Study On Technology Of Video Analysis For Intelligent Surveillance

Intelligent Video Surveillance(IVS) is an important direction of computer vision. It aims to make the video surveillance system to be high intelligent by accurately controlling the video surveillance system based the result come from the description, analysis and understanding of the surveillance video’s content. It rapidly becomes the research hot spot of security because of its obvious effect in all-sidedly preventing and effectively managing the terrorism. Thus, it is of great importance and valuable to study the technology of video surveillance in depth.The basic problem faced IVS, which must be key solved, is moving object detection, tracking, classification and recognition. In this paper, these three key links are deeply studied with different methods, and the main work includes:(1) In the research of moving object detection, we put forward a novelty moving object detection’s method based on Low-Rank Representation. This method gets the forward-ground moving object labels by modeling the moving object detection as outlier detection, and modeling the complication background with Low-Rank Representation and utilizing the alternating optimization algorithm. The experimental results on the public data have validated the detection accuracy of proposed algorithm is higher than the classical methods.(2) For moving object tracking, we presented an object tracking algorithm based on metasample sparse representation. A set of metasamples were firstly extracted to construct the object dictionary. And then the overcomplete dictionary was built by adding trivial template. For tracking, an iterative algorithm was proposed to solve 1l-norm minimization. The experimental results indicate that the proposed method is more robust than the existing methods, and is able to dealing with the object tracking under occlusion too.(3) Last, this paper studied the object classification problems. Multi-view pedestrian samples often contains such high intra-class variances that multi-view pedestrian classification suffers from high classification error. To solve this problem, a novel multi-view pedestrian recognition algorithm based on non-negative matrix factorization(NMF) and least square is proposed in this paper. Firstly, by using the NMF, the subspace of multi-view pedestrian samples is acquired and base vectors are extracted. Secondly, by introducing the collaborative representation the sparse presentation of the subspace is performed constrainedly by the least square, and then sparse coefficients are obtained. Finally, multi-viewpoint classification is completed using sparse coefficients based on the nearest subspace rule. The comparison experimental results on the self-established multi-view pedestrian dataset show that the proposed method outperforms several state-of-the-art methods in terms of accuracy and effectiveness.