Spatio-Temporal Isolation-Based Online Anomalous Trajectory Detection

Abstract

The ubiquity of GPS data from taxis has spurred research in spatio-temporal trajectory analysis. Anomaly detection within these trajectories is vital for ensuring safety and efficiency. Existing methods address this by analyzing spatial and temporal features like detouring, delays, and inconsistent orientations, across various scales. However, online anomaly detection of subtrajectories, crucial for real-time response in streaming environments, presents a challenge due to data complexity and the need for swift action. While prior studies focused on spatio-temporal subtrajectory anomalies, they often suffer from over-detection due to limited coverage between specific source-destination pairs. This paper introduces the Spatio-Temporal Isolation-Based Online Anomalous Trajectory Detection (ST-IBOAT) methodology. ST-IBOAT offers a novel approach for adaptive online detection of subtrajectory anomalies. It employs a dynamic reference set coupled with an adaptive working window strategy, effectively capturing realistic driver decision-making based on the spatio-temporal context.