Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GLC-SLAM: Gaussian Splatting SLAM with Efficient Loop Closure

Ziheng Xu, Qingfeng Li, Chen Chen, Xuefeng Liu, Jianwei Niu

TL;DR

GLC-SLAM is introduced, a Gaussian Splatting SLAM system that integrates global optimization of camera poses and scene models that achieves superior or competitive tracking and mapping performance compared to state-of-the-art dense RGB-D SLAM systems.

Abstract

3D Gaussian Splatting (3DGS) has gained significant attention for its application in dense Simultaneous Localization and Mapping (SLAM), enabling real-time rendering and high-fidelity mapping. However, existing 3DGS-based SLAM methods often suffer from accumulated tracking errors and map drift, particularly in large-scale environments. To address these issues, we introduce GLC-SLAM, a Gaussian Splatting SLAM system that integrates global optimization of camera poses and scene models. Our approach employs frame-to-model tracking and triggers hierarchical loop closure using a global-to-local strategy to minimize drift accumulation. By dividing the scene into 3D Gaussian submaps, we facilitate efficient map updates following loop corrections in large scenes. Additionally, our uncertainty-minimized keyframe selection strategy prioritizes keyframes observing more valuable 3D Gaussians to enhance submap optimization. Experimental results on various datasets demonstrate that GLC-SLAM achieves superior or competitive tracking and mapping performance compared to state-of-the-art dense RGB-D SLAM systems.

GLC-SLAM: Gaussian Splatting SLAM with Efficient Loop Closure

TL;DR

GLC-SLAM is introduced, a Gaussian Splatting SLAM system that integrates global optimization of camera poses and scene models that achieves superior or competitive tracking and mapping performance compared to state-of-the-art dense RGB-D SLAM systems.

Abstract

3D Gaussian Splatting (3DGS) has gained significant attention for its application in dense Simultaneous Localization and Mapping (SLAM), enabling real-time rendering and high-fidelity mapping. However, existing 3DGS-based SLAM methods often suffer from accumulated tracking errors and map drift, particularly in large-scale environments. To address these issues, we introduce GLC-SLAM, a Gaussian Splatting SLAM system that integrates global optimization of camera poses and scene models. Our approach employs frame-to-model tracking and triggers hierarchical loop closure using a global-to-local strategy to minimize drift accumulation. By dividing the scene into 3D Gaussian submaps, we facilitate efficient map updates following loop corrections in large scenes. Additionally, our uncertainty-minimized keyframe selection strategy prioritizes keyframes observing more valuable 3D Gaussians to enhance submap optimization. Experimental results on various datasets demonstrate that GLC-SLAM achieves superior or competitive tracking and mapping performance compared to state-of-the-art dense RGB-D SLAM systems.
Paper Structure (28 sections, 15 equations, 3 figures, 7 tables)

This paper contains 28 sections, 15 equations, 3 figures, 7 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Visual SLAM
  4. 3DGS-based SLAM
  5. PRELIMINARY
  6. Scene Representation
  7. Uncertainty Modeling
  8. SYSTEM
  9. Tracking
  10. Local Mapping
  11. Map Building
  12. Uncertainty-minimized Keyframe Selection
  13. Loss Function
  14. Loop Closing
  15. Loop Detection
  16. ...and 13 more sections

Figures (3)

  • Figure 1: Reconstruction results on ScanNet 30054. Our method effectively mitigates the severe map drift inherent in Gaussian-SLAM 9, while also providing superior scene geometry and detail compared to GO-SLAM 14.
  • Figure 2: System Overview. Our system consists of three processes: tracking, mapping and loop closing. The tracking process estimates and refines camera poses $\{\text{R, t}\}$ by minimizing the tracking loss. The scene is managed as Gaussian submaps while the local mapping process select keyframes with an uncertainty-minimized strategy to optimize the active submap. If a loop is detected, the loop closing process triggers loop closure online, followed by efficient map adjustment to correct accumulated error and mitigate map drift.
  • Figure 3: Mesh Evaluation on ScanNet 3. The red boxes show map drift or poor details.