Abstract: Based on image blocks,a method for constructing background models is presented to reduce computation redundancy arising from pixel-background model and to improve execution speed of the system.After reviewing the main methods of background extraction up to now,we present a partitioning method and some common features for the image blocks,and construct some adaptive mixture Gaussian models with these features.Experimental comparison between this method and the traditional pixel-background models is made with a group of videos.The results show that this method enhances system execution efficiency greatly at the same finding-out rates.