Bootstrap algorithms for dynamic stereo vision

Summary form only given. The use of stereo vision is significantly simplified by using depth information computed from a narrow-baseline image pair to constrain the search for correspondence in wider-baseline images. Such images can be acquired by using a combination of motion and multiple cameras. This approach has been in terms of random field models and Bayesian estimation. Experimental results have demonstrated the success of the approach. Algorithms that are efficient, produce accurate depth maps, and impose fewer constraint on scene geometry than previous approaches to stereo have been obtained and demonstrated with images of a realistic, outdoor scene model. The algorithms were developed as part of a larger scenario in which small camera motions will be used to bootstrap stereo correspondence