Anomaly Detection On Intelligent Video Surveillance Based On Bayesian Hierarchical

With the rapid development of computer vision processing technology, electronic technology, Information and Communications Technology and intelligent information processing technology, intelligent video surveillance system has been widely applied in many fields such as national defense construction, traffic control, intelligent security, etc. And most of the existing video surveillance system still relies on on-site monitoring operation, which resulting in a waste of human resources. Meanwhile, it affects the efficiency of the entire operation of the system. Therefore, to study the key technology of intelligent video surveillance system and improve the performance of video surveillance has important theoretical significance and practical value.Currently, the research and application of intelligent video surveillance are facing many problems, many scholars are engaged in research in the field and made a lot of results. Based on these results, this article mainly aimed at the feature extraction of moving target and abnormal event detection in intelligent video surveillance system. The main work is summarized as follows:Firstly, this article briefly introduces the main task, related technology and its applications of intelligent video surveillance; describes the Bayesian approach, hierarchical Bayesian methods and basic arithmetic; summarizes the application and the advantages and disadvantages of clustering methods commonly used; lists out several clustering algorithm performance.Secondly, specific to the problems of the unstructured video file, the great amount of information and data stored in the form of color pixels, brightness and location of the target object, and the huge diversity of the performance characteristics of the content. This article realized the analysis of the video files by using the techniques of prepossessing and a more mature text processing techniques.Thirdly, Aiming at moving target feature extraction problem in intelligent video surveillance, this article introduce the method of improved Pyramid Lucas-Kanada(PLK) optical flow. The traditional methods of Horn-Schunck optical flow belongs to the dense optical flow algorithm, its calculating amount of is quite large; Lucas-Kanada is a sparse optical flow method, which solves the problem of big calculation. However, this method has many restrictions, thus affecting the validity of the optical flow method. The basic idea of PLK optical flow method is a pyramid structure of the image sequence, the higher layers is lower smoothed down sampling form the original image layer is equal to zero. The method greatly induces the possibility of movement satisfies the assumption, so as to realize the characteristics of fast moving object extraction. Under the consideration of the poor robustness of least square method used in PLK optical flow algorithm, this article improve the PLK optical flow method by using the method of weighted least squares. The experimental results show that, compared with the traditional optical flow method, improved PLK optical flow method has good effect of feature extraction.Fourthly, under the considerations of the problem of anomaly events detection of intelligent video surveillance, the paper puts forward the weighted hierarchical Bayesian model. The core idea of the model is using hierarchical prior, the selection of prior distribution, and the basic idea of it is, when people grasp the overall structure and the details of a priori information at the same time, it build a model in a phased manner, when the prior distribution is given by hyper parameters is difficult to determine. And this new transcendental decide by a priori and super prior together can be called the hierarchical prior. And this model will be hierarchical Bayesian analysis theory used to model the prior distribution assumption, which help to eliminate prior distribution to estimate the result of excessive impact, enhance estimation robustness, make the model has strong applicability. The experimental results show that compared with traditional Bayesian, the model has good effect of abnormal event detection.