Research On Driver Fatigue Detection Technology Based On Multi-feature Fusion

With the acceleration of urbanization process,the demand and ownership of cars in China have increased sharply,and road traffic safety has become an urgent social problem in China.Research shows that fatigue driving is one of the main causes of traffic accidents,this paper studies the driver fatigue detection technology based on multi feature fusion method.The main work is as follows:Firstly,according to the anchor box structure,the K-mean clustering method is used to modify the yolov3 algorithm,which increases the average accuracy(AP)of face detection by 16.7%.Then,the modified Yolov3 algorithm is compared with the Adaboost algorithm based on Haar-like.The comparison results show that the former is significantly better than the latter in face detection accuracy and detection speed.At the same time,in order to solve the problem of missing detection in the detection process of face detection algorithm,Kalman filter algorithm is applied to the tracking of face motion trajectory to predict the position of face detected in the next frame image.Secondly,a cascade lifting tree based method is used to detect 68 key points of human face.Then the eye aspect ratio(EAR)is calculated by using the method based on the combination of key points and image processing.Compared with using only key points to calculate EAR,this method can effectively solve the problem of excessive EAR error caused by dark light or dark eye color,and increase the EAR difference when the eye is closed.In addition,this method increases the accuracy of eye opening discrimination from 86.8%to 92.4%.The key point based method is used to calculate the mouth aspect ratio(MRA),and three fatigue characteristics are calculated according to EAR and MAR:percentage of eyelid closure over the pupil time(PERCLOS),the longest continuous eye closure time and the number of mouth openings.Then,taking the three fatigue characteristics PERCLOS,the longest continuous eye closure time and the number of mouth opening as the input and the fatigue state as the output,a three-layer BP neural network has been constructed.The number of hidden layer nodes is determined to be 4 through simulation.The experimental results show that the accuracy of this method is up to 90%and can complete the detection task.Finally,in order to solve the problem of fatigue misjudgment caused by the decrease of eye EAR value due to the change of facial expression or other factors in awake state,this paper proposes a fatigue detection method based on time accumulation,and constructs a fatigue detection model.The specific method is to divide the sample to be tested into multiple segments,judge its fatigue state by BP neural network,calculate the fatigue cumulative value according to the fatigue cumulative model,and judge the fatigue state of the sample according to the cumulative value.Through the design of long and short video detection experiments,it is verified that the improved method improves the accuracy of fatigue detection to a certain extent,which has research reference significance.