Background subtraction for surveillance systems using an iterative uniform operator

Variations of the illumination in a structurally stationary scene, noise of cameras, shadows and small fluttering of the background objects are main causes that usually challenge the foreground object detections for most of the surveillance applications. In this paper, a novel background subtraction approach using a simple iterative approximation process with a uniform operator is presented. The process, assuming that the region is uniformly illumination-distributed when illumination of the scene varies, will normally derive two categories of results, one uniform, and the other non-uniform. The foreground objects will be extracted from the non-uniform regions after 3~4 iterative steps. The detection rule is very simple and uses only one reference image. The list of the regions, following by sizes of these regions, is generated and updated by the iterative process itself. All sub-regions at the same iterative step can be processed in parallel. 5~6 background images under similar illumination conditions are helpful at the beginning to evaluate the basic parameters of the estimation. Experiments show that this approach is effective in detecting the foreground objects in illumination-variant scenes.