Study On Vehicle Lane Changing Behavior Recognition And Lateral Trajectory Prediction Based On Support Vector Machine

Vehicle lane changing is a common driving behavior,which greatly affects the safety of vehicle operation.Traffic congestion and traffic accidents caused by lane changing are very common,especially traffic accidents caused by lane changing on highways,which seriously endanger people’s lives and property safety.This also makes the model construction and analysis of vehicle lane changing behavior become one of the research hotspots.Therefore,in order to accurately and efficiently identify lane changing condition,it is of great significance to establish a recognition model of lane changing behavior.Based on the analysis of the above problems,from the point of view of data-driven,this paper fully extracts the main feature information of vehicle lane changing trajectory data with the help of the powerful non-linear learning and high-level pattern recognition ability of support vector machine.The identification and modeling of the actual lane changing process and the prediction of vehicle lateral movement trajectory are carried out.The main contents of the article include:(1)Analyze the behavior of vehicle lane changing.From the perspective of vehicle lane changing trajectory,the lane changing process of the vehicle is divided into the vehicle following stage,the vehicle lane changing preparation stage,and the vehicle lane changing execution stage.Determine the 10 influencing factors that can characterize the lane changing behavior,and extract the data of the lane changing related variables based on the NGSIM(Next Generation Simulation)trajectory data set,including the trajectory data of the lane changing vehicle and its surrounding vehicles.The original trajectory data of the extracted from lane changing vehicle is normalized and the principal component analysis dimension reduction processing is used as the basic experimental data for constructing the vehicle lane changing recognition model.(2)Establish a multi-classification support vector machine vehicle lane recognition model based on grid search combined with particle swarm optimization(Grid Search-PSO).Based on the training sample data set,the rough search method is used to locate the approximate interval ranges of the corresponding parameters C and g,which are used as the benchmark values of the optimization range of the parameters of particle swarm optimization.Then,the optimal combination of training parameters is determined by using particle swarm optimization algorithm in this range.The model is established in Matlab environment,and the test accuracy of the model is 97.68%,and the corresponding recognition accuracy of SVM model without optimization parameters is only 80.87%.The results show that the model has strong classification ability and robustness.(3)Through the analysis of vehicle lane changing time and lane changing lateral trajectory,the polynomial model is used to fit the lane changing trajectory,the K-fold cross-validation method is used to test the fitting polynomial model,and to realize the prediction the time of vehicle lane changing and the lateral movement distance of the lane changing vehicle,the prediction model with support vector regression mechanism is built.