Dense Estimation and Object-Oriented Segmentation of the Optical Flow with Robust Techniques

In this paper we address the intricate issue of recovering and segmenting the apparent velocity field between consecutive frames of an image sequence. As for motion estimation, we design a global cost functional to be minimized among the set of possible velocity fields. In the same spirit as the energy function proposed by Black et al, we consider two terms, both involving a robust M-estimator. The first one reinforces the fragile modeling of the optical flow constraint equation, while the second (a priori) term incorporates a discontinuity preserving smoothness constraint. A multiresolution formulation of this differential estimation method aims at accessing long range displacements in a coarse-to-fine incremental way. As for the minimization associated with robust estimators, we define a very efficient deterministic multigrid relaxation algorithm which converges fast toward estimates of good quality. Besides, the resulting iterative estimator is able to produce, at very low cost, "crude" estimates..