Epiflow quadruplet matching:Enforcing epipolar geometry for spatio-temporal stereo correspondences

Hongsheng Zhang, Shahriar Negahdaripour

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Scopus citations

Abstract

Given a pair of left-right views of a calibrated stereo rig at two different poses, we have previously proposed two robust and efficient feature matching methods. These so-called quadruplet feature matching schemes use RANSAC for outlier rejection in establishing spatio-temporal matches in four views, and computing the motion of the stereo rig from one pose to the next [12]. However, a significant number of valid quadruplet matches may be missed in part due to the inherent difficulties of temporal feature matching/tracking, and in part because of the conservative nature of the applied epipolar mutual-enforcing quadruplet matching technique in exchange for increased robustness. Having determined some initial quadruplet matches, an optimum motion is then determined by applying all the constraints embedded in the geometry of the two stereo pairs [11]. In this paper, we exploit the estimated motion to identify a larger number of valid quadruplet features under the so-called Epiflow Quadruplet Matching framework. Here, local matching dissimilarities of the left and right views individually and in the left-right views at the second stereo pose are expressed in terms of an energy function of discrepancies between optimal and estimated feature locations. The deviations of the feature locations in the second stereo views are defined as 4-D vectors that are constrained by quadratic constraints arising from the stereo epipolar geometry. Experiments with two sets of real data are presented to show the performance from the proposed energy minimization framework. Subsequent improvement in motion estimation, accommodated by utilizing more correct quadruplets matches, leads to a a natural framework for iterative motion estimation and feature matching.

Original languageEnglish (US)
Title of host publicationProceedings - Seventh IEEE Workshop on Applications of Computer Vision, WACV 2005
PublisherIEEE Computer Society
Pages481-486
Number of pages6
ISBN (Print)0769522718, 9780769522715
DOIs
StatePublished - Jan 1 2005
Event7th IEEE Workshop on Applications of Computer Vision, WACV 2005 - Breckenridge, CO, United States
Duration: Jan 5 2005Jan 7 2005

Publication series

NameProceedings - Seventh IEEE Workshop on Applications of Computer Vision, WACV 2005

Other

Other7th IEEE Workshop on Applications of Computer Vision, WACV 2005
CountryUnited States
CityBreckenridge, CO
Period1/5/051/7/05

    Fingerprint

ASJC Scopus subject areas

  • Computer Vision and Pattern Recognition
  • Software

Cite this

Zhang, H., & Negahdaripour, S. (2005). Epiflow quadruplet matching:Enforcing epipolar geometry for spatio-temporal stereo correspondences. In Proceedings - Seventh IEEE Workshop on Applications of Computer Vision, WACV 2005 (pp. 481-486). [4129521] (Proceedings - Seventh IEEE Workshop on Applications of Computer Vision, WACV 2005). IEEE Computer Society. https://doi.org/10.1109/ACVMOT.2005.47