Virtual Running of GPS Vehicles for Trajectory Analysis

Mining moving patterns from object trajectories is an crucial task for understanding movement characteristics. In this study, we propose a novel trajectory analysis method to detect split-and-merge patterns based on a virtual running method. A virtual run is a rearranged moving order of objects that minimizes the cost function and is used to perform sub-trajectory clustering from whole trajectories. We reformulate the problem by locating branching and junction points as the covering convex hulls. We manipulating virtual runs to maintain the perimeter of the covering convex hulls. We can then easily locate branching and junction points by determining whether the perimeter of the convex hulls exceeds or falls below a given threshold parameter. Once these points are detected, the trajectories are reclustered by splitting or merging the convex hulls. We conducted experiments on a real-world GPS vehicle trajectory dataset, and the results indicate that the proposed method effectively detects major features on the road network.