Research On Risk Identification Of Vehicle Lane Change Behavior

Lane changing behavior is a random movement of the vehicle,which may cause traffic clashes,leading to accidents of various degrees.There are two types of lane changing behaviors-safe lane changing behavior and risky lane changing behavior-depending on whether the movement is smooth or violent.If the lane changing behavior is too violent,risks of traffic accidents will rise.With the continuous development of automobile safety and lane-changing warning systems,it is possible to accurately identify the two types of lane changing behaviors at an early stage and make automatic warnings or operations against potential risky lane changing behaviors.In this paper,the support vector machine pattern of recognition technology was adopted to establish a model that can effectively recognize and distinguish safe lane changing behaviors and risky lane changing behaviors.With the help of KMRTDS driving simulation platform,a simulation experiment was conducted on the two types of lane changing behaviors,based on training and testing samples of recognition models extracted from driving and operating data.The significance of the study lies in the development of a system that warns the driver immediately when the lane changing behavior is too violent or has exceeded general safety limits,thus preventing accidents.The performance of classification model for recognition is determined by the size of time window for recognition,characteristic parameters and effects of model parameters.In this paper,the ROC classification results were utilized,combined with a comprehensive analysis on the recognition effects based on AUC value and the accuracy of the model.The establishment of the optimal recognition model will take three steps:1.Identify the optimal time window.2.Extract optimal feature parameters.3.Find model parameters through optimal algorithm.First of all,The start of the lane changing behavior is defined as the midpoint of the time window,and the same time interval(0.5s,1s,1.5s)was formed by forward and backward 3 time windows(1s,2s,3s),Then a comparison of model identification effect was made.The optimal time window was determined as 2s.Secondly,by using stepwise regression analysis,factor analysis,and multidimensional preference analysis,the dimension of the original feature parameters was reduced.Among them,the performance of the classifier trained by the stepwise regression analysis method is the best,so the parameters extracted by this method are the optimal feature parametersFinally,a comparison was made among the enumeration algorithm,particle swarm algorithm to find out the best approach.The classification accuracy of genetic algorithm was the lowest,while that of the enumeration algorithm and the particle swarm algorithm were quite close.In contrast,the AUC value was 0.992,close to the original data:0.996.It compensated for the loss of information in the process of dimensionality reduction,hence the selection of particle swarm algorithm was found to be the optimal algorithm.Having identified the optimal time window,the optimal parameters,and the optimal model parameters,the LIBSVM algorithm was adopted for the training and verification of the recognition model in MATLAB.The final model of the overall recognition rate was 92.55%,which can accurately identify the lane to maintain,safe lane change,the risk of changing.The results were obtained with a small volume of samples yet achieved high recognition rate model.Therefore,a good recognition effect has been realized.