A probabilistic approach for an active contour-based background subtraction

Abstract

Contour-based background subtraction algorithms are always focusing on the edge of objects in the scene. Most of them may fail when an initial contour is accidentally located inside the foreground object. From the fact that background and foreground image are usually independent to each other, we introduce a contour-based background subtraction algorithm which minimize the dependency between them. Not only guiding contour out to cover foreground object when the predicted contour is accidentally located in the foreground object, the algorithm can be run along the video frames without human labor required by predicting initial contour from the dependency between foreground and background.