Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DG-SLAM: Robust Dynamic Gaussian Splatting SLAM with Hybrid Pose Optimization

Yueming Xu, Haochen Jiang, Zhongyang Xiao, Jianfeng Feng, Li Zhang

TL;DR

DG-SLAM is proposed, the first robust dynamic visual SLAM system grounded in 3D Gaussians, which provides precise camera pose estimation alongside high-fidelity reconstructions and delivers state-of-the-art performance in camera pose estimation, map reconstruction, and novel-view synthesis in dynamic scenes.

Abstract

Achieving robust and precise pose estimation in dynamic scenes is a significant research challenge in Visual Simultaneous Localization and Mapping (SLAM). Recent advancements integrating Gaussian Splatting into SLAM systems have proven effective in creating high-quality renderings using explicit 3D Gaussian models, significantly improving environmental reconstruction fidelity. However, these approaches depend on a static environment assumption and face challenges in dynamic environments due to inconsistent observations of geometry and photometry. To address this problem, we propose DG-SLAM, the first robust dynamic visual SLAM system grounded in 3D Gaussians, which provides precise camera pose estimation alongside high-fidelity reconstructions. Specifically, we propose effective strategies, including motion mask generation, adaptive Gaussian point management, and a hybrid camera tracking algorithm to improve the accuracy and robustness of pose estimation. Extensive experiments demonstrate that DG-SLAM delivers state-of-the-art performance in camera pose estimation, map reconstruction, and novel-view synthesis in dynamic scenes, outperforming existing methods meanwhile preserving real-time rendering ability.

DG-SLAM: Robust Dynamic Gaussian Splatting SLAM with Hybrid Pose Optimization

TL;DR

DG-SLAM is proposed, the first robust dynamic visual SLAM system grounded in 3D Gaussians, which provides precise camera pose estimation alongside high-fidelity reconstructions and delivers state-of-the-art performance in camera pose estimation, map reconstruction, and novel-view synthesis in dynamic scenes.

Abstract

Achieving robust and precise pose estimation in dynamic scenes is a significant research challenge in Visual Simultaneous Localization and Mapping (SLAM). Recent advancements integrating Gaussian Splatting into SLAM systems have proven effective in creating high-quality renderings using explicit 3D Gaussian models, significantly improving environmental reconstruction fidelity. However, these approaches depend on a static environment assumption and face challenges in dynamic environments due to inconsistent observations of geometry and photometry. To address this problem, we propose DG-SLAM, the first robust dynamic visual SLAM system grounded in 3D Gaussians, which provides precise camera pose estimation alongside high-fidelity reconstructions. Specifically, we propose effective strategies, including motion mask generation, adaptive Gaussian point management, and a hybrid camera tracking algorithm to improve the accuracy and robustness of pose estimation. Extensive experiments demonstrate that DG-SLAM delivers state-of-the-art performance in camera pose estimation, map reconstruction, and novel-view synthesis in dynamic scenes, outperforming existing methods meanwhile preserving real-time rendering ability.

Paper Structure

This paper contains 14 sections, 13 equations, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Approach
  4. 3D Gaussian map representation
  5. Motion mask generation
  6. Coarse-to-fine camera tracking.
  7. SLAM system
  8. Experiments
  9. Evaluation of generating motion mask
  10. Evaluation of mapping performance
  11. Evaluation of tracking performance
  12. Ablation study
  13. Time consumption analysis
  14. Conclusion

Figures (3)

  • Figure 1: Overview of DG-SLAM. Given a series of RGB-D frames, we reconstruct the static high-fidelity 3D Gaussian map and optimize the camera pose represented with lie algebra $\xi_i$.
  • Figure 2: Qualitative results of the motion mask generation. By fusing the semantic mask and depth warp mask, the final mask will be more precise.
  • Figure 3: Visual comparison of the rendering image on the TUM and BONN datasets. Our results are more complete and accurate without the dynamic object floaters.