A Driving Risk Evaluation Method Considering Driver Characteristics

With the development of economy and automobile industry,the number of vehicles has been increasing,bringing great convenience to people's life.However,traffic safety has become a big social issue.Traffic accidents have caused a lot of casualties and property losses.According to statistics,more than 90% of traffic accidents were caused by drivers.Dangerous driving behaviors became a major factor of traffic accidents.Dangerous driving existed in a variety of ways.Although some typical dangerous driving behaviors have been classified as illegal behaviors such as overspeed and drunk driving.There are still many dangerous driving behaviors that are not listed in traffic regulations.It means these behaviors cannot be effectively supervised.Therefore,it is urgent and important to evaluate the driving risk effectively,to improve the driving safety and reduce traffic accidents.However,according to investigation,most of the existing driving risk evaluation methods do not take into account the different definitions of dangerous driving behavior in different driving states and the differences between driver characteristics.In order to obtain a more accurate,more reasonable and personalized result,this paper proposes a driving risk evaluation method considering driver characteristics.The main contents are as follows:(1)Considering that the driving process of vehicles is extremely complex,in order to accurately evaluate the driving risk,it is necessary to recognize the current driving state first.Driving states were divided into three types: free driving,car following and lane changing.It's easy to distinguish free driving and car following through sensors data like radar.Thus,the lane changing state recognition model was established.The recognition model applied LSTM(Long Short Term Memory Network),and trained on real-world lane changing data which consists thirteen features of both vehicle's driving data and their neighbor's current states from NGSIM(The Next Generation Simulation)data.The accuracy of lane changing state recognition model is 0.9560.And the comparison of different samples and different algorithms shows this model has obvious advantages.(2)In order to quantify the risk degree of dangerous driving behavior,the driving risk evaluation models of each driving state were established.And considering the complexity of lane changing state,it was further subdivided into free-driving lane changing and following lane changing.The dangerous behaviors in each driving state were analyzed.The characteristic parameters of dangerous driving behaviors were selected to input the driving risk fuzzy inference system of each driving state.The driving risk was output in range [0,1].(3)Considering that applying the same algorithm to different types of drivers would leads to inaccurate results,the driving style was classified and the driving style optimization factor calculation model was established.To get the driving style,thirteen features were selected and reduced to four common factors by the factor analysis.K-means algorithm was applied to cluster the common factors and gained the centers of each diving style.The driving style optimization factor can be calculated through the distance of the common factors coordinate points.The final driving risk degree was obtained after the output of driving risk evaluation models added the driving style optimization factor.Finally,the NGSIM data were used to verify the driving risk evaluation method considering driver characteristics.The results show that this method can achieve effective and accurate driving risk evaluation for different drivers in each driving state.