Research On The Cause Analysis And Identification Method Of Road Risk Based On Near-Crash Events

Traffic accidents bring irreparable casualties and property losses to the world every year.How to reduce traffic accidents and improve traffic safety is one of the primary concerns of various countries for traffic.The most important type of traffic accident is the driving collision accident.The traditional research on the cause of accident risk and risk prediction is mostly based on such data for analysis and modeling.However,due to the fact that the driving accident data is not easy to obtain,the overall number is small,and the detailed driving data is lacking in the accident data,the characteristic variables used in the modeling of the models proposed by these studies are often limited to conventional traffic flow variables,which cannot be make full use of a series of data during vehicle operation.Near-crash event refers to the event that requires the target vehicle or any other vehicle,pedestrian,bicycle or animal to quickly avoid,can provide more controllable laboratory data as an effective supplement to traffic safety research,Furthermore,near-crash event analysis has the potential to further understand crash causality and improve road safety.Therefore,this paper aims to propose a road risk causative analysis model and road segment risk identification model based on Near-crash events for different road sections.Research in the field brings new ideas and modeling methods for the use of data frames.First,analyzing the data from which Near-crash events can be obtained.In order to extract Near-crash events from vehicle operation data,a near-crash method based on braking deceleration and Time to Collision(TTC)is proposed;and using DBSCAN clustering algorithm to divide Near-crash events into high,medium and low severity levels;using a real vehicle experimental platform equipped with Advanced Driver Assistant System(ADAS)to collect vehicle operation data,88 real vehicle experiments including highways,urban roads,and urban expressways were carried out;the collected vehicle operation data were preprocessed to form a usable operation data set.Then,the collision accident was replaced by the Near-crash event,and a comprehensive database was established to explore the influencing factors related to the Near-crash event through literature research and related research experience,and the causative factor information was coded through video and questionnaires;After coding the database,the common factors and other information of different levels of Near-crash events were extracted through validity and reliability tests;the common factors and other information were used as empirical factor models,and AMOS software was used to establish a structural equation model to study the influencing factors of different levels of Near-crash event;the final result proves the feasibility of the structural equation model,and the causative factors of the Near-crash events of different levels are different.In order to match the vehicle operation data with the road sections,a method of dividing road sections based on three different road types is proposed.The sampling of expressways and urban expressways is equal to the interval rules,and the adjacent intersections of urban roads are sampled as one.According to the principle of road sections,the near-crash events of different levels are matched to the road sections,and a method of dividing the risk of road sections based on the number of near-crash events of different levels on the road section is proposed;the operation data of all vehicles are matched to the road sections,and the speed,lateral acceleration,longitudinal acceleration,yaw angle,accelerator,steering wheel angle,steering wheel angle,a total of seven indicators to build a road section risk identification database.Finally,based on Netica software,a Bayesian network model was constructed for three different road types,namely urban roads,urban expressways and expressways,and the sensitivity and effectiveness of the models were analyzed;the final model accuracy was 78.120%;The number of wrong road sections in the evaluation of urban road sections is 30,and the accuracy rate is 75.806%.Among them,the number of wrong road sections in the evaluation of expressway sections is 64,and the accuracy rate is 76.727%.Among them,the number of wrong road sections in the evaluation of expressway sections is 41 the accuracy rate is 81.192%.The model has the worst ability to identify risks in urban road sections,which may be due to the complex traffic conditions in urban road sections,and it can be seen from the video that Near-crash events mostly occur when vehicles stop or start at traffic lights.