ICP-3DGS: SfM-free 3D Gaussian Splatting for Large-scale Unbounded Scenes

TL;DR

This paper introduces ICP-3DGS, a novel method that combines ICP with scene densification to enable SfM-free 3D Gaussian Splatting for large-scale scenes, improving camera pose estimation and view synthesis.

Contribution

It presents a new approach integrating ICP and voxel-based densification to reconstruct large-scale scenes without relying on SfM or preprocessed poses.

Findings

Outperforms existing methods in camera pose estimation.

Achieves superior novel view synthesis in large-scale scenes.

Effective in both indoor and outdoor environments.

Abstract

In recent years, neural rendering methods such as NeRFs and 3D Gaussian Splatting (3DGS) have made significant progress in scene reconstruction and novel view synthesis. However, they heavily rely on preprocessed camera poses and 3D structural priors from structure-from-motion (SfM), which are challenging to obtain in outdoor scenarios. To address this challenge, we propose to incorporate Iterative Closest Point (ICP) with optimization-based refinement to achieve accurate camera pose estimation under large camera movements. Additionally, we introduce a voxel-based scene densification approach to guide the reconstruction in large-scale scenes. Experiments demonstrate that our approach ICP-3DGS outperforms existing methods in both camera pose estimation and novel view synthesis across indoor and outdoor scenes of various scales. Source code is available at https://github.com/Chenhao-Z/ICP-3DGS.