Stereo matching of corresponding images is a fundamental process in photogrammetry. In the last three decades many attempts have been made to automate it. These attempts yielded limited success. However, the majority of the problems experienced in the early matching systems still exist today. This is a consequence of using the area-based image correlation method for solving the problem. In this method matches were found by correlating two images (windows), from the left and right photographs, based on raw gray level values of the pixels. Recently, the interest operator method has been implemented to solve the stereo matching problem. In this method distinct pixels are evaluated rather than areas. The major drawback of these methods is that they depend heavily on geometrical restrictions and requires the human operator to approximately match the images.; In this research edges have been used to solve automatically the relative orientation problem. Edges of objects are primitives which are commonly used in computer vision and also evidently used by the human operator to solve the stereo matching problem. A coarse to fine strategy has been used in order to achieve the objective. This is similar to the procedure used by the human operator who first matches some pronounced objects in a large field of view and only later performs exact matching by concentrating on a small area.; Results show that the initial alignment of the two photographs has been improved from an average Y parallax of 123 pixels to less than 6. This is a remarkable improvement given the fact that it does not require an operator intervention. This research demonstrates the advantages in solving the stereo matching problem by emulating processes of the human operator. This is in contrast to past approaches where solutions were based on purely mathematical basis which yielded limited results under highly restricted circumstances.
展开▼
