Open Collections

Interpreting severe occlusions in region-based stereo vision

University of British Columbia

This thesis presents a theory of correspondence matching in region-based stereo vision under severe occlusions. The general purpose of stereo vision is to perform distance measurements based on correspondence between slightly offset cameras. Region-based stereo vision performs matching in terms of regions in each image. Regions are generally obtained using a segmentation algorithm and matching is done by comparing the properties of the regions. The goal of this thesis is to analyze and explicitly model the effects of occlusions on the region-based stereo matching process. In order to explicitly model only occlusions it is assumed that segmentation can be done correctly and that properties of the regions can be altered only by occlusions. In order to systematically solve the problem, occlusions are separated into categories depending on the effect they had on the properties of the regions. Each occlusion type is modeled in terms of a number of constraints. A constraint satisfaction process is introduced in order to identify the existence and kind of occlusion. The matching process allows establishing one-to-one, one-to-none, one-to-multiple, as well as, multiple-to-multiple region correspondences. A software system has been developed in order to test the correctness of the presented theory. The system allows the user to create a virtual scene with severe occlusions. Stereo images of the scene are then generated and passed onto the stereo algorithm. The system is able to correctly identify all occlusion types and reliably establish correct correspondences between regions. Further, the system was able to produce explanations for the match or lack of a match between regions. The system has also been successfully tested using real stereo images.

Thesis/Dissertation

application/pdf

2009-04-28

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/7665

Computer Science

University of British Columbia

UBCV

DSpace