Abstract: Aimed at the problem of fatigue driving detection, a method of eye fatigue detection based on facial image features is proposed. We used the RetinaFace network to detect the facial area. The Ensemble of Regression Trees (ERT) algorithm was used to obtain 68 key feature points, and the eye regions were divided. We calculated the eye aspect ratio to detect the blink behavior. According to the PERCLOS criterion, the detection and determination of the fatigue state was realized. The experimental results on the YawDD dataset show that the average accuracy, precision, and recall rate of this method are 90.24%, 92.41% and 91.90%, which can effectively identify eye fatigue.