Digital regeneration of historical buildings using UAV photogrammetry and 3D Gaussian splatting for metaverse environments

The digital preservation of cultural heritage faces a trilemma balancing high-precision modeling costs, extensive production cycles, and the urgent need for rapid public accessibility. While Unmanned Aerial Vehicle photogrammetry offers scalable data acquisition, standard 3D Gaussian Splatting (3DGS) struggles with geometric occlusions and non-Lambertian materials (i.e., surfaces with view-dependent reflections, such as glazed tiles) inherent in East Asian timber-and-stone architecture. We propose a Cloud-to-Local hybrid 3DGS framework optimized for architectural heritage regeneration. A heuristic opacity modulation strategy deployed within the Unreal Engine 5 Niagara GPU buffer dynamically culls multi-view-deficient “floater” artifacts without altering baseline photogrammetric data. Furthermore, anisotropic covariance optimization and Physically-Based Rendering calibration preserve the morphological integrity of weathered masonry and woodwork. Evaluations demonstrate visualization-grade relative geometric consistency and real-time rendering efficiency over 110 FPS. By integrating the digital twin into a serious game, this research pioneers a Constructive Authenticity paradigm, transforming static reality capture into an experiential platform.