A stereo matching algorithm and uncertainty based on homography

To avoid the feature points matching confliction under the epipolar constraint of stereo image pair, this paper presents a feature points correspondence method under the constraint of homography associated with epipolar geometry constraint. First the projections of the 3-D planes were detected within the image pair. Feature points belong to these regions can find its corresponding points uniquely without the ambiguity by the epipolar geometry alone. Delete these coplanar points and much of the correspondences were corrected. The matching points of those points set that didn´t belong to any 3-D plane were estimated approximately by its nearest correctly matched point pairs. Postior information was used to test the accuracy of both fitting the plane and estimating the corresponding points. A cost function was given during the matching process to compute the corresponding points´ uncertainty without reconstruction and the optimal rule of the two constraints was given. Experiments demonstrate that the algorithm is efficient and practical.