Trajectory Anomaly Detection with Multi-feature and Multi-resolution Based on Direction Grouping

• Published in: 2021 16th International Conference on Intelligent Systems and Knowledge Engineering (ISKE) pp. 503 - 508
• Main Authors: Guo, Jiangbo, Zhu, Zhixiang, Wang, Chenwu, Wang, Pei, Xu, Rongyao, Hou, Gen
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.11.2021
• Subjects: Anomaly detection; Clustering methods; DBSCAN clustering; Feature extraction; Fitting; Knowledge engineering; Motion segmentation; Multi-feature; Multi-resolution; Surveillance; Trajectory; Trajectory clustering
• DOI: 10.1109/ISKE54062.2021.9755325

Unreasonable trajectory segmentation will lead to inaccurate results about pedestrian anomaly detection. To solve this problem, this paper proposed a multi-feature and multiresolution feature representation method combining global and local features of the trajectory. First, the trajectories are grouped using the DBSCAN clustering algorithm according to the direction of movement. Generate a Fitting trajectory based on all trajectories in the group. Then, the trajectory is segmented to use different resolutions, The offset distance of the sub-trajectory and the Fitting trajectory at each resolution is calculated using the position, velocity, and direction features Abnormal trajectories are detected using DBSCAN clustering. The motion pattern of the trajectory is accurately represented using a multi-resolution method combined with contextual features. The abnormal trajectory can be detected more comprehensively through the multi-feature method. The experimental results show that the anomaly detection effect of our method is better than the existing trajectory anomaly detection methods using clustering.