Moving Object Detection Based on Improved Codebook Model

Background modeling is a key technology in video surveillance field. Conventional methods used to detect the moving object have drawbacks in the aspects of accuracy and robustness. In order to address these problems, we change the RGB columnar structure in traditional codebook model and establish codebook background model in the YUV color space. We cluster every pixel on a timeline in an image and extract background templates. Then the current image and the background templates are compared. At the same time, the background templates are renewed. We integrate the frequency information to improve the process of symbol decision, deletion and matching. In the meantime, the spatial information of current pixel is added into the prediction of the foreground, making the results of foreground detection more reliable. The experimental results indicate that the proposed method can detect the moving objects effectively and correctly under various conditions. In addition, the proposed algorithm also has the good noise robustness. Keywords-background modeling; codebook; moving object detection; noise robustness