Intelligent Detection Technology For Object Of Interesting In Video

With the development of the intelligent city, video surveillance system is playing more and more important role for its highly effective and intelligent, convenience and swift features in the aspect of public security, traffic management, environment protection, etc. It not only brings convenience for people but also vast amounts of video information redundancy. Moreover in the actual application, only some specific moving objects are cared by people. In order to improve the detection efficiency of video data and satisfy the needs of application, in this paper we research the intelligent detection technology for the object of interesting (OOI) in video surveillance based on the existing research results from the aspect of computer vision. The specific research jobs are as follows:Assuming that people are interested in all the objects in video, Graph-Cuts algorithm is one of the common detection methods adopted. But most Graph-Cuts optimization methods are based on the hypothesis that foreground obeys the uniform distribution, which results in a fixed boundary term and the deviation appears on the target edges. This paper aims to modify the boundary term and the optimized detection method is proposed. The proposed method constrains the weighted gradient by the target edge length extracted in the previous frame so that it can adaptively modify the boundary term of the next frame. Thus can solve the edge deviation of target and improve the detection performance at the same time.Because that people are not interested in all the moving objects in the actual video surveillance, so detection for OOI needs to rely on the needs of applications. Markov Random Field (MRF) theory is used to train the object model for its characters of low training complexity and regardless of illumination variation and noise. The detection method based on the OOI model is proposed in this paper. The proposed method obtains the high-order prior information of object through the trained model, and optimizes the parameters of energy function by the Pseudo-Boolean theory during the detection process, and then the OOI could be extracted. This method can satisfy the detection requirement for OOIs in actual application for its model is robust and lower complexity.In this paper we research two kinds of OOI detection methods based on the feature and model of OOI. In the experiments, we perform the tests for video sequences in different situations, such as the complex and simple background, single and multiple moving objects, with and without illumination change. Finally the results of simulation and detection indicators'analysis verify that the two methods could detect the OOIs well.