Mining Trajectory Corridors Using Fréchet Distance and Meshing Grids

With technology advancement and increasing popularity of location-aware devices, trajectory data are ubiquitous in the real world. Trajectory corridor, as one of the moving patterns, is composed of concatenated subtrajectory clusters which help analyze the behaviors of moving objects. In this paper we adopt a three-phase approach to discover trajectory corridors using Fréchet distance as a dissimilarity measurement. First, trajectories are segmented into sub-trajectories using meshing-grids. In the second phase, a hierarchical method is utilized to cluster intra-grid sub-trajectories for each grid cell. Finally, local clusters in each single grid cell are concatenated to construct trajectory corridors. By utilizing a grid structure, the segmentation and concatenation need only single traversing of trajectories or grid cells. Experiments demonstrate that the unsupervised algorithm correctly discovers trajectory corridors from the real trajectory data. The trajectory corridors using Fréchet distance with temporal information are different from those having only spatial information. By choosing an appropriate grid size, the computing time could be reduced significantly because the number of sub-trajectories in a single grid cell is a dominant factor influencing the speed of the algorithms.