ATDP: An Adaptive Spatio-Temporal Trajectory Clustering Algorithm Based on Density Peak

Abstract

With the development of mobile positioning technology, a large number of trajectory data with spatio-temporal information are rapidly produced and collected. Density peak clustering (DP) is a simple and effective trajectory data clustering algorithm, but the algorithm requires manual intervention in clustering process and is sensitive to parameters. Based on this, the paper proposes an adaptive spatio-temporal trajectory clustering algorithm (ATDP) based on DP. ATDP proposed an improved Hausdorff distance measurement algorithm combined with the time dimension, which can generate sub-trajectories with time synchronization for similarity measurement, reducing the algorithm complexity and reflecting the time synchronization of trajectory data. At the same time, ATDP introduces KNN idea to redefine the local density and enhance the robustness of the algorithm. In view of the selection problem of clustering center in DP process, a stable adaptive selection method was proposed to avoid manual intervention in the clustering process and realize the visualization of clustering results. Finally, ATDP algorithm is tested on several real trajectory datasets. Compared with the classical trajectory clustering algorithm, the experimental results show that the ATDP algorithm can realize the adaptive trajectory clustering based on the density peak, and the clustering results can accurately reflect the common motion trend of moving targets and have better clustering quality.