Anomaly Detection Of Vehicle Trajectory Flow Based On Space Time Clustering

In recent years,mobile positioning technology and mobile communication systems have developed vigorously,and various mobile terminal devices have been widely used,and a large number of services based on the location information of mobile objects have been derived.The location information mainly includes the longitude and latitude information of the object and the timestamp.As the collection continues,a large-scale complex trajectory data set will eventually be formed.Therefore,it is necessary to conduct data mining to reveal the rules hidden behind the object behavior patterns.Commonly used data mining techniques include: cluster analysis,the purpose is to divide massive trajectory data into several clusters with similar movement trends.Trajectory anomaly detection,the purpose is to detect the trajectory flow data of moving objects with reference to the characteristic trajectory,thereby to select the abnormal behavior of the research object.This thesis is dedicated to designing a vehicle trajectory flow anomaly detection method based on trajectory space-time clustering,deeply analyzes the vehicle trajectory data,and performs efficient and accurate anomaly detection on the trajectory flow.The traditional trajectory clustering algorithm(Tra Clus)only clusters based on the spatial characteristics of the object.In order to better analyze the behavior characteristics of the moving object,the overall idea of the trajectory clustering algorithm in this paper is that firstly,the massive historical trajectories of objects are simply cleaned,after the preprocessing is completed,the efficient K-MEANS clustering algorithm is used for clustering according to the time label of the object,and then the trajectory clustering of the spatial dimension is performed in different time clusters,and the trajectory division strategy with angle limitation is used to reduce the algorithm LDBSCAN algorithm,which combines the local outlier factor(LOF)and the density-based spatial clustering algorithm(DBSCAN)to eliminate the influence of outlier sample points,uses the clustering algorithm with outlier removal and LDBSCAN algorithm to improve the accuracy of clustering results.In order to realize the anomaly detection of vehicle trajectory flow,this thesis develops an application equipped with trajectory flow anomaly detection algorithm on the vehicle edge device,and realizes the edge calculation of the data analysis result.This method efficiently utilizes system-on-chip(So C)performance,completes computing tasks on edge devices and uploads them to the cloud server,reducing the loss of data transmission and the computing pressure of the server.This thesis uses the trajectory clustering algorithm to obtain the characteristic trajectory of the object movement,and uses the sliding window as a reference to detect the trajectory flow anomaly of the moving target.Finally,a trajectory flow anomaly detection model combining spatial and temporal characteristics is proposed to realize the detection of moving objects.Effective detection of abnormal trajectory flow.Experimental results show that the proposed clustering algorithm effectively reduces the time of data processing,improves the quality of clustering and makes the clustering result smoother.The anomaly detection model combined with the Internet of Vehicles platform can effectively detect the anomaly of the vehicle trajectory flow based on the results of the trajectory clustering analysis,and update the anomaly detection model through the newly collected trajectory to improve the reliability of the model.