Research on Driver Fatigue Early Warning System Based on YOLO Algorithm

Driving fatigue is one of the key factors that cause major traffic accidents. In this paper, a real-time driver fatigue early warning system based on YOLO (You Only Look Once) algorithm is proposed and investigated to address the limitations of traditional fatigue detection methods. The core of the system adopts the improved YOLOv7 model to efficiently detect the key facial regions, combines with the dlib library to accurately locate 68 facial feature points, and extracts the eye aspect ratio and mouth aspect ratio as the key fatigue indicators. By introducing the Perclos model, the frequency of eye closure and yawning is analysed in 150 frames to quantify the fatigue state. Experiments on the NTHU-DDD dataset show that the system has an accuracy of 89.2% for distracting behaviour recognition and 91.5% for fatigue state determination, with good real-time performance. Meanwhile, this paper provides an in-depth analysis of the challenges faced by the system in complex scenarios such as extreme lighting and occlusion, and looks forward to future research directions such as lightweight deployment, multimodal fusion and cross-scene generalisation.