STL: Online Detection of Taxi Trajectory Anomaly Based on Spatial-Temporal Laws

Abstract

Aiming to promote the standardization of taxi services and protect the interests of passengers, many methods have been proposed to detect taxi trajectory anomaly based on collected large-scale taxi traces. However, most existing methods usually employ a counting-based policy to differentiate normal trajectories from anomalous ones, which may give rise to high false positives. In this paper, we propose an online detection method, named Spatial-Temporal Laws (STL). The basic idea of STL is that, given the displacement from the source point to the current point of a testing trajectory, if the current point is normal, either its driving distance or driving time should lie in a normal range. STL learns the two ranges from historical trajectories by defining two spatial-temporal models: one characterizing the relationship between displacement and driving distance, and another depicting the relationship between displacement and driving time. Consequently, STL is more precise compared with the counting-based methods, greatly reducing the number of false positives. Based on large-scale real-world taxi traces, STL is evaluated through a series of experiments which demonstrate its effectiveness and performance.