Abstract: The clustering of moving object trajectories has important theoretical and practical value in the fields of urban planning,public space design and mobile object behavior prediction.In this paper,aiming at the poor clustering effect of traditional clustering algorithms (e.g.,k-means,DBSCAN) on moving object trajectories,a new trajectory clustering algorithm named iBTC was proposed.The algorithm segments the trajectory firstly,and based on the principle of minimum description length,the trajectory segmentation problem is modeled as the shortest path problem for undirected graphs.Then,Dijkstra algorithm is used to find the optimal segmentation.Next,based on the idea of Isolation Forest,the trajectory clustering problem is transformed into a special anomaly detection problem,and the algorithm uses partition-merging procedure to cluster trajectory data.Finally,experiments were performed on the simulated data set and pedestrian track data recorded by monitor.The results show that the algorithm can achieve good clustering effect.

Key words: Clustering, Isolation forest, Moving object, Trajectory segmentation