Anomalous Taxi Trajectory Detection Based on Experiential Constraint Rules and Evidence Theory

In the context of urban transportation, large-scale collections of floating car trajectories are constrained by low sampling rates due to concernsabout data processing and storage. This creates uncertainty when identifying movement trajectories that reflect true route choice behaviors. To reduce uncertainty, this paper presents an approach using experiential constraints based on evidence theory to detect anomalous trajectories in taxi GPS trajectories. The approach employs three factors including the ratio of travel length between GPS traces and shortest distance path, the cost index of experiential avoid roads, and travel start times. The evidences based on the three measurements are combined in an evidence theory framework in order to get the anomalous degree of each trajectory so that the anomalous trajectories whose travel distance and travel route are significantly different from normal ones can be detected. A case study is presented using real world GPS trajectories of over 11, 000 taxis. The experimental results in Wuhan show that our method, which is not influenced by the number of trajectories between a single OD pair, has the ability to detect anomalous trajectories and can be applied to clean biased data before route choice analysis using a large fleet of floating cars.