SEMI-GLOBAL MERGING of DIGITAL SURFACE MODELS from MULTIPLE STEREOPAIRS

Abstract

The semi-global optimization algorithm, which approximates a global 2D smoothness constraint by combining several 1D constraints, has been widely used in the field of image dense matching for digital surface model (DSM) generation. However, due to occlusion, shadow and textureless area of the matching images, some inconsistency may exist in the overlapping areas of different DSMs. To address this problem, based on the DSMs generated by semi-global matching from multiple stereopairs, a novel semi-global merging algorithm is proposed to generate a reliable and consistent DSM in this paper. Two datasets, each covering 1 km2, are used to validate the proposed method. Experimental results show that the optimal DSM after merging can effectively eliminate the inconsistency and reduce redundancy in the overlapping areas.