Abstract:A feature-constrained stereo matching algorithm for lunar rover navigation is presented based on the analysis of the stereo vision system and working environments of lunar rover. In feature-matching phase, edge points are extracted with wavelet transform and are used as the primitives for matching. Then three criterions are utilized in turn to select the correct matching points with the pyramidal searching strategy. As a result, the algorithm finds corresponding points successfully for large numbers of edge points. Area-matching is accomplished under the constraint of edge-matching results, and the correlation is selected as the criterion. Experimental results with real images of natural terrain indicate that the algorithm provides dense disparity maps with fairly high accuracy.