The Research Of Regional Moving Object Detection Algorithms Based On Video Scenes

In recent years, with the development of technology, video surveillance has gradually entered a smart era. Smart video surveillance technologies mainly include the follows:moving object detection, moving object identification and classification, moving object tracking, behavior recognition and event recognition. Moving object detection is the first task in video processing sequences, the rest tasks rely greatly on its result for further analysis. Therefore, the research of more accurate moving object detection algorithm has great significance.Among all the moving object detection algorithms, single pixel based modeling detection algorithm is widely studied; it ignores the relevancy between adjacent pixels, builds up models based on single pixel, so the detection results is often inaccurate. Regional moving object detection algorithm takes the relevancy between pixels into consideration, and makes overall judgment about whether the pixel is moving, this kind of algorithm can get more accurate results than single pixel based modeling detection algorithm. However, the level of relevancy between the pixel and its surroundings of this algorithm is still quite low, and the relevancy is independent of the scene and the detection accuracy is easily affected by noise.In this paper an effort was made to study the regional moving object detection algorithm from a higher level of relevancy, and proposed to apply moving object recognition technologies to moving object detection. Moving object recognition algorithms are often based on scenes and objects; they are freer of noise and can get better results. The main researches in this paper are as follows:(1) A Hu momentum based regional moving object detection algorithm was proposed. First, the features of Hu momentum of a region with or without moving objects were studied, to analyze the feasibility of applying Hu momentum to regional moving object detection. Then the Hu momentum based regional moving object detection algorithm based on single Gaussian model as well as Gaussian mixture model was proposed. The experimental results show that the algorithm can achieve higher accuracy.(2) A SSIM based regional moving object detection algorithm was proposed. In order to analyze the feasibility, at first the SSIM features of a region with or without moving objects was studied. Then the SSIM based regional moving object detection algorithm based on single Gaussian model as well as Gaussian mixture model was proposed. The experimental results show that the SSIM based regional moving object detection algorithm can achieve higher accuracy.(3) The emphasis of chapter four was on the performance analysis of the two previous algorithms when applied to pixel-level detection. To compare it with the traditional Single Gaussian Model moving object detection method, some experiments were conducted. The experiment shows that our algorithm has better results. On the basis of the previous research, these algorithms were applied to dual-camera moving object detection systems. At first some problems that will be encountered in single camera detection were discussed, and then the cooperative dual-camera regional moving object detection algorithm was designed, and then analyzed its principles and advantages. At last, experiments were conducted to compare the algorithm with traditional single camera detection method, and the result shows that our algorithm has better performance.