Background subtraction for the moving camera: A geometric approach

Background subtraction is a commonly used technique in computer vision for detecting objects. While there is an extensive literature regarding background subtraction, most of the existing methods assume that the camera is stationary. This assumption limits their applicability to moving camera scenarios. In this paper, we approach the background subtraction problem from a geometric perspective to overcome this limitation. In particular, we introduce a 2.5D background model that describes the scene in terms of both its appearance and geometry. Unlike previous methods, the proposed algorithm does not rely on certain camera motions or assumptions about the scene geometry. The scene is represented as a stack of parallel hypothetical planes each of which is associated with a homography transform. A pixel that belongs to a background scene consistently maps between the consecutive frames based on its transformation with respect to the “hypothetical plane” it lies on. This observation disambiguates moving objects from the background. Experiments show that the proposed method, when compared to the recent literature, can successfully detect moving objects in complex scenes and with significant camera motion.