Abstract
This dissertation describes techniques for efficiently matching corresponding areas of a stereo pair of images. Measures of match which are suitable for this purpose are discussed, as are methods for pruning the search for a match. The mathematics necessary to convert a set of matchings into a workable camera model are given, along with calculations which use this model and a pair of image points to locate the corresponding scene point. Methods are included to detect some types of unmatchable target areas in the original data and for detecting when a supposed match is invalid. Region growing techniques are di9 for extend matching areas into regions of constant parallax and for delimiting uniform regions in an image. Also, two algorithms are presented to show some of the ways in which these techniques can be combined to perform useful tasks in the processing of stereo images.