Integration of Yolov5 for Enhanced Driver Safety through Facial Recognition and Drowsiness Monitoring

Driver fatigue is a primary contributor to traffic accidents in various countries. To address this issue, the research developed a computer vision system using facial feature recognition and blink rate monitoring to promptly detect signs of driver fatigue. While facial recognition technology has proven effective in fatigue detection and mask compliance, it faces challenges in adapting to diverse environmental conditions and ensuring efficient device utilization. The study integrated YOLOv5 into machine learning to enhance fatigue detection using a Raspberry Pi 4, demonstrating significant improvements in driving safety through real-time object identification. The research documented significant outcomes from training the YOLOv5 model, with consistent reductions in bounding box prediction losses, object detection, and classification throughout the training process. Precision and recall metrics showed steady increases, while Mean Average Precision (mAP) scores indicated progressive gains, confirming the model's ability to accurately predict labels with minimal bounding box overlap. Precision-recall and recall-recall curves for "focused" and "microsleep" drivers illustrated how the YOLOv5 model adjusted its performance based on varying confidence levels in object recognition. In summary, this study underscores the effectiveness of YOLOv5 in enhancing driving safety by efficiently detecting and identifying potential signs of driver fatigue in real-time scenarios.