The Research Of Driver Fatigue Detection Based On Deep Learning

The detection of driver's fatigue and to warn the driver real-time is an important guarantee for safe driving.This paper focus on four aspects: driver's face detection,face tracking,eye detection and fatigue state recognition.(1)In the face detection stage,traditional face detection algorithms can be easily affected by driver's attitude change,the illumination change during the driver's face detection.Thus,a convolutional network combined with adaptive sliding window method is proposed to extract the driver's face features and to locate the face region.(2)In the face tracking stage,the improved DSST object tracking algorithm is utilized to make the driver's face tracking.In the meanwhile,a region prediction strategy is established.Combined with the improved DSST tracking algorithm,we can track the driver's face quickly.What's more,this paper also introduces a YOLO tracking algorithm based on GPU acceleration.The simulation experiments show that the improved DSST algorithm together with the region prediction strategy and GPU acceleration based YOLO detection and tracking algorithm have achieved good tracking results.However,the YOLO tracking algorithm consumes more computing resources.Therefore,region prediction strategy with improved DSST algorithm is utilized in this paper.(3)In the stage of eye detection,a driver's eye location model combined with rough location and precise location is proposed by studying the location of human eyes.Firstly,according to the approximate extent of eye location in face image,we initially locate the location area of the eye in the face.Secondly,we build a driver's eye database,build a convolutional neural network for driver's eye recognition,and combine the sliding window method to locate the eyes accurately.(4)In the stage of fatigue recognition,the driver's eye location model,eye features extraction model based on morphology and convolution neural network is proposed.Besides that,the parameters of the height and width ratio of the driver's eye and the percentage of eye closure are used to detect the fatigue state of the driver.