3D Reconstruction of Underwater Shipwrecks: 3D Gaussian Splatting and Structure from Motion for the Melania shipwreck
Keywords: Underwater, Photogrammetry, 3D-Gaussian Splatting, Melania shipwreck, Artificial Intelligence, Agisoft Metashape
Abstract. Underwater cultural heritage documentation presents significant challenges due to optical distortions inherent to aquatic environments. This study presents a dual approach for 3D reconstruction of underwater environments using two image-based 3D reconstruction techniques: Structure from Motion (SfM) and 3D Gaussian Splatting (3DGS). The goal is to evaluate the respective efficacy of each in terms of geometric accuracy and texture fidelity. The comparison is performed on the pilot case study of the Melania shipwreck, off the coast of Vada in Livorno, Italy. The results demonstrate that, while SfM proves more effective in structural reconstruction of submerged remains, 3DGS might enhance texture photo-realism and more effectively mitigate underwater lighting distortions. Conducted within the framework of the Sub-IA project, this research contributes to the development of integrated methodologies for the high-fidelity and realistic, image-based digital documentation of underwater heritage.