Search space reduction in the edge based stereo matching by context of disparity gradient limit

Abstract

Traditionally, finding the corresponding points has considered to be the most difficult part of stereo matching algorithms. Usually, the correspondence for a feature point in the first image is obtained by searching in a predefined region of the second image, based on the epipolar line and the maximum disparity. The reduction of the search region can increase the performance of the matching process, in the context of the execution time and the accuracy. We proposed a new matching strategy to reduce the search space for the edge-based stereo correspondence algorithms. Considering the maximum of the disparity gradient in the real scene, we formulated the relation between the maximum search space in the second images with respect to the relative displacement of the continuous edges (as the feature points) in the successive scan lines of the first images. Then we developed some very fast stereo matching algorithms, based on the nonhorizontal edges as feature points, and the normalized cross correlation criteria (NCC) with different sizes of the matching block (as the similarity measures). We applied these new algorithms on the Renault stereo image and compared the result with those of a traditional matching algorithm (20 pixels search regions and NCC with size of 15×15). The speed up of these new algorithms is between 2.8 to 13.8 and the percentage of errors is between 0.5 to 5.4.