Vision-Based End-to-End Deep Learning for Autonomous Driving in Next-Generation IoT Systems

Abstract

AbstractThe Internet of Things (IoT) systems have grown and become an essential part of everyday lives. Since most computations are performed in the cloud, traditional IoT systems are not designed to meet the requirements of high reliability and low latency in autonomous driving systems. With the help of deep learning, IoT network is evolving into the Internet of Autonomous Things (IoAT). IoAT as the next-generation IoT network is expected to process more information locally and transform computers into autonomous things, not only freely roaming in and interacting with the physical environment without human guidance but also making proactive decisions based on the IoAT network information. This chapter presents an overview of research on using end-to-end deep learning technologies for computer vision-based autonomous driving systems. It briefly discusses the ethics of autonomous driving; it also describes autonomous driving paradigms and the associated deep learning methodologies. Furthermore, it proposes an IoAT-compatible low-cost, low-latency, high-accuracy, and high-reliability CNN-LSTM-based autonomous driving model that incorporates temporal information, transfer learning, and navigational command. It also provides a detailed analysis against existing models. Finally, the chapter draws its conclusions and discusses future research directions to further improve system performance.KeywordsComputer visionDeep learningAutonomous drivingConvolutional neural networks (CNN)Internet of Autonomous Things (IoAT)Long short-term memory (LSTM)