Drunk Driving Behavior Detection Based On The Deep Learning And Wavelet Analysis

In recent years,with the rapid increase in the number of motor vehicles,it has brought extremely serious road safety problems.Among the fatal traffic accident factors,95% are caused by human factors,among which: drunk driving is the most serious factor,and the traffic accident rate caused by it shows a trend of increasing year by year.Based on the above background,this paper proposes a deep extreme learning machine model based on kernel function for accurate detection of drunk driving behavior.The model detects drunk driving behavior by detecting typical abnormal driving behavior cues related to drunk driving,including: weaving,drifting,swerving,turn with a wide radius,accelerating rapidly,decelerating rapidly,and speeding behavior.The problem of drunk driving behavior detection is converted into the detection of seven kinds of abnormal driving behavior cues.Among them,as the detected drunk driving cues increase,the possibility of drunk driving increases correspondingly.This paper uses a car simulator to collect the driving data of the driver,including speed,acceleration,steering wheel angle and distance from the center of the lane,to provide data support for model training and verification.Wavelet transform is used to pre-process the collected data;feature extraction and comparison of vehicle behavior data using wavelet transform and wavelet packet transform,respectively;then by b uilding a support vector machine model based on kernel function,a long-short-term memory network model,deep extreme learning machine model based on kernel function model detect drunk driving behavior.Finally,the comparison shows that the proposed deep extreme learning machine model based on kernel functions in this paper is more accurate and sharp,achieving the detection rate of 92.4%.