Exploration of 3D Reconstruction Techniques for the UPOU FMDS’ Perma G.A.R.D.E.N.: Contributing to Its Digital Twin Development

Abstract

This project aims to explore emerging 3D reconstruction techniques applied to the UPOU FMDS’ Perma G.A.R.D.E.N., in contribution to its digital twin development. The objectives include providing a simplified overview of how the explored 3D reconstruction methods work, testing these techniques to the garden, and assessing their output models in terms of visual accuracy and usability. The project also addresses and acknowledges various issues, explores an optimal methodology, and includes prototyping and recommendations to provide insights for future developers and researchers working on the garden’s digital twin. Among the three explored techniques—Photogrammetry, Neural Radiance Fields (NeRF), and Gaussian Splatting—Gaussian Splatting was found to be the most suitable for the garden, based on visual accuracy and usability. The optimal methodology utilized the Nerfstudio framework with its “splatfacto” method, which is its Gaussian Splatting procedure. For prototyping, PlayCanvas was used with its support for editing and integrating Gaussian Splatting 3D models. Various issues were addressed and acknowledged during these phases. The insights from this project could benefit future developers and researchers involved in the garden’s digital twin project by demonstrating the differences between the 3D reconstruction methods and their application to the garden. The prototypes provided, which are available as open educational resources, could inspire the enhancement of the current digital twin of the garden, which is its 360-degree photo based Virtual Reality (VR) tour, by integrating walkability through a 3D model, thus increasing its immersive experience.