Moving target detection is one of the critical issues in video surveillance. Background subtraction is a typical approach to detect moving targets in surveillance using only one static camera. The performance of background subtraction algorithms depends on how to construct background model. In this paper

codebook is used to estimate the background model using a long clip under limited memory. First

an initialization codebook is constructed for each pixel. Second

codewords in the initialization codebook are selected to construct background codebook based on the assumptions about the background model. The image can be partitioned to foreground and background according to the background codebook

and in the meanwhile codebook is updated correspondingly. Experimental results demonstrate the effectiveness of the method presented in this paper

even when there are moving clutters in the background and illumination variations.