Shadows removal by edges matching

Abstract

Motion detection algorithms usually detect moving regions in a rough way; in some application contexts it could be mandatory to obtain the exact shape of such objects by removing cast shadows as well as ghosts and reflections due to variations in light conditions. To address this problem we propose an approach based on edge matching. The basic idea is that edges extracted in shadow (or ghost) regions in current image exactly match with edges extracted in the correspondent regions in the background image. On the contrary, edges extracted on foreground objects have not correspondent edges in the background image. In order to remove all shadow regions instead of only shadow points, we firstly segment the foreground image into subregions, according to uniformity of photometric gain between adjacent points. The algorithm has been tested in many different real contexts, both in indoor and outdoor environments. © 2009 Springer-Verlag.