Foreground objects detection using multiple difference images

Abstract

In visual surveillance, robust foreground object detection is an essential step for further processing such as segmentation, tracking, and extraction of a scene's contextual information. Typical approaches continuously update background images and use then for detecting foreground objects. They involve many parameters that should be adjusted according to the situation where surveillance cameras are operating. We propose an algorithm for the robust detection of foreground objects using multiple difference images that requires only one parameter to adjust. We show that the proposed algorithm gives comparable results with less computation time through experimental results using test images with groundtruths.