Open Collections

A unified recognition and stereo vision system for size assessment

University of British Columbia

This paper presents a unified recognition and stereo vision system which locates objects and determines their distances and sizes from stereo image pairs. Unlike other such systems, stereo information is not the input to the recognition stage. Instead, recognition is performed first and its output forms the input to stereo processing. This permits successful analysis of images captured in poor conditions (murky, specularities, poor camera alignment) and reduces time requirements. Model-based recognition is accomplished in two stages. The first stage seeks feature matches by comparing the absolute orientation, relative orientation and relative length of each image segment to those of the model segments in order to find chains of adjacent segments in the image which match those of the models. The absolute orientation constraint can be suppressed to permit identification of randomly-oriented objects. The second stage verifies candidate matches by comparing the relative locations of matched image features to the relative locations of the corresponding model features. The models them selves are generated semi-automatically from images of the desired objects. In addition to providing distance estimates, feature-based stereo information is used to disambiguate any multiple or questionable matches. Although the system and issues presented herein are quite general, the discussion and testing are primarily related to the motivating task of non-invasively assessing the size of sea-cage salmon.

Thesis/Dissertation

application/pdf

2009-03-06

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/2429/5682

Computer Science

University of British Columbia

UBCV

DSpace