Recognition Analysis Of Human Behavior Based On Video

Human behavior recognition is through a computer to video or image sequence of human intelligent analysis, understanding and recognition, computer vision belongs to the Advanced section. With the popularity of video technology in the field of computer vision gradually toward the white-hot, more and more scholars began to get involved in this field. Recognition of human behavior in human behavior based on video recognition technology refers to the interception of a video image sequence frame by frame, and then find the target in the image sequence, tracking the target until the last track by continuing to extract a series of motion characteristics, thus determining the action category. After many years of effort in domestic and foreign researchers, and human behavior based on video recognition technology has made remarkable progress. But practice shows that the recognition stage can only make a better analysis of some simple background environment, their level of maturity far unable to meet the real needs of life, so he joined a more robust algorithm in video recognition technology used in the past meet the real needs of nature become imperative. The main topic of human behavior based on video recognition, by using image detection, video tracking, target identification and a series of effective means to analyze information from a video, and make specific judgments such behavior. The main innovation of this thesis research points are summarized below: Target detection algorithm in a lot of the pros and cons of comparative analysis found that the algorithm based on Gaussian mixture model to better reflect the robustness of the target detection process. When using Gaussian mixture model to do background subtraction method, real-time updates will bring huge background model of computation, this time, the first inter-frame difference method using motion regions and background region roughly divided can greatly reduce the area at the time of update calculation.(1) at present, the main target detection method mainstream light flow method, background subtraction, inter-frame difference method and Gaussian template method. Optical flow method for calculating hardware requirements and time consuming too large; inter-frame difference method is relatively simple, but practice shows that the effect is not ideal; background subtraction method based on Gaussian mixture model is a better choice, first create a Gaussian background template, and then match the image frame, if the difference is small, it is determined that the background of the target, if the difference is large, it is determined that the foreground object. When using this method, the template can be updated in real time in favor of the subsequent tracking and matching, which is a big key advantage of background subtraction method based on Gaussian mixture model.(2) in terms of video tracking, the paper used the Meanshift algorithm and algorithm for a detailed analysis and research. This topic when doing video track successfully used Meanshift algorithm, a large number of experimental data show Meanshift algorithm processing edge occlusion, target rotation, can achieve good results during deformation and background motion insensitive and other issues. However Meanshift algorithm also has some bottlenecks, it’s the size of the target area has been restricted, in dealing with the problem of robust template update is not strong.(3) this thesis gradient algorithm and features multi-feature fusion algorithm based on multi-class movement were to identify the key points based on behavioral research, get a lot of experimental data, experimental results show that the effect of the above two methods to identify relatively satisfactory final in this paper, the merits of these two algorithms are described in detail.