Analysis of scene anomaly recognition method based on machine learning
Ruiying Liu, Xinyu Yuan, Bingjie Yu, Yuxue Yang, Teng Wang · 2021
Among the many video surveillance problems, we usually pay the most attention to anomalies. The so-called exception can have different definitions for different scenarios. The detection and judgment of abnormal phenomena often need to rely on other intelligent video surveillance technologies: moving target detection and tracking, human action recognition and behavior analysis, interactive behavior understanding, crowd analysis, etc. The research of this paper is mainly to detect and analyze abnormal behaviors in monitoring scenarios. For different monitoring scenarios, two different anomaly detection ideas are proposed. One is based on target detection and tracking, which extracts the motion information of the target area by locating the position of the person in the scene, and analyzes abnormal behavior on the basis of action recognition. Another idea is to analyze the scene based on the abnormalities of the video image texture features. The former has a relatively robust detection effect when it can accurately track and locate moving targets, while the latter is suitable for abnormal detection in crowded people.