Research On Dynamic Background Segmentation Based On Virtual-Real Cooperation

In virtual-real cooperative system, people are more concerned about the movement of foreground image, so we need to remove the background image in real time. Due to background has a lot of relative motion,it is an important and difficult work for us to achieve the real-time dynamic background separation. To achieve the real-time moving foreground extraction from the dynamic background of video sequence, this paper utilizes Codebook method to construct background model for dynamic background sampling and extract the foreground image in real time by computation between the current coming video streaming and background model in the GPU. on this basis, This paper designs and achieves a virtual-real cooperative system in real time.This paper proposes a dynamic background separation technology based on GPU programming utilizing Codebook algorithm. Although the Codebook background modeling method can deal with the dynamic change of each pixel in the background, it does a lot of operations in the process of creating codebook, which affects algorithm processing speed. According to this problem, this paper proposes a dynamic background separation technology utilizing Codebook method based on GPU programming. Codebook background modeling method is divided into three steps: Background construction initialization in GPU; Compute and match between the current video frame pixel and background model in GPU and detect the foreground target image; update background Codebook model the dynamically at the same moment of foreground detection. In this way, this paper achieves the real-time background segmentation by the parallel processing technology.This paper puts forward to a method that deletes the obsolete codewords unmatched for a long time at the same moment of foreground detection and background learning. The current frame image need to be updated by the following image, In the process of foreground detection, this paper deletes the obsolete codewords unmatched for a long time. Compared with traditional Codebook algorithm, the method this paper proposes match more efficiency and implements the dynamic background segmentation.This paper proposes a method that removes the foreground image noise and enhances the connectivity based on the Gibbs-MRF random field theory. Utilizing Gibbs-MRF algorithm, This paper translates the research of prior model into energy function and the issue of foreground image parameters into the optimal solution of the original image, By this way, eliminate the noise in the image and enhance the image connectivity so as to realize accurate background separation.Based on the above proposed algorithms, this paper achieves the dynamic background separation in real time. On this basis, this paper develops and achieves a virtual-real cooperative system in a real time, further more, this paper designs several applications, including identification of human movement and virtual smiling faces accompany, the apple slice game, the real-time vehicle and character detection on the public road. The experimental result shows that the proposed algorithms can achieve the foreground detection from dynamic background better, and have a real-time performance.