Interactive Media Systems, TU Vienna

Region-based Optical Flow Estimation with Treatment of Occlusions

By Michael Bleyer, Margrit Gelautz, and Christoph Rhemann

Abstract

This paper describes an algorithm for computing the optical flow field between two consecutive frames. The algorithm takes advantage of image segmentation to overcome inherent problems of conventional optical flow algorithms, which are the handling of untextured regions and the estimation of correct flow vectors near motion discontinuities. Each segment's motion is described by the affine motion model. Initial motion segments are clustered to derive a set of robust layers. The assignment of segments to layers is then improved by optimization of a global cost function that measures the quality of a solution via image warping. Occlusions in both views are detected and handled in the warping procedure. Furthermore, the cost function aims at generating smooth optical flow fields. Since finding the assignment of minimum costs is $\mathcal{NP}$-complete, an efficient greedy algorithm searches a local optimum. Good quality results are achieved at moderate computational expenses.

Reference

M. Bleyer, M. Gelautz, C. Rhemann: "Region-based Optical Flow Estimation with Treatment of Occlusions"; Talk: Joint Hungarian-Austrian Conference on Image Processing and Pattern Recognition (HACIPPR), Veszprem, Hungary; 05-11-2005 - 05-13-2005; in: "Joint Hungarian-Austrian Conference on Image Processing and Pattern Recognition (HACIPPR)", (2005), 235 - 242.

BibTeX

Click into the text area and press Ctrl+A/Ctrl+C or ⌘+A/⌘+C to copy the BibTeX into your clipboard… or download the BibTeX.