Anomaly Detection by Analyzing the Pedestrian Behavior and the Dynamic Changes of Behavior

作者：Shen, Maying; Jiang, Xinghao^*; Sun, Tanfeng

来源：13th International Conference on Intelligent Computing (ICIC), 2017-08-07 To 2017-08-10.

DOI：10.1007/978-3-319-63309-1_20

• Anomaly detection
• Adaptive Anomaly Weight (AAW)
• Pedestrian segmentation
• Dynamics of Pedestrian Behavior (DoPB)
• Locality Sensitive Hashing (LSH)

Since there is an increasing demand of security and safety assurance for public, anomaly detection has been a great focus in the field of intelligent video surveillance analysis. In this paper, a novel method is proposed for anomaly detection through the analysis of the pedestrian behavior with motion-appearance features and the dynamic changes of the behavior over time. Locality Sensitive Hashing (LSH) functions are used in the method to finally detect the abnormal behaviors. Compared to the relative works, the main novelties of this paper mainly includes: (1) the pedestrians in the image are segmented with the method of Robust Principal Component Analysis (RPCA) in the preprocessing step; (2) in order to describe the dynamic changes of behavior, the Dynamics of Pedestrian Behavior (DoPB) feature on Riemannian manifolds is proposed as the individual descriptor; (3) during the detection process, the Adaptive Anomaly Weight (AAW) with block-based optical flow tracking is used to measure the anomaly saliency. Experimental results and the comparisons with state-of-the-art methods demonstrate that the proposed method is effective in anomaly detection and localization.

• 出版日期2017
• 单位上海交通大学