Research Of Video Background Modeling Algorithm

In video surveillance, intelligent transportation and other fields, the first step is usually to detect moving objects in the video image, and the detection results is used in target tracking,target classification,behavior understanding and other subsequent processing.The moving target detection is the foundation and key in the analysis of video sequences. Many scholars have done a lot of research for foreground detection, which is still one of the hot field of computer vision.The main works and research results are summarized as follows:(1)To get a good background modeling quickly in complex scenes, and detect moving objects effectively is very difficult.In this paper we have done a more in-depth research for the question,and place emphasis on Gaussian mixture model,Code Book,vibe,and difference algorithm.(2)For lack of difference algorithm, according the changing of neighborhood pixel values,we have improved difference algorithm.The difference algorithm improved has a stronger adaptability for complex background and noise.At the same time, through the foreground image detected by vibe algorithm, we fill the "hole" of image detected by differential algorithm.By the combination of difference algorithm improved and vibe algorithm, we have effectively solved the problem of double shadow of difference algorithm, and also make vibe algorithm having a faster convergence rate to "ghost" phenomenon.Selecting the appropriate segmentation threshold, the shadow suppression will have a certain effect.(3)Generally complex scenes have multimodal distribution,and distribution of each pixel has its similarities.This article has improved CSDP clustering algorithm.In the establishment phase of the background model, we cluster the pixels and get a background model by the CSDP clustering algorithm improved. In order to adapt to changes in local and global illumination in the foreground detection,the algorithm,from the two dimensions of brightness and color, matches the pixel values. After successfully matching pixel,the background model is updated,in order to adapt to slow changes of the background. When the static background model changes suddenly, our algorithm can quickly update the background model, remove non-existent model in the background, or add a new model, solving the problem such as vehicle suddenly starts, or stays a long time.