A Novel Algorithm for Detecting Spatial-Temporal Trajectory Outlier

As an important area of spatial-temporal data mining, trajectory outlier detection has already attracted broad attention in recent years. In this paper, we present a novel distance measurement between spatial-temporal sub-trajectories and propose algorithm STOD for detecting outliers in both spatial and temporal dimensions jointly. Each trajectory is divided into line segments at first, and the corresponding minimal boundary boxes are constructed. After combination, the outlier index is computed with our distance measurements including overlapping volume, angle and speed. To improve the efficiency of algorithm STOD, we present algorithm PSTOD for parallel detecting spatial-temporal trajectory outlier, which is implemented using Spark framework. The experiment results on real taxi dataset show that the two algorithms are effective and efficient.