Binocular 3D reconstruction of local ancient architectural structures enhanced by geometric priors

Abstract

Traditional stereo matching algorithms suffer from low matching accuracy and severe disparity map holes when reconstructing local ancient architectural structures, especially within widespread weak-texture regions. This paper proposes a binocular 3D reconstruction method for local ancient architectural structures enhanced by geometric priors. Firstly, a multi-scale fusion strategy is embedded into the Semi-Global Block Matching (SGBM) algorithm to generate an initial disparity map, which balances the smoothing effect for weak-texture regions and the preservation of edge details. Secondly, the progressive probabilistic Hough transform is utilized to extract horizontal and vertical structural lines of ancient buildings. Combined with geometric regularization priors, our method constrains disparity continuity and reduces mismatches at depth discontinuities. Finally, taking advantage of the inherent left-right symmetrical layout of ancient architecture, we adopt symmetry priors to fill missing disparity values in weak-texture areas. Experimental results demonstrate that the proposed method effectively increases the proportion of valid pixels and greatly improves the overall integrity of 3D reconstruction for local ancient architectural structures.