Research on Fatigue Detection Algorithm Based on Improved YOLOv8 for Face Detection

This study focuses on developing a fatigue detection algorithm based on face recognition. The algorithm first optimizes the YOLOv8 target detection framework, and significantly improves the accuracy and efficiency of face recognition by adding the CA (coordinate Attention) mechanism and replacing it with the SIoU (scaled Intersection Over Union) loss function. The algorithm accurately locates 75 facial key points, especially focusing on the feature points around the eyes and mouth. By using the changes in the aspect ratio of these key points, it can deeply analyze the eye opening and closing state, mouth activity, etc. as important indicators for evaluating fatigue. In addition, the algorithm further mines the information of head posture, and judges whether the subject has fatigue-related head movements such as nodding through key point data. Finally, by integrating multi-dimensional features such as eye, mouth state, and head posture, a comprehensive judgment of the individual fatigue state is achieved, and an efficient and accurate fatigue detection system is constructed. Experimental verification shows that after the above improvements, the mAP (mean average precision) of YOLOv8 on the face detection task is improved by 3.3%, and the accuracy of the entire fatigue detection algorithm is as high as 94.6%, which fully meets the high-precision requirements in actual application scenarios.