Discrete-continuous optimization for optical flow estimation

Stefan Roth, Victor Lempitsky, Carsten Rother

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

13 Citations (Scopus)

Abstract

The accurate estimation of optical flow is a challenging task, which is often posed as an energy minimization problem. Most top-performing methods approach this using continuous optimization algorithms. In many cases, the employed models are assumed to be convex to ensure tractability of the optimization problem. This is in contrast to the related problem of narrow-baseline stereo matching, where the top-performing methods employ powerful discrete optimization algorithms such as graph cuts and message-passing to optimize highly non-convex energies. In this chapter, we demonstrate how similar non-convex energies can be formulated and optimized in the context of optical flow estimation using a combination of discrete and continuous techniques. Starting with a set of candidate solutions that are produced by either fast continuous flow estimation algorithms or sparse feature matching, the proposed method iteratively fuses these candidate solutions by the computation of minimum cuts on graphs. The obtained continuous-valued result is then further improved using local gradient descent. Experimentally, we demonstrate that the proposed energy is an accurate model and that the proposed discrete-continuous optimization scheme not only finds lower energy solutions than traditional discrete or continuous optimization techniques, but also leads to very accurate flow estimates.

Original languageEnglish
Title of host publicationStatistical and Geometrical Approaches to Visual Motion Analysis - International Dagstuhl Seminar, Revised Papers
Pages1-22
Number of pages22
DOIs
Publication statusPublished - 2009
Externally publishedYes
EventInternational Dagstuhl Seminar on Statistical and Geometrical Approaches to Visual Motion Analysis - Dagstuhl Castle, Germany
Duration: 13 Jul 200818 Jul 2008

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume5604 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

ConferenceInternational Dagstuhl Seminar on Statistical and Geometrical Approaches to Visual Motion Analysis
Country/TerritoryGermany
CityDagstuhl Castle
Period13/07/0818/07/08

Fingerprint

Dive into the research topics of 'Discrete-continuous optimization for optical flow estimation'. Together they form a unique fingerprint.

Cite this