An anomaly detection algorithm for taxis based on trajectory data mining and online real-time monitoring

Taking the prevention of taxi frauds as a motivating example, an anomalous spatio-temporal trajectory detection method that combines offline mining and online detection was proposed. A city roadmap was partitioned into a grid based on the longitude and latitude, using Pathlet sequences to express taxi trajectories instead of the traditional GPS sequences. Then, K-racial classes’ normal sequences were clustered in the same origin-destination pair from history data sets. The incoming online GPS data was transformed into Pathlet sequences and matched with K-racial classes’ normal sequences. The distance was computed and scored. Distance along with spatial and temporal factors together forms the criterion for determing anomalous taxi trajectories. Finally, based on the real taxi GPS data sets in Beijing area during March, 2011 to May, 2011, experimental results indicate that the proposed method is able to detect online anomalous trajectories efficiently and quickly.