Research On Fatigue Driving Detection Method Based On Video Analysis

With the rapid development of China’s economy and the improvement of people’s life experience,the number of cars is increasing day by day.Traffic accidents caused by fatigue driving are increasing year by year,which pose a great threat to the life safety of drivers and become the focus of attention in today’s society.Therefore,fatigue driving detection is of great significance.Through real-time detection and early warning of drivers’ fatigue driving behavior,the damage of traffic accidents can be effectively reduced.This paper makes a comparative analysis of various fatigue driving detection methods at home and abroad.A face detection and face landmark detection network based on deep learning have been proposed,which combine eye and mouth state characteristics to determine driver fatigue.The main research contents are as follows:(1)Aiming at the problem of poor generalization ability using a single neural network for face detection,a cascaded architecture with three stages of deep convolutional networks was proposed.The network can predict face in a coarse-to-fine manner.We replace the standard convolution with a combination of separable convolution and residual structure in the network to make our method achieves competitive accuracy while keeps real-time performance.(2)Aiming at the problems of unclear illumination or image quality and unbalanced distribution of training data,a new face landmark detection network was proposed.The network is composed of a backbone network and an auxiliary network.The backbone network is designed with a lightweight structure to reduce the number of parameters.The auxiliary network participates in training during the training phase to improve the accuracy of face landmark detection.In addition,the loss function was improved by assigning different weights to samples of different data volumes to balance the sample data.(3)According to the face landmark detection results,a state recognition method of the eye area and the mouth area is proposed.For eye information,the PERCLOS parameter was first used,and then the aspect ratio EAR was used to determine the fatigue blink of eye feature points and calculate the blink times.For the mouth information,the aspect ratio MAR of the mouth was proposed using the same strategy to conduct yawn judgment and yawn frequency statistics.(4)Aiming at the problem that fatigue driving detection using a single feature is easily affected by object occlusion,PERCLOS parameters,blink times and yawn times were combined to conduct fatigue driving detection.In this paper,experiments are carried out on each step of the fatigue driving detection method.The experimental results show that the proposed face detection method has higher detection accuracy while maintaining detection speed on the public data set.In the aspect of face landmark detection,the proposed method has higher detection accuracy and better robustness for complex background and illumination.Finally,the fatigue driving detection method combining the eye and mouth states can effectively detect the driver’s fatigue state.