Anomaly Detection via Trajectory Representation

Abstract

Trajectory anomaly detection is a vital task in real scene, such as road surveillance and marine emergency survival system. Existing trajectory anomaly detection methods focus on exploring the density, shapes or features of trajectories, i.e., the trajectory characteristics in geography space. Inspired by the representation of words or sentences in natural language processing, in this paper we propose a new anomaly detection in trajectory data via trajectory representation model ADTR. ADTR first groups all GPS points into semantic POIs via clustering. Afterwards, ADTR learns POIs context distribution via algorithm of distributed representation of words, which aims to represent a trajectory as a vector. Finally, building upon the derived vectors, the PCA strategy is employed to find outlying trajectories. Experiments demonstrate that ADTR yields better performance compared with state-of-the-art anomaly detection algorithms.