Initial disparity estimation using sparse matching for wide-baseline dense stereo

We propose a triangulation based initialization met hod for dense disparity estimation from uncalibrated wide-b aseline image pairs using sparse correspondences. The metho d includes: (a) sparse correspondence retrieval, (b) Delaunay triangulation and homography estimation, and (c) ob taining a dense initial disparity map to initialize dense ste reo algorithms. Comparison with existing methods demons trates improvement in results with lower computational com plexity.