A novel real-time framework for extracting patterns from trajectory data streams

The rapid development and deployment of location-acquisition equipment such as GPS systems and GSM communication networks has made collection of spatio-temporal trajectory datasets possible and led to the demand of managing and mining patterns from trajectory datasets to discover objects' movement behavior. As trajectories are generated continuously without limitation and boundaries, they form stream data. Though there are lots of research work done on mining trajectory datasets, none of them considers trajectory data as streams. They treat trajectory data as static data and run multiple scans on the data. In this paper, we present our efforts in facilitating this demand by developing a novel stream data mining algorithm to discover spatio-temporal sequential patterns from trajectories in real time; our algorithm is the first on-line trajectory mining algorithm and only needs to scan the trajectory dataset one time. We also propose a new data structure, called trajectory stream mining tree (TSM-tree), to store and represent up-to-date trajectory patterns. We conduct experiments using real life trajectory datasets to evaluate the performance of our algorithm.