Edge-based Algorithm Innovations for Intelligent Transportation Systems - A Safety and Efficiency Perspective

Abstract

In this thesis, we tackle unique and compelling challenges within Intelligent Transportation Systems (ITS) by leveraging low power edge computing devices. alongside AI and machine learning innovations. This work delves into two important problems in ITS pertaining to efficiency of traffic flow prediction and safety in two wheeler driving. For traffic management, we propose LSTM-based autoencoders, equipped with contextual attention mechanisms, to precisely identify and respond to anomalous traffic patterns. This approach, in analyzing vast datasets like the PeMS, not only enhances the accuracy of anomaly detection but also the efficiency of traffic flow management in urban settings. Turning our focus to two-wheeler safety, we employ simple sensor technologies to develop models that excel in the real-time classification of driving events and fall detection. By meticulously testing various machine learning models, we’ve proposed time-series-based LSTM and Bi-LSTM networks for their superior accuracy in recognizing critical safety incidents. The practical deployment of these models on edge devices, such as Raspberry Pi, underscores their viability for instant safety interventions, a crucial step towards mitigating accidents before they occur. Moreover, we developed a predictive model utilizing the Isolation Forest algorithm to anticipate fall events based on rider behavior, an innovation aiming at preemptive safety measures rather than reactive responses. This predictive capability represents a paradigm shift in how vehicular safety technologies are conceptualized and deployed, focusing on accident prevention. Our comprehensive study illustrates the transformative potential of integrating edge computing with AI in ITS. By addressing the unique challenges of anomaly detection in both traffic management and two-wheeler safety, we contribute significantly to the advancement of intelligent transportation systems. This research not only paves the way for future innovations in vehicular safety and traffic optimization but also promises to enhance the efficiency and safety of transportation globally.