Parametric multiple-motion estimation for motion segmentation

We describe a parametric multiple-motion estimation approach for motion segmentation. Towards such a goal, a reliable optical flow estimation method is introduced which deals with the sparse optical field problem from the view of missing data reconstruction. Multiple-motion parameter estimation is carried out on the basis of constancy of motion field for each motion model. Also to achieve a smooth motion boundary, spatial segmentation information is utilized. Experimented results are presented on a cluttered video sequence with fast moving object motion.