Progressive refinement for clustering spatio-temporal semantic trajectories

The discovery of spatio-temporal patterns in trajectories of moving objects can greatly influence many fields, such as animal migration analysis, weather forecasting, and mobile marketing. But Existing works have mainly focused on the geometric properties of trajectories, while the semantics and the background geographic information has rarely been addressed. The deep researches urgently need to deploy in order to effectively utilize the knowledge extracting for semantic trajectories. A progressive refinement clustering algorithm is proposed to extract meaningful patterns in those similar semantic trajectories. Firstly, the trajectory data are translated into stop sequences which can be handled as a traditional categorical dataset (a row-by-column Boolean table) by treating each stop as an attribute and each trajectory as a row. The generic transaction data clustering algorithm is employed to group the trajectories traversing common stops. Secondly, it clusters chronological sequence of stops-based trajectories again within the former results. Finally, the time attribute is added and obtain the refined clusters. Experiments show the proposed progressive refinement method can reduce the workload and runtime dramatically.