A stereo matching algorithm for lunar rover navigation

Abstract

The goal of designing stereo matching algorithms for lunar rover navigation on natural terrain is to produce disparity maps with both denseness and accuracy. We present a multi-stage matching algorithm in which matching results of points with larger gradient in earlier stages are used to assist in matching of points with smaller gradient in later stages for the purpose of improving matching accuracy in regions lacking texture. Two new constraints together with other constraints and statistical means are employed to achieve this matching strategy. The effectiveness of the algorithm is verified by experiments with real images of natural terrain in various conditions