Moving Region Detection by Transportation Problem Solving

In this paper, we propose a novel moving region detection method from the viewpoint of solving the transportation problem. This method extracts the relations between regions as a solution to the transformation problem between pixels belonging to adjacent frames. Moving regions are detected by utilizing the properties of these relations. This method does not require any models such as prior knowledge or particular assumptions about moving objects or backgrounds in a video. Since the method adaptively detects moving regions from input frame data, it can deal with the fluctuations of moving objects or backgrounds. We demonstrate the effectiveness of the proposed method through several experiments conducted using actual videos.