Research And Application Of Driving Fatigue Detection Method Based On Multi-source Behavior Fusion

Under the rapid economic development,the number of motor vehicles on the road has increased dramatically,and accidents caused by fatigue driving often occur.The detection method based on EEG and ECG needs to be worn in a special state,which is not conducive to the user experience.Popularization,the detection method for the eyes or for yawning is not only complicated and too simple,and it cannot comprehensively detect the fatigue state of the driver.In response to this problem,this paper is devoted to applying the face recognition algorithm based on convolutional neural network to the research of this paper,and then to study the positioning method and recognition method of the eyes and mouth,and to study how to integrate multi-source behavior more comprehensive to detect the fatigue state of the driver,and finally put forward a set of application algorithms applied in the actual fatigue driving detection.The specific work has the following aspects:(1)Firstly,the research status of related work is investigated,and the basic research work is summarized,including the composition of artificial neural networks,training methods,CNN network structure and its application in face detection,and image denoising relevant algorithms.This all make basic preparations for subsequent research;(2)Secondly,for the positioning of the driver’s mouth and eyes,a method based on facial geometric distribution was used to roughly locate the mouth,and a regression tree model based on statistical classification was used to accurately locate the eyes.In terms of mouth state recognition,in order to improve the accuracy of the recognition of the driver ’s mouth opening action,a qualitative recognition of the opening and closing of the mouth based on the aspect ratio index after the binarized mouth was proposed.For different people’s mouths,the calibration method was used to find the discrimination threshold,and the experiment compared the method based on the area of the mouth was designed to verify its effectiveness and accuracy.In the recognition of the eye state,on the basis of detecting the six feature points of the human eye,a recognition method based on the opening angle of the eye was proposed to quantitatively reflect the degree of eye opening.For the case of different eye sizes,design a calibration method to find the recognition threshold for a specific human eye,and identify the blinking action or the long-term closing action according to the time of closing the eyes.The experimental results verify the effectiveness and accuracy of the proposed method;(3)Finally,the significance and definition method of three multi-source fatigue driving indicators are analyzed and studied,which are P80 indicator of PERCLOS,blink frequency indicator(times / minute)and yawn frequency indicator(times / two minutes).According to the definition of the P80 index,a method of mapping from the proposed eye state recognition method to the P80 index was designed,and the normalization method of blink frequency and yawn frequency was designed according to a priori knowledge.Combining the advantages of artificial neural network,a multisource facial behavior fatigue driving detection model was proposed,and simulation experimental data was designed to verify the validity and accuracy of the model.Finally,a set of self-learning application algorithms were designed for system development and use.