Trajectory Outlier Detection Based on DBSCAN and Velocity Entropy

Pedestrian trajectory outlier detection is of great significance in the field of video surveillance and public safety. However, due to various behavior patterns of pedestrians and complex video scenes, the trajectories obtained by real-time tracking algorithm are often complicated and accompanied by missing data. In this paper, a trajectory outlier detection algorithm based on DBSCAN and velocity entropy is proposed to detect outliers from real-time tracking trajectories. On the one hand, an improved DBSCAN algorithm is used to detect abnormal sub-trajectories, which identifies trajectory outliers with local features. On the other hand, by comparing the velocity entropy of the trajectories, trajectory outliers are detected from the overall perspective. The concept of trajectory confidence is proposed to evaluate the reliability of the results of trajectory outlier detection, thereby further reducing the false detection rate and improving accuracy. Finally, an experiment is carried out and shows that our method has a better performance than TRACLUS.