TS-DBSCAN: To Detect Trajectory Anomaly for Transportation Vehicles

Abstract

Deep exploration of the potential characteristics from vehicle trajectory data benefits transportation safety management and improves transport efficiency. Therefore, trajectories anomaly detection plays a pivotal role in transport enterprises. In this paper, we proposed a novel density clustering model named TS-DBSCAN to detect outliers of trajectory data, which is a DBSCAN-based method for clustering time-series data. We first analyzed the time correlation of the trajectory data of transportation vehicles. Then the distance between two adjacent timestamps is considered as the training data of DBSCAN clustering algorithm, determining the $\upvarepsilon$-neighborhood radius $\left( {Eps} \right)$ and the minimum neighbor number $\left( {MinPts} \right)$ according to the distance density distribution. Finally, anomaly clusters are detected from the trajectory data. We conducted experiments based on real trajectory data of transportation vehicles to evaluate the effectiveness. The experimental results show that TS-DBSCAN algorithm can detect abnormal trajectory data with both efficiency and accuracy.