Hierarchical Background Subtraction Algorithm For Foreground Background Segmentation

Segmentation of moving objects from a video sequence is one of the fundamental and critical task in automated video surveillance, and it gets challenging when the back-ground is non-stationary. Background subtraction is a widely used algorithm for real-time segmentation of moving objects in the presence of the static camera. Various pixel-based or block-based background subtraction approaches are available in the literature. Pixel-based methods generate smooth contours of foreground objects while block-based methods are more robust to the dynamic background. In this dissertation, we propose to combine both block-based and pixel-based background subtraction techniques in a hierarchical manner for better foreground detection. Motion segmentation problems such as dynamic background, illumination variations, and noise are addressed at block-level. Then a fine level processing of the foreground regions is done on the fore-ground background model obtained at block-level processing to smoothen the foreground objects