In this article, a novel pixel based object detection framework is proposed that leverages dual type pixel-level information to construct the background model.