Research On Multi-feature Fusion Vehicle Fatigue Detection Method

With the development of economy,the number of private cars is rising,and the number of traffic accidents per year in China is also rising.Relevant reports show that the number of traffic accidents with fatigue driving as the main reason accounts for about 20% of the total number of accidents.If the driver can be alerted to fatigue in time while driving,the number of traffic accidents caused by fatigued driving can be reduced.In the current fatigue detection algorithm,the fatigue detection method based on the characteristics of the driver’s physiological signal will interfere with the driver,making it a factor endangering driving safety.Fatigue detection based on vehicle behavior characteristics is affected by road conditions and weather,and the parameters of different vehicles are different,so its robustness is low.Fatigue detection based on driver behavior characteristics,combined with certain image processing technology,can perform fatigue detection without disturbing the driver,and is not affected by the environment and vehicle parameters,so it has the advantages of high precision and high robustness.This paper studies a method to analyze the driver’s facial behavior characteristics to judge the driver’s fatigue state.The research content mainly includes the following aspects:1.Fatigue feature area localization.Optimized multi-feature extraction process,proposed multi-task convolutional network to detect face regions from images and obtain head pose,use the ERT algorithm to obtain the face feature points to locate the eye area and the mouth area.The multi-task convolutional neural network integrates face classification,bounding box regression and head pose estimation tasks.The multi-task convolutional neural network has an accuracy rate of 99.2% for face tasks,the bounding box regression task achieves a score of 0.665 in the official WIDER FACE evaluation,and the average absolute error of the head pose estimation task is8.021.Compared with using MTCNN+FAS-Net,the frames per second detection is increased from 12 FPS to 25 FPS.2.Fatigue parameter extraction.After obtaining the feature area,use the eye aspect ratio to obtain the blink information in the eye area,and use the PERCLOS mean and the number of blinks to measure the eye fatigue state.Use the mouth aspect ratio to obtain yawn information in the mouth area,and use the number of yawns to measure mouth fatigue status.The head posture(pitch angle,yaw angle,and roll angle)was used to judge the nodding and tilting of the head,and the number of nods and the duration of abnormal postures were used to measure the fatigue state of the head.3.Multi-feature fusion fatigue detection.Because a single fatigue index has the problem of low accuracy and easy misjudgment,this paper adopts the method of multi-feature fusion to judge the fatigue state.Random forest based on CART tree is used as a multi-feature fusion method,and 5 kinds of fatigue parameters are used as features.Process the UTA data set,take 30 s short videos as samples,and select some of them as training sets and test sets.The experimental results show that the fatigue detection method adopted in this paper can achieve an accuracy of 96.2%,which has good feasibility for fatigue driving detection.Figure 36 Table 13 Reference 70...