Detecting static occlusion edges using foreground patterns

Static occlusions are a common impediment to successful object tracking in many realistic scenes. Knowledge about the locations of occlusions in the field of view of video cameras can allow tracking algorithms to successfully handle occlusion events. We present a simple and efficient rule-based method for finding large, rigid occluders in a scene by analysis of images from a single camera. Pixels along occlusion edges are identified through specific spatiotemporal patterns occurring in the binary foreground segmentation masks obtained from the input video. The final output of our algorithm is a binary mask indicating the locations of static occluders in the scene. We present experimental results from several outdoor scenes and compare the performance of the algorithm with a previously proposed method.