Research On Driver Fatigue State Recognition Method Based On Visual Features

With the continuous increase of car ownership in our country,the number of traffic accidents is also increasing year by year,which has been widely concerned by people.The main cause of traffic accidents is that drivers have behaviors that endanger driving safety,such as fatigue driving,drunk driving and so on.Drunk driving has been effectively curbed.However,due to the dynamic nature of fatigue driving,it is difficult for traffic management departments to intervene effectively,and the accidents caused by fatigue driving has caused huge losses of life and property.In recent years,with the improvement of traffic safety regulatory requirements and the development of technology,the construction of driver fatigue recognition system to prevent fatigue driving has become one of the research hotspots.In the process of driver fatigue state recognition,the recognition accuracy of fatigue state is easily affected by the complex illumination environment and head posture changes.This paper makes an in-depth study,and proposes a driver fatigue state recognition method based on visual features,at the same time the fatigue state are divided into different fatigue grades,which improves the recognition accuracy of driver fatigue state.The research contents of this paper are as follows:Firstly,by collecting driver’s driving video,the driver’s driving state data set is made,and the driver’s face image data is preprocessed.The driver’s face detection and tracking method under complex illumination(strong light,weak light,uneven illumination)and different head posture are studied.On the basis of them,the improved residual network face detection method is used for face detection,which has high detection accuracy in complex environment.Then,fatigue features are extracted.By locating facial feature points,the eye region and mouth region are obtained,and the aspect ratio of eyes and mouth are calculated.The camera used in the experiment is calibrated to obtain the internal parameters of the camera.Combined with the plane image of face feature points,the 3D points in the camera coordinates are calculated by Pn P algorithm to estimate the head posture,and obtain the Euler angle of the head posture,then construct the 3D eyeball model.According to the Euler angle of the head posture and the error compensation of the eyeball model,the driver’s line of sight direction and gaze area are determined.Experimental results show that the algorithm can extract eye information well under complex lighting conditions,and has high detection speed and accuracy.Finally,the driver fatigue state recognition method is studied by fusing fatigue features.A fatigue state recognition method based on improved residual network and PERCLOS parameters is proposed.The fatigue state recognition method based on hog feature and SVM is studied.A fatigue state recognition and fatigue classification judgment method based on improved yolov4 algorithm is proposed.The extracted features are classified,and then the fatigue state is distinguished.In this paper,based on the experimental platform and system design,a large number of experiments are carried out to verify the proposed method of driver fatigue driving detection.The experiments show that the improved yolov4 algorithm has high robustness which can effectively prevent driver fatigue driving.