Anomaly Detector Based On The Global-Local Anomaly Score For Trajectory Data

With the rapid proliferation of GPS and the increase of data volume in the information age,the importance and challenges of the research on trajectory anomaly detection algorithm are gradually highlighted.In recent years,this topic has attracted a lot of research attention.A plenty of anomalous trajectory detection algorithms have been developed in the literature.However,most current detection methods cannot well handle the complex trajectory set with varying local densities.In addition,trajectory similarity measure,which is a vital part in trajectory anomaly detection,is usually very difficult,especially for trajectories with different lengths.In this paper,we propose solutions to address these two issues.First,we propose a new measure of anomaly score,called Global-Local Anomaly Score(GLAS),to sensitively quantify the abnormal degree of complex trajectories and eliminate negative effects caused by the variation of local densities.Second,we design an effective and efficient trajectory similarity measure,the Extended Power enhanced Euclidean Distance(EPED).EPED is more sensitive to anomalies in trajectory data,and can better capture the difference between normal trajectory and abnormal trajectory.EPED can also handle unequal length trajectories.Based on the proposed anomaly score GLAS and the new similarity measure EPED,we develop a novel unsupervised anomalous trajectory detection technique,called Global-Local Anomaly Score based Anomalous Trajectory Detector(GLAS-ATD).The proposed GLAS-ATD is evaluated on both synthetic and real-world trajectory data sets.Extensive experimental results show that the new anomalous trajectory detection technique outperforms both classic and state-of-the-art methods.