MRF-motion segmentation based on dominant motion estimation and the detection of uncovered regions

This paper presents an algorithm for the segmentation of multiple moving objects. Our proposal is based on dominant motion estimation, static segmentation and Markov random field (MRF) classification of the regions obtained by static segmentation. Dominant motion estimation is based on efficient variants of the Hough transform applied over a hierarchy of motion models with increasing complexity. The final segmentation is only performed after all the motion models have been determined and is based on motion information, which includes the explicit detection of uncovered regions and on contextual properties between neighboring static regions