Research On Evaluation Methods Of Driving Risk In Different Driver Fatigue States

The Driver is the leading factor in road traffic safety,and 90% of traffic accidents are related to human factors.Traffic accidents caused by driver’s fatigue have especially caused serious loss of life and property to the society.Researchers have carried out extended studies on driving fatigue and driving risks.It is proved that driver’s driving performance will decline under fatigue.However,there is no quantitative assessment of the driving risk caused by driving fatigue.In addition,previous studies have ignored driver’s status and surrounding environment to the driving risks.Therefore,this paper designs an experiment to evaluate driving risk levels under different driving fatigue states,building typical driving scenarios based on a driving simulator and establishing a quantitative index system for driving risks in different driving scenarios,which provides a way for classifying driving risk levels.Thirty-one subjects participated in the experiment,and their driving fatigue subjective data based on the Stanford Sleepiness Scale(SSS)and driving behavior data were collected.The significant difference of driving risk indicators under different fatigue levels were analyzed.The results show that with the deepening of driving fatigue,the driver’s multiple driving behavior indicators show an upward trend,and there are significant differences under different driving fatigue levels,indicating that driving fatigue has an impact on the driver’s longitudinal and lateral motion control ability of vehicles.In order to evaluate the driving risk levels,the multi-dimensional driving risk index characteristic parameters are reduced to obtain driving risk factors based on the Factor Analysis.Then,based on the driving risk factors,the K-means Clustering is used to divide the driving behavior into three types: low-risk,medium-risk,and high-risk,which are named according to the final cluster center.The driving fatigue level and the driving risk level are mapped.The results show that as fatigue deepens,the probability of low-risk decreases,and the probability of medium-risk and high-risk increase,and there is a strong correlation between driving fatigue and driving risk levels.Finally,a deep neural network(DNN)is used to train and test the driving behavior data with driving risk level labels after K-means clustering.The results show that the recognition accuracy of the DNN model on the test set in the straight scenario is 95.93%,92.31% in the detouring scenario,and 96.49% in the pedestrian scenario.A verification experiment was designed to collect driving behavior data of two subjects under safe and dangerous driving operations whose driving risk levels were manually assessed and identified through the DNN model.The results show that the driving risk levels identified by the DNN model has a high consistency with that evaluated manually,and it can even more accurately distinguish the medium-risk and high-risk.