A real-time motion detection for video surveillance system

This paper describes an approach to moving objects detection in video stream with the usage of improved background subtraction and model updating methods as well as hierarchical data structure. The improved method of background model updating allows to change dynamically the speed of such updating depending on the average change of pixels´ values. Applying of the modified algorithm of quadtree creation allows to speed up the performance by comparing not all the pixels, but only the random ones and also to use the quadtree structure created on the previous frame instead of creating from scratch. Experimental results show that proposed method is up to 10% faster than simple difference, up to 45% faster than running average.