DIRECT INCREMENTAL MODEL-BASED IMAGE MOTION SEGMENTATION FOR VIDEO ANALYSIS

Dynamic analysis of image sequences is an important task in object-ori ented video applications. It often relies on the segmentation of each image of the sequence into region entities of apparent homogeneous mot ion. In this paper, we present an original motion segmentation algorit hm based on 2D polynomial motion models, a multiresolution robust esti mator to compute these motion models, and appropriate local observatio ns supplying both motion relevant information and their reliability. M otion segmentation is formulated as a contextual statistical labeling problem exploiting multiscale Markov random field (MRF) models. One of its main features is that it avoids time consuming alternate iteratio ns between motion model estimation and spatial support identification. An original detection step allows us to estimate and to update the nu mber of required motion models, and thus to handle the appearance of n ew objects. Numerous experiments performed with real indoor and outdoo r image sequences demonstrate the efficiency of the method. (C) 1998 P ublished by Elsevier Science B.V. All rights reserved.