Robust 3-D motion estimation and depth layering

Two problems are addressed in this paper: camera 3-D motion estimation and image depth layering. For the first problem, the method we adopt is the use of a hierarchy of motion models in combination with robust estimation methods. A dense motion vector field is assumed to be granted. For the second, an iterative deterministic relaxation algorithm is used and the extraction of the image depth layers is based on both optical flow and luminance information.