Research On Anomaly Recognition Based On Vehicle Track Data At Intersection

Intersection brings together different directions of traffic flow,traffic environment is complex,so it has a significant impact on the road network traffic safety and operation efficiency.With the development of big data and information technology,in the complex intersection environment,it is an effective way to improve the traffic at future intersections to dig and analyze the vehicle track data deeply and make timely warning for the abnormal behavior of vehicles.At present,the commonly used trajectory data mining techniques can be divided into two categories,including cluster analysis and anomaly detection.Trajectory clustering is to cluster massive trajectory data into several relatively homogeneous groups according to a certain law.The anomaly detection is based on the characteristic track as reference to detect individual tracks that do not conform to the normal track mode,so as to judge whether the vehicle track is abnormal.The purpose of this paper is to use the different characteristics of intersection trajectory to cluster the trajectory data,and propose a real-time and efficient trajectory anomaly recognition method,so as to provide new ideas for improving traffic order at intersections and solving driving safety problems,and contribute to the research and development of intelligent transportation..In terms of clustering analysis,this paper analyzes the characteristics of trajectory velocity,direction,acceleration,and so on,and selects significant features such as turning angle as the characteristic indexes of clustering,which makes up for the deficiency of traditional trajectory clustering algorithm which only cluster the spatial features of trajectory data.In order to study the behavior characteristics of running vehicle trajectories more accurately,this paper improved the aggregation hierarchical clustering algorithm and proposed a hierarchical clustering algorithm based on trajectory multi-features,which carried out trajectory clustering from the perspective of trajectory direction and spatial position.In this method,Bhattacharyya distance is used for the first layer clustering of trajectory turning angle,and the preliminary clustering results are further clustered based on the starting position.Due to the high dimension of trajectory data,Laplacian mapping is added in this paper to reduce data dimensions and improve computational efficiency.Compared with the traditional trajectory clustering algorithm,the efficiency of the improved clustering algorithm is verified.In the aspect of anomaly detection,the current trajectory anomaly detection methods usually only focus on the spatial anomaly of the trajectory,and rarely consider the direction and angle anomaly of the trajectory.This paper identifies track anomalies from two aspects of starting point position and steering angle,and builds an anomaly recognition model based on track data of running vehicles.This model can not only improve the accuracy of recognition results effectively,but also identify the track subsegment anomalies with insignificant differences.The specific content is to establish the location recognition algorithm of the trajectory starting point through the GMM model,and then take the moving window as the basic comparison unit,establish the recognition algorithm based on the trajectory turning Angle,so as to sum up a real-time trajectory anomaly recognition model.Finally,the anomaly recognition algorithm proposed in this paper is verified by using vehicle track data in real intersection scenarios.The results show that the anomaly recognition algorithm proposed in this paper can identify the vehicle operating mode at intersections in real time and efficiently and find out the anomaly points,and has a higher accuracy than the traditional method.