Adaptive change detection approach for object detection in outdoor scenes under variable speed illumination changes1

In this paper an adaptive algorithm is presented for the detection of changes in outdoor video-surveillance images. The proposed method can be considered as basis for a low level image processing stage in a advanced video-surveillance system. The main feature of the algorithm is the robustness to the illumination changes in the scene: this robustness is achieved by using a background updating module working in cooperation with the change detection algorithm. The background updating method works in a different way according to speed of illumination changes. If a sudden illumination change is detected the background is heavily updated while if a slow lightning variation occurs, a continuous soft updating module is used to take into account the long-term slow illumination changes. Experimental results of an automatic people counting system including the proposed low level image processing stage indicate that the method detects changes accurately in case of time-varying illumination.