Maritime traffic flow clustering analysis by density based trajectory clustering with noise

Most of the existing ship trajectory clustering algorithms focus on the properties of single AIS point or sub-trajectories: the trajectory point clustering does not consider the spatio-temporal correlation between neigh-boring points on the same ship trajectory, and is incapable to portray the overall characteristics of ship motion; the ship sub-trajectories clustering needs to discard some points in the ship trajectories, which may lose the vital part of trajectories for clustering purpose. In order to solve the mentioned problems, this paper proposes a DBTCAN (Density based Trajectory Clustering of Applications with Noise) algorithm. This algorithm is suitable for clustering complete trajectories or sub-trajectories of different lengths by using Hausdorff distance as a similarity measure, and can recognize noise trajectories. In addition, DBTCAN algorithm can adaptively determine its optimal input parameters by using adaptive parameter algorithm. We test this method by real AIS data from Bohai Sea, and the experimental results show that DBTCAN algorithm can cluster ship trajectories and extract the main routes of Bohai Sea. Furthermore, the results can provide guidance for the VTS and other agents for carrying out route planning, vessel traffic separation and regulating traffic flows. •DBTCAN is suitable for clustering complete trajectories or sub-trajectories of different lengths.•DBTCAN algorithm can process massive amounts of ship trajectory data, and can eliminate the interference of cluttered trajectories.•DBTCAN algorithm can adaptively determine its optimal input parameters by using adaptive parameter algorithm.•The clustering results can provide guidance for carrying out route planning and vessel traffic separation.