【摘要】 针对疲劳驾驶检测问题，提出了一种改进YOLOv5模型的人脸疲劳检测方法。首先，对YOLOv5模型增加检测层和添加CA注意力机制的改进，用于检测驾驶员的面部区域。其次，使用Dlib库中的级联回归算法实现人脸部68个特征点的标定和眼部、嘴部的定位。最后，计算驾驶员眼部（EAR）和嘴部（MAR）的纵横比，依据眼睑闭合程度百分比（Percentage of Eyelid Closure Over the Pupil Over Time,PERCLOS）法则进行疲劳判定并进行预警处理。实验结果表明，改进后的YOLOv5算法的平均准确率达到92.5%，能够满足人脸疲劳检测对精度和速度的综合要求。更多还原

【Abstract】 A face fatigue detection method with improved YOLOv5 model is proposed for the fatigue driving detection problem. First, the YOLOv5 model is improved by adding detection layers and adding CA attention mechanism for detecting the driver’s facial region. Second, the cascade regression algorithm in the Dlib library is used to implement the calibration of 68 feature points on the human face and the localization of the eye and mouth. Finally, by calculating the driver’s eye(EAR) and mouth(MAR) aspect ratios, fatigue determination and early warning treatment are performed based on the Percentage of Eyelid Closure Over the Pupil Over Time(PERCLOS) law. The experimental results show that the accuracy of the improved YOLOv5 algorithm is 92.5%, which can meet the combined requirements of accuracy and speed for face fatigue detection.更多还原