Traffic Scene Perception And Prediction Based On Vehicle Trajectory Analysis

Traffic scene perception and prediction mainly include traffic flow parameter acquisition,traffic abnormal behavior recognition and traffic congestion state prediction,which is a research hot spot of scholars in the transportation field of various countries.Because the traffic scene has some features such as complex variability and the vulnerability of weather and traffic participants.Therefore,accurate perception and prediction of traffic scenarios is a huge challenge.Vehicle trajectory is visual,understandable,all-day and all-weather.The analysis of its rich temporal and spatial characteristics provides a new idea for the traffic scene perception and prediction.This paper mainly analyzes and predicts traffic scenes by extracting and analyzing vehicle trajectories video data and vehicle GPS trajectories.The main contents of the research include:(1)Aiming at the video data of traffic scene,a method based on trajectory analysis is constructed to acquire urban intersection traffic flow parameter.Firstly,in order to obtain the region of urban intersection and the movement model of vehicle,the characteristics of vehicle trajectory are analyzed by using K-Means clustering,and then the statistics of each intersection traffic flow parameters are realized.Since the above method requires to set the input parameters,this paper also studies a method of vehicle trajectory motion direction recognition based on support vector machines.In order to obtain the accurate traffic flow parameters in each directions of urban intersections,such as the left,right turn and straight travel of the vehicles.(2)A method of vehicle abnormal behavior detection based on trajectory analysis is proposed for the video data of traffic scene.Firstly,the reverse projection of image vehicle trajectory on three-dimensional space pavement is obtained by using camera calibration technology,and then the abnormal vehicle behavior detection based on individual trajectory pattern is realized.Aiming at the problem of insufficient initiative of the above method,a vehicle anomaly behavior detection method based on the trajectory vector field is proposed.The trajectory vector field is established by using the vehicle trajectory feature voting,and the abnormal behavior and motion trend detection of the vehicle are realized.(3)Aiming at the floating car GPS data of traffic scene,a method of urban road traffic congestion analysis and prediction based on vehicle GPS trajectory is proposed.Firstly,KMeans algorithm is used to cluster and analyze the trajectory sample data to obtain the actual traffic flow state.Secondly,LSTM-CNN model of traffic flow state congestion prediction based on deep learning network is constructed,and the vehicle trajectory is transformed into a series of static images to satisfy the input characteristics of convolutional neural network.Finally,the traffic congestion prediction of urban road in the future is realized by training and optimizing model.Finally,this paper validates the proposed method by using various typical traffic scenarios,and the results show that the traffic scene perception and prediction method proposed in this paper based on vehicle trajectory analysis has higher precision and better practicability.