Real-Time Object Recognition For Advanced Driver-Assistance Systems (ADAS) Using Deep Learning On Edge Devices

Abstract

Self-driving cars utilize sensors and artificial intelligence to navigate to destinations autonomously, thus enhancing safety. As autonomous vehicles advance swiftly, accurately detecting objects in real-time is essential to avoid collisions. Advanced driver assistance systems boost vehicle safety and efficiency by providing real-time warnings. In addition, autonomous vehicles improve the decision-making processes to reach the destination. This proposed work detects real-time objects such as cars, bikes, trucks, buses, lorries, autos, barrier cones, and pedestrians using a deep learning model implemented on an AI board. The performance metrics of the model are evaluated by calculating the mean average precision (mAP), recall, and precision. The results show a mean average precision of 91.9%, with precision and recall values of 98.6% and 96% respectively, compared to the different versions of Yolo models.