Vehicle Lane Change Intention Recognition And Trajectory Prediction In Highway Merge Zone

Traffic safety is a common concern in society,and lane-changing behavior is one of the major causes of traffic accidents.With the continuous development of self-driving technology,self-driving cars will be expected to reduce the number of road accidents and improve road safety in the near future.Early recognition of vehicle driving intention and accurate prediction of vehicle trajectory are the keys to realize the safe implementation of autonomous driving.Therefore,this paper focuses on lane change intention recognition and vehicle trajectory prediction.This paper takes AD4 CHE dataset as the research object,selects the vehicle trajectory data of 9 sections of highway congestion merging area for analysis,screens the trajectory anomaly through box line diagram to get 2843 natural driving trajectory data,refers to the existing related research and then combines the data and the actual situation of this paper,establishes this paper’s lane change and following fragment extraction rules,extracts a total of 2104 following fragments,414 lane change fragments,and analyzes the distribution characteristics of following and lane change duration and distance.The features that can characterize driving style are obtained from vehicle trajectory data,the features are normalized and downscaled,the DBSCAN algorithm and the K-means++clustering algorithm are compared,and the K-means++ algorithm with larger CH score and higher contour coefficient is selected as the clustering algorithm for driving style in this paper,and the driving style is divided into three categories: conservative,stable and aggressive through cluster analysis,and each trajectory is labeled with driving style to analyze the driving behavior characteristics under different driving styles.Using the three mainstream algorithms of GBDT algorithm as the base learner,the Stacking model and the Voting model are fused to determine 2.2s as the optimal time window for lane change intention,extract the vehicle interaction information and driving style information in the time window as the input features of the model,and establish two types of lane change intention recognition models considering the memory effect and the driving style.The results show that the model considering driving style has higher accuracy,better stability and less error fluctuation than the model considering only memory effect.In particular,the accuracy of the model considering driving style improves by about 3.1% between 2.2s and 3s,and the prediction accuracy of the model improves by 6.7% at 3s.A CNN-GRU model for vehicle lane change trajectory prediction was constructed based on lane change intention recognition,driving style was introduced as the feature input of the model,and the error in lateral and longitudinal coordinates was used as the evaluation index.The results show that the overall level of vehicle trajectory error in x-axis direction and y-axis direction is small,and the prediction accuracy of the model considering driving style improves27.3% in x-axis direction at 3s;26.4% in y-axis direction.19% in x-axis direction at 5s;19.1%in y-axis direction.In summary,considering driving style can effectively improve the accuracy of vehicle lane change trajectory prediction.