3D Reconstruction of Transparent and Reflective Surfaces, Through the Use of SfM Processes Supported by 3D Gaussian-Splatting and 2D Gaussian-Splatting

This study evaluates three advanced approaches for 3D reconstruction of challenging transparent surfaces: conventional photogrammetry, enhanced 3D Gaussian Splatting, and novel 2D Gaussian Splatting (2DGS). Through a detailed case study of a glass artifact, the research demonstrates 2DGS's superior performance in geometric reconstruction and multi-view consistency, leveraging its innovative planar representation to outperform alternative methods in capturing fine surface details and complex internal structures. While the enhanced 3D approach shows advantages in visual rendering quality, it exhibits surface artifacts, and traditional photogrammetry proves inadequate for complete reconstruction. The comparative analysis highlights 2DGS's balanced capabilities in structural accuracy and perceptual quality, albeit with higher computational demands. These findings establish 2DGS as a significant advancement for cultural heritage documentation, particularly for transparent and reflective objects requiring precise digital preservation. The study identifies key directions for future development, including performance optimization and accessibility improvements, to facilitate broader adoption in heritage conservation and remote sensing applications where accurate 3D documentation is essential.