Large Occlusion Stereo

  • Authors:
  • Aaron F. Bobick;Stephen S. Intille

  • Affiliations:
  • College of Computing, Georgia Institute of Technology, Atlanta, GA 30332. afb@cc.gatech.edu;MIT Media Laboratory, Cambridge, MA 02139. intille@media.mit.edu

  • Venue:
  • International Journal of Computer Vision
  • Year:
  • 1999

Quantified Score

Hi-index 0.00

Visualization

Abstract

A method for solving the stereo matching problem inthe presence of large occlusion is presented. A data structure—thedisparity space image—is defined to facilitatethe description of the effects of occlusion on the stereo matchingprocess and in particular on dynamic programming (DP) solutions thatfind matches and occlusions simultaneously. We significantly improveupon existing DP stereo matching methods by showing that while somecost must be assigned to unmatched pixels, sensitivity toocclusion-cost and algorithmic complexity can be significantly reducedwhen highly-reliable matches, or ground control points, areincorporated into the matching process. The use of ground controlpoints eliminates both the need for biasing the process towards asmooth solution and the task of selecting critical prior probabilitiesdescribing image formation. Finally, we describe how the detection ofintensity edges can be used to bias the recovered solution such thatocclusion boundaries will tend to be proposed along such edges,reflecting the observation that occlusion boundaries usually causeintensity discontinuities.