GSORB-SLAM: Gaussian Splatting SLAM benefits from ORB features and Transmittance information

Wancai Zheng, Xinyi Yu*, Jintao Rong, Linlin Ou, Yan Wei, Libo Zhou,
Zhejiang University of Technology *denotes equal contribution
Paper Code

Overview

In this paper, we propose GSORB-SLAM, a dense SLAM framework that integrates 3DGS with ORB features through a tightly coupled optimization pipeline. To mitigate the effects of noise and artifacts, we propose a novel geometric representation and optimization method for tracking, which significantly enhances localization accuracy and robustness. For high-fidelity mapping, we propose an adaptive Gaussian expansion and regularization method that facilitates compact yet expressive scene modeling while suppressing redundant primitives. Furthermore, we design a hybrid graph-based viewpoint selection mechanism that effectively mitigates overfitting and accelerates convergence. Extensive evaluations across various datasets demonstrate that our system achieves state-of-the-art performance in both tracking precision—improving RMSE by 16.2% compared to ORB-SLAM2 baselines—and reconstruction quality—improving PSNR by 3.93 dB compared to 3DGS-SLAM baselines.

Replica Room Sequence Renderings

Room0

Room1

Room2

Replica Office Sequence Renderings

Office0

Office2

Office4

Render Comparisons

Ours
Photo-SLAM
Ours
Photo-SLAM
Ours
SplaTAM
Ours
SplaTAM
Ours
Photo-SLAM
Ours
Photo-SLAM
Ours
SplaTAM
Ours
SplaTAM

Myself dataset

Training (Speed Up)

Rendering

Rendering result

The results on the Replica and TUM RGB-D datasets show that our approach outperforms existing dense neural SLAM and 3DGS-SLAM methods on the commonly reported rendering metrics.

Tracking Result

The best results will be highlighted in red, while the second-best will be highlighted in blue. The result is the average of six times data.

Surface depth VS. Depth

Surface depth not only improves tracking real-time performance but also enhances localization accuracy.

Acknowledge

[1]GS-SLAM: Dense Visual SLAM with 3D Gaussian Splatting
[2]SplaTAM: Splat, Track & Map 3D Gaussians for Dense RGB-D SLAM
[3]Orb-slam2: An open-source slam system for monocular, stereo, and rgb-d cameras

BibTeX

@misc{zheng2024gsorbslamgaussiansplattingslam,
            title={GSORB-SLAM: Gaussian Splatting SLAM benefits from ORB features and Transmittance information}, 
            author={Wancai Zheng and Xinyi Yu and Jintao Rong and Linlin Ou and Yan Wei and Libo Zhou},
            year={2024},
            eprint={2410.11356},
            archivePrefix={arXiv},
            primaryClass={cs.RO},
            url={https://arxiv.org/abs/2410.11356}, 
}
}