE sensors to gather data from human body in the form of electronic skin tactile sensors [54, 55], pressure sensors [56–58], etc. Capacitive sensors in WE have been realized with polymeric substrate mainly to achieve flexibility. Pressure sensing has been accomplished with the principle used to achieve

178

R. Matthew et al.

touch sensing in typical electronics available in the market. Tremendous advancements have been made in the field of capacitive sensor-based WE, especially at the material level [59, 60]. Typical materials that have been reported to realize flexible electrodes and dielectric include silver nanowire, CNT, PDMS, and ITO thin films [56, 61, 62], to cite a few. The major challenge in design of such sensors is the fringing field effect, parasitic capacitance, poor linearity, and susceptibility to external noise components from both body and environment. An improvised capacitive sensing system depicting high sensitivity and noise immunity is supercapacitors or iontronics sensors. The electrical double layer (EDL)-based iontronics sensors are relatively a new technology. Compared to a traditional capacitor structure, for similar dimensions the EDL capacitor configuration results in higher capacitance. In addition, the EDL capacitor structure results in high noise immunity, especially in the case of wearable sensors in which the parasitic capacitance is significant. Typically, ion gels and ionic liquid-based EDL that  have been reported as body interface pressure sensors are 3D contact force measurement, as tactile sensors, to cite a few [63–65]. Literature encompasses examples of EDL-based iontronics sensors depicting ultrahigh sensitivity in the order of 3.1 nF kPa−1 enough to measure very low body interface pressure changes [66]. Even though, such sensors depict ultrahigh sensitivity, their integration into WE still has limitation primarily due to biocompatibility issues, scalability constraints, and restrictions in mass production. 4.2.1.3  Piezoelectric Mechanical Wearable Sensors Piezoelectric mechanical wearable sensors are based on the phenomenon of piezoelectric effect in which the electric charges in a material change with an external applied force/pressure. Under an external mechanical stimulus, there is reorientation of electric dipoles in a piezoelectric material resulting in the development of voltage due to surface charge reconfiguration. In recent times, various mechanical wearable sensors utilizing piezoelectric effect have been reported in the literature. A few applications include tactile sensor [67], bending sensor [68, 69], pressure sensor [70], etc. Piezoelectric mechanical sensors depict high sensitivity toward dynamic measurement with fast response times, i.e., useful in measuring high-frequency dynamic signals. However, such sensors depict poor static measurement response mainly due to charge leakage. Typically used piezoelectric materials include PVDF [71], ZnO [72], and PbTiO3 [73]. Apart from sensing applications, piezoelectric materials like PZT have been used to realize energy-harvesting devices for self-powered wearable electronics. Innovative techniques to improve the static response of piezoelectric devices such as PbTiO3 nanowire and graphene hetero-structure-based piezoelectric device with reduced carrier mobility induced due to carrier scattering phenomenon in graphene composite have been also reported [74].

Intelligent Wearable Electronics: A New Paradigm in Smart Electronics

179

4.2.2  Electrical Wearable Sensor As discussed earlier, in a human body, skin acts as the information/data extraction site. In the case of electrical wearable sensor, also human skin is the source for data extraction. An electrical sensor measures change in electrical entity (resistance/conductance or capacitance) of the skin. In the case of typical biological signals, these changes in electrical entities are weak signals demanding high complexity/specifications of electronics signal processing and conditioning circuitries. In order to extract/sense data from the skin surface, the most important parameter to be considered while designing an electrical wearable sensor is the impedance matching between the skin surface and the sensing unit, i.e., the electrode. To accomplish impedance matching, either dry or wet electrodes are used, in which the former forms a direct contact with the skin surface, whereas the latter uses gels for electrical impedance matching. Typically, dry electrode-based electric wearable sensors are preferred due to their ease of integration and usage, especially in the case of consumer electronics with repeated usage. In addition to mechanical and electrical sensors, optical sensors are also integrated in WE for sensing applications.

4.3  WE Operation Principle Smart wearable electronics design is complex and involves primarily the following modules: (1) an integrated energy source (battery), (2) sensors, (3) signal processing/conditioning modules, and (4) wireless signal transmitting/receiving module. The sensor modules convert the vital physical/chemical/biological signals from the body into an equivalent electrical signal for measurement. This electrical signal is fed to signal processing/conditioning circuits either for display or transmission. The electrical signal obtained by the sensor is typically processed by an integrated circuit (IC) technology-based circuitry. The processed signal is either displayed or transmitted using a display unit or transmitter module, respectively. For WE, flexible printed circuit board (FPCB) is an attractive alternative since it provides an ideal platform to be integrated with other modules. Literature encompasses reports of FPCB-based WE in which the signal processing, conditioning, and transceiver circuity are integrated on the flexible substrate. For instance, Gao et al. [75] reported a fully integrated wearable sensor array that encompasses signal processing, conditioning, and transmitter circuitry based on FPCB. Energy source for any portable WE systems is critical since such systems do not have the provisions to be charged continuously from an external power source. In recent years, much focus has been on developing energy source for WE system. A possible solution for energy source in WE is nano-generators based on piezoelectric and triboelectric effects. Nano-generators are energy harvesters which convert

180

R. Matthew et al.

Fig. 5  A typical computing model for healthcare system (source from [77])

thermal or mechanical energies into electric energy. Nano-energy generators with delivering power density of 35 mW/m2 and an open circuit voltage of 145 V have been reported [76]. Typical tele-healthcare and monitoring systems utilize pervasive computing as depicted in Fig. 5. Such systems include various stages like health-state data collection, data analysis/transmission, and data collection/storage/distribution followed by analysis of the received data. Such a model is not limited to healthcare systems alone but also extendable to other WE systems. Efforts are underway to integrate smart WE with IoT system. A typical architecture of smart WE interface with IoT and its various sub-systems is shown in Fig. 6. The main modules of an IoT-based WE include the following: (1) communication gateway, (2) cloud server, and (3) block chain. Premise is divided into three layers: (1) body area network (BAN), (2) local/personal area network (LAN)/(PAN), and wide area network (WAN). Communication between various modules may be wired or wireless. The sensing sub-system constitutes several sensors gauging various entities such as body parameters like temperature, moisture, body motion/movement, gesture, etc.; physiological parameters like glucose levels, blood pressure, pulse rate, etc.; and environmental parameters like light, humidity, pressure, and radiation levels, to cite a few. Further, there are actuation sub-systems that may include sound-, motion-, vibration-, and temperature-based actuators. An inherent part of such system is control sub-system that mainly encompasses system on a chip (SoC) and programmable devices like application-specific integrated circuits (ASICs)- and field programmable gate arrays (FPGAs)-based units. In addition to the aforementioned modules, there are communication, power, and storage sub-­ systems along with display units.

Intelligent Wearable Electronics: A New Paradigm in Smart Electronics

181

Fig. 6  A typical architecture of IoT-based smart garment (source from [24])

5  Popular WE In recent years, wearable electronics have found versatile applications for detecting and continuous monitoring of vital human body signals. In this section, we will discuss a few promising applications of WE.

182

R. Matthew et al.

5.1  WE in Healthcare Domain Over the years, flexible substrate-based healthcare devices have been highly explored for monitoring body signal both in the invasive and noninvasive modes. Various health body parameters that have been monitored/recorded include body temperature, glucose level, ECG, and EEG, to mention a few. Continuous monitoring of these vital signals is critical to predict any failure since feeble signatures of the abnormalities are visible at an early stage. In healthcare applications, the following segments of WE have been reported: (1) force sensors for detecting/monitoring pulse, voice, and motion and (2) biochemical sensors for measuring/analyzing glucose level and sweat, and temperature sensor to monitor/measure body temperature, body temperature profile, etc. In addition to the aforementioned classes of WE for healthcare applications, multifunctional sensor platform with multi-sensor modules and multiplexed signal detection operation have been also reported. The vital body parameters are monitored using personal area network (PAN) as depicted by Fig. 7. The data may be transferred using different communication protocols wirelessly. The communication devices are typically planted in medical body area network (MBAN) ensuring less interference from other electronic systems. 5.1.1  Sensors The various sensor modules include the following.

Bluetooth communication

3G/4G Communications

Wi fi Communications

Smartphone

Fig. 7  Wireless design in personal area network (PAN) (source from [78])

Internet

Intelligent Wearable Electronics: A New Paradigm in Smart Electronics

183

5.1.1.1  Pressure/Force Sensors Unlike the conventional designs based on metals and semiconductors, recently much focus has been on developing plastic/polymeric substrate-based force sensors due to their unprecedented advantages such as flexibility, stretchability, biocompatibility, and wearability, to mention a few. Flexible force sensors typically convert mechanical entities like pressure, tension, stress, etc. into an equivalent electrical signal. Typically such sensors complete the conversion of mechanical stimulus into an equivalent electrical signal with a readout mechanism such as piezoresistive, piezoelectric, and capacitive. Premise has been covered in the previous sections of this chapter. As mentioned such sensors have been reported with metals, semiconductors, and polymeric materials in the literature. However, in recent times, much focus has been only plastic/polymeric- or paper-based WE. A typical example of paper-based wearable sensor is reported by Tao et al. [43]. The sensor is based on graphene as active material with paper substrate for gauging pressure induced due to body movements such as pulse from the wrist, breathing rhythm, motion states, etc. The active material used was graphene oxide and paper composite realized by mixing paper with graphene oxide followed by a subsequent drying/heating step. The dynamic range of measurement of the reported sensor was observed to be in the range of 0–20 kPa. The maximum sensitivity reported was 17.2 kPa−1 in the input pressure range of 0–2 kPa. An electronic skin sensor device was reported by Wang et al. [31] with natural microcapsule actuator. In the sensor, the conductive active material was realized by preparing composite of sunflower pollen microcapsule with multiwalled CNT in PDMS. The e-skin was realized by sandwiching the active composite in-between two conductive electrodes. The sensor was used to gauge dynamic variation of pressure/strain in the human finger and throat. Carbon paper and PDMS composite active material and PDMS substrate-­ based sensors integrated with belt and gloves were reported by Li et al. [79]. The carbon paper was realized from tissue papers with a high-temperature pyrolysis process. The sensor was shown to have high sensitivity in gauging breath and gestures. Improved sensitivity capacitive e-skin sensor integrating organic field effect transistor (OFET) and micro-PDMS dielectric has been reported in treatise [80, 81]. The sensor design depicted very high sensitivity (8.4 kPa−1) in the pressure range of