Date of Award
CC BY-NC-ND 4.0
Stereo vision is aimed at recovering 3D structure from two images taken with cameras positioned at different viewpoints. To obtain the depth of a scene, we need to establish the correspondence of pixels or features between the two stereo images. This process is called matching. Dense matching uncalibrated images is a difficult task; it requires the matching of each and every pixel between images, while no knowledge of the camera parameters is available. Most existing methods for dense matching uncalibrated images are impractical and time consuming. In order to develop a fast, accurate and practical method, we attempted to use image interest points to provide a disparity estimate. Then we proposed a fast dense matching algorithm which integrates edge features of the image. The matching was carried out separately for edge areas and non-edge areas. In order to match the non-edge areas of the image, matching constraints were combined to restrict the search region. This approach effectively reduces the computational time and improves the matching quality. Our hybrid method has been tested on several indoor and outdoor scenes and the results demonstrate its capability. Paper copy at Leddy Library: Theses & Major Papers - Basement, West Bldg. / Call Number: Thesis2001 .J56. Source: Masters Abstracts International, Volume: 40-03, page: 0723. Adviser: B. Boufama. Thesis (M.Sc.)--University of Windsor (Canada), 2001.
Jin, Hongxuan., "Dense matching of uncalibrated images for stereo vision." (2001). Electronic Theses and Dissertations. 1632.