Human Behavior Analysis Algorithm Based On Video Surveillance

Video-based intelligent monitoring system is one of the important issues about the social security, the impact of production and the academic research.Firstly, we elaborate on the prospects of video detection technology and it points out that the existing GMM has two problems:the one is the calculation speed; the other is that it's difficult to deal with the effect of light on the background. At the beginning of this article, we propose scanning lines update and template update algorithm to increase speed. The scanning lines update algorithm is similar to the template update algorithm. But the experimental result shows template update algorithm is better which can also meet the need of real-time monitoring. Next, the thesis deals with the light by the combination of three difference method and a Gaussian mixture model. We do the experiment at unmanned walking and someone walking scene with dramatic light changes. We find that the result is much better than the original Gaussian mixture model algorithm. Then, based on three-step search algorithm, we propose motion vector evaluation algorithm which is appled to the number of on and off, effectively detecting the direction of movement of the human body.In the thesis, a large number of original video data are collected, we study the shelter from the camera, area detection, detection of rising, and fights and other scenes to identify the human behavior. Firstly, we introduce the camera occlusion, elaborate the judgment algorithm based on gray value and gradient difference and show the experimental results under the simple scene the camera is completely occluded. Then, we introduce the rising detection algorithm and regional detection algorithm by using human behavior detected foreground image, introduce the basic principles and experimental procedures of these two algorithms and give the experimental results. Finally, focusing on the fights behavior characteristics, from global and local features of the fights behavior, we give the judgement rules and a detailed description of the experiment. This article gives the judgement of behavior from the entropy, the maximum amplitude and the violent fights proportion of movement area. Experimental results show that the algorithm not only has a good scenario-based detection effect of the above scenarios and provides experimental evidence to more complex scenes study, but it also lays the basis for the future algorithm.