Vehicle Trajectory Anomaly Detection Method Based on Trajectory Representation and Clustering
Xinrui Zhong · 2024
Trajectory anomaly detection is an important technique for identifying abnormal patterns that significantly differ from normal behavior in the trajectory data of moving objects. This paper explores a trajectory anomaly detection method that integrates trajectory representation and clustering algorithms. This research employs LSTM-RNN to represent the GPS trajectory data of urban vehicles, converting variablelength vehicle trajectories into fixed-length vectors. The DBSCAN algorithm is then applied for trajectory clustering, effectively identifying anomalous trajectories. Experimental results demonstrate that our method can accurately extract abnormal patterns from large-scale trajectory data, achieving a silhouette coefficient of 0.336, indicating good clustering performance.