Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

FGS-SLAM: Fourier-based Gaussian Splatting for Real-time SLAM with Sparse and Dense Map Fusion

Created by
  • Haebom

Author

Yansong Xu, Junlin Li, Wei Zhang, Siyu Chen, Shengyong Zhang, Yuquan Leng, Weijia Zhou

Outline

3D Gaussian splatting-based SLAM technology enables real-time position estimation and high-quality map generation. However, it requires an iterative convergence process due to the uncertainty of Gaussian position and initialization parameters, and there is a problem that the Gaussian representation may be excessive or insufficient. In this paper, we propose a novel adaptive density method based on Fourier frequency-domain analysis to set Gaussian priors and achieve fast convergence. In addition, we present a method to independently and integratedly construct a sparse map that supports efficient tracking using the Generalized Iterative Closest Point (GICP) and a dense map that generates high-quality visual representations. This is the first SLAM system that achieves high-quality Gaussian mapping in real time by utilizing frequency-domain analysis. We achieve an average frame rate of 36 FPS on Replica and TUM RGB-D datasets, and show competitive accuracies in both position estimation and mapping.

Takeaways, Limitations

•
Takeaways:
◦
First proposal of a Gaussian splatting-based SLAM system utilizing Fourier frequency domain analysis.
◦
Effectively setting Gaussian prior information and achieving fast convergence through adaptive density method.
◦
Achieving both efficiency and high-quality visualization through integrated configuration of sparse maps and dense maps.
◦
Demonstrating real-time processing and competitive accuracy on Replica and TUM RGB-D datasets.
•
Limitations:
◦
Further evaluation of the generalization performance of the proposed method on various environments and datasets is needed.
◦
Further validation of robustness in real-world environments is needed.
◦
A more comprehensive comparative analysis with other state-of-the-art SLAM techniques is needed.
👍
Made with Slashpage