Experiments with edge-based stereo matching

Edge matching is the most critical step in feature-based stereo vision. Edges, as obtained by various operators on the image, are discontinuities in the light intensity function. In general, they correspond to physical boundaries in the object space. Here we are concerned with two methods of matching edges in digital stereopairs. In one method the edges are segmented and approximated by polygons. Corresponding edges are found by comparing vertices, leg orientation, and approximate location within the image. The second method described is based on the ψ - s representation (to be defined later) of the edge. Thus, the search space for matching is reduced by one dimension. -Authors