A stereo matching algorithm for lunar rover

In order to satisfy the requirement of navigation and localization of the lunar rover, a new quasi-dense matching algorithm based on SIFT features and region growing is presented in the paper. Firstly, the original images are preprocessed by Gauss filter and CLAHE method to weaken the effect of noise and contrast, then the invariant SIFT feature points in each image are extracted and reliably matched with common constraints in stereo matching, such as epipolar constraint, uniqueness constraint and continuity constraint. Finally, matched SIFT features with high accuracy are taken as the seeds from which corresponding relations propagate towards other regions of the image by best first strategy. Experimental results indicate that the algorithm performs well when used in the matching of stereo image pair of simulated lunar surface terrain environment and produces a dense disparity map with rather high accuracy.