<title>Adaptive vergence for stereo matching</title>

Abstract

Using oblique imagery in the cartographic extraction of man-made features introduces several problems to the process of stereo matching that are not found with near-nadir imagery. Even after the images are brought into a collinear epipolar alignment, the ground-plane is not perpendicular to camera axis and the assumption that building roofs are similarly perpendicular to the camera axis is even less valid than with near-nadir imagery. In addition, walls are now commonly visible rather than being exceptional features. These problems require changes to the processing of the imagery and in the operation of the stereo matching. The obliquity of the ground-plane causes the most problem for stereo matchers. To accommodate the larger range of disparity, the stereo process must either increase its search window, or utilize an adaptive vergence mechanism to control the search window on a pixel- by-pixel basis. Increasing the search window increases both the search time and the chance for aliasing to occur. We show that adding an adaptive vergence mechanism to an existing stereo system provides a substantial improvement in its ability to process oblique imagery.