YOLOv5-Based Driver Behavior Monitoring System for Safer Roads on Jetson Xavier NX

Recent advancements in computer vision and deep learning have led to the widespread adoption of driver assistance systems (ADAS), which play a crucial role in detecting critical situations to ensure driving safety and comfort. However, achieving real-time monitoring of both the driver and the environment remains a significant challenge. This study addresses this gap by developing a real-time ADAS utilizing images from embedded platform cameras. The system employs a driver-oriented approach, analyzing driver conditions, including phone and cigarette use, as well as eye tracking, to detect fatigue and sleep, thus providing timely warnings. Models were trained on GPU using custom datasets, and detection speeds were compared across different embedded platforms and a computer environment. The study culminates in the development of a real-time ADAS prototype boasting a remarkable 95% accuracy rate.