Sub-trajectory- and Trajectory-Neighbor-based Outlier Detection over Trajectory Streams.

Precisely and efficiently anomaly detection over trajectory streams is critical for many real-time applications. However, due to the uncertainty and complexity of behaviors of objects over trajectory streams, this problem has not been well solved. In this paper, we propose a novel detection algorithm, called STN-Outlier, for real time applications, where a set of fine-grained behavioral features are extracted from the sub-trajectory instead of point and a novel distance function is designed to measure the behavior similarity between two trajectories. Additionally, an optimized framework(TSX) is introduced to reduce the CPU resources cost of STN-Outlier. The performance experiments demonstrate that STN-Outlier successfully captures more fine-grained behaviors than the state-of-the-art methods; besides, the TSX framework outperforms the baseline solutions in terms of the CPU time in all cases.