Assume a 3D surface point that is projected onto its corresponding image points . The inverse process holds for triangulating from the corresponding point pair . We can in fact exploit the calibrated camera geometry and express the 3D point as a depth value along the known line of sight that extends from the camera projection center through the image correspondence . Triangulation computes the depth as the length of connecting the camera projection center and the locus of minimum distance between the corresponding lines of sight. The triangulation is computed for each image point and stored in a dense depth map associated with the viewpoint.
The depth for each reference image point is improved by the correspondence linking that delivers two lists of image correspondences relative to the reference, one linking down from and one linking up from . For each valid corresponding point pair we can triangulate a consistent depth estimate along with representing the depth uncertainty. Figure 7.17(left) visualizes the decreasing uncertainty interval during linking. While the disparity measurement resolution in the image is kept constant (at 1 pixel), the reprojected depth error decreases with the baseline.