Integrated region- and pixel-based approach to background modelling

In this paper a new probabilistic method for background modelling is proposed, aimed at the application in video surveillance tasks using a monitoring static camera. Recently, methods employing time-adaptive, per pixel, mixture of Gaussians (TAPPMOG) modelling have become popular due to their intrinsic appealing properties. Nevertheless, they are not able per se to monitor global changes in the scene, because they model the background as a set of independent pixel processes. In this paper, we propose to integrate this kind of pixel-based information with higher level region-based information, that permits one to manage also sudden changes of the background. These pixel- and region-based modules are naturally and effectively embedded in a probabilistic Bayesian framework called particle filtering, that allows a multi-object tracking. Experimental comparison with a classic pixel-based approach reveals that the proposed method is really effective in recovering from situations of sudden global illumination changes of the background, as well as limited non-uniform changes of the scene illumination.