Anomaly Behavior Detection Based On Intelligent Video Analysis

As a cutting edge technology in computer vision and image processing, Intelligent video surveillance has been utilized in multiple realms, including large infrastructures such as military bases, supermarket, and small facilities as convenience stores, classrooms. However, lots of surveillance systems are not intelligentized, known as the traditional surveillance system, which are not able to detect anomaly behaviors instantly and automatically. In other words, staffs have to search the video recordings manually to look for information related to abnormal activities, which is less efficient and accurate. The traditional surveillance calls for people to monitor the screens all along for precaution, and this kind of non-intelligent surveillance system is not only a consuming of time but manpower. Therefore, it’s vital to research and develope the technology of intelligent video surveillance, intelligent video technology is implanted with one or multiple video picture processing algorithms, which implements real-time monitoring and scene learning. Based on the result of machine learning and analysis, the surveillance system will generate alarms when anomaly behaviors take place, or call out staffs to cope with situation before it is too late.Human action recognition is conceived the core technology of Intelligent video surveillance, which possesses extremely high scientific research value and application prospect. Specially, the research of fall detection become hotspot due to population aging. Fall in elderly people my cause a series physical damages, such as unconsciousness, disability. In addition, fall behavior may trigger disease complications. Therefore, it’s totally necessary to develop a method to prevent people from suffering secondary damage when they fall by setting alarm and call for help.This paper separately discusses two main subjects. Firstly, this paper developes an intelligent surveillance system autonomously, secondly, an algorithm of fall detection is researched, which can be implanted in the surveillance system we develop with corresponding interfaces as an algorithm module.1. The surveillance system is developed on the Windows7system, using Visual Studio2010and Opencv function library as tools. This system is compatible with multiple kinds of video cameras and can play real-time video swimmingly. It is embedded with several human action recognition algorithms to detect anomaly behaviors in the monitoring scene. Besides, the intelligent surveillance system can also manage the warning messages, capture pictures and video segments of abnormal activities, implement functions such as displaying digital-map, accessing NVR/DVR remotely, playing video recordings,.etc.2. Through investigations and researches relate to fall detection, this paper proposed an algorithm to monitor fall behavior. The algorithm is combined with the threshold method and the Gaussian Mixture Model method to fulfill detecting and tracking moving objects, and remove the shadow and noise in the segmented image sequences. When the outline rectangle of the moving objects is acquired, the effective area ratio can be calculated. By setting the threshold, fall detection algorithm can analyze the result of both outline aspect ratio and effective area ratio. When these two scenario take place at the same timeframe, it can be determined as fall. Experimental results shows that the algorithm can accurately detect fall behavior in time, and has low rate of false detect rate.The Intelligent video surveillance system and the algorithm have been tested in multiple situations, and the results indicates that both the system and the algorithm can operate steadily, and show good robustness.