An efficient method for driver fatigue state detection based on deep learning

Abstract

The driver fatigue detection method based on traditional image processing technology suffers from low accuracy due to object rotation, illumination change, and object occlusion. This paper proposed an efficient method for detecting driver fatigue state based on face infrared image and deep learning. The method performed face detection and feature point locating through a cascaded network with depthwise separable convolution as the core, then acquired the eye region according to the feature point coordinates and then performed eye state recognition. Finally, the PERCLOS criterion was used for fatigue judgment, and then the result was output. The experimental results showed that the accuracy of the proposed method is 96.2%, and the average time is 32.4 milliseconds, which is better than the existing methods and can effectively detect the fatigue state of the driver. This method is important for the protection of drivers and traffic safety.