Background Modeling Based on Subpixel Edges

We propose an approach to model the background of images in a video sequence based on subpixel edge map. This work is motivated by the observation that intensity based background models are sensitive to changes in illumination and camera parameters, e.g., gain control. In addition, the false positive rate is higher due to accidental alignment of figure intensities with the background model. Background models of edge maps, however, are more localized and thus reduce the likelihood of accidental alignment. We argue that the discretization error in pixel-level background models is also responsible for some of the false positives and develop a method based on subpixel edges whose background is thus highly selective. This method models the edge position and orientation using a Mixture of Gaussians model. This approach has been tested on a wide range of videos and the resulting background models are a much more selective figure-ground segregation.