Research On Drowsiness Detection Based On Steering Wheel Operating Characteristics Under Real Road Conditions

Drowsiness driving is a key incentive for traffic accidents,especially under highway conditions,where the monotonous driving environment intensifies the driver drowsiness and ultimately causes the tragedy of car crash.Therefore,developing driver's drowsiness test technologies for real vehicle applications is of great significance for safe driving.Due to the risk of real car experiments,the previous research is mostly based on driving simulator,for which our research group has previously developed experimental experiences and data analysis methods.This paper conducts the real vehicle drowsiness experiments,obtains the test data,grasps the drowsiness driving phenomenon under real vehicle test condition,analyzes the operation characteristics of the drowsiness driving,extracts the drowsiness index in both time and frequency domain,select optimal feature set based on the candidate index set,designs SVM classifier,and develops driver drowsiness detection algorithm for real vehicle application.Firstly,the real vehicle drowsiness experiments in the highway carried based on a Golf7 is carried,and the steering wheel angle,speed,driver facial video and other experimental data are collected.A sample set of driver status data is constructed by scoring the sample data using expert subjective scoring method based on face video.The existing research usually identify the drowsiness based on the time-domain characteristics of driving behavior variables,lacking the frequency domain characteristics of operational behavior parameters.In this paper,the drowsiness driving characteristics of real vehicle experiment are analyzed from both time and frequency domain.Based on the analysis of drowsiness driving characteristics,77 quantitative drowsiness indexes are extracted from the time and frequency domain respectively for the highway and simulator experimental data.To avoid the negative influence on the classifier caused by low correlation between the selected index and the fatigue or the interdependence of the features,the sequential forward floating selection algorithm(SFFS)is adopted to optimize the index set,and the accuracy of 10 rounds 10-folds cross validation is obtained under different index screening conditions based on the cross validation of Fisher and SVM classifiers.The results show that the SVM classifier's detection accuracy is less dependent on the sample,and the accuracy is relatively high.In order to furtherly improve the accuracy of drowsiness detection,the joint optimization method is used to optimize the parameters of the classifier and index set,and the optimal index subset and the highest accuracy are obtained under different SVM parameters.For the highway data set,the optimal penalty weight C = 8 and RBF(radial basis function)parameter ?= 4 are obtained after optimizing the SVM parameters and the indicator set.The maximum accuracy of the drowsiness detection is 89.27%.Finally,to verify the effectiveness of the above algorithm,the drowsiness driving test is carried out based on the Golf test vehicle on the expressway.After drowsiness scoring of the sample data,index extraction and optimization and parameter optimization,the accuracy of cross validation reaches 81.82%,which shows that the presented algorithm can effectively detect the drowsiness.