publications

2024

1. CG-SLAM

   

   CG-SLAM: Efficient Dense RGB-D SLAM in a Consistent Uncertainty-aware 3D Gaussian Field

   Jiarui Hu^*, Xianhao Chen^*, Boyin Feng, Guanglin Li, Liangjing Yang, Hujun Bao, Guofeng Zhang, and Zhaopeng Cui^†

   In Proc. of the European Conf. on Computer Vision (ECCV), Jul 2024

   Abs PDF Code Website

   This paper presents an efficient dense RGB-D SLAM system, i.e., CG-SLAM, based on a novel uncertainty-aware 3D Gaussian field with high consistency and geometric stability. Through an in-depth analysis of Gaussian Splatting, we propose several techniques to construct a consistent and stable 3D Gaussian field suitable for tracking and mapping. Additionally, a novel depth uncertainty model is proposed to ensure the selection of valuable Gaussian primitives during optimization, thereby improving tracking efficiency and accuracy. Experiments on various datasets demonstrate that CG-SLAM achieves superior tracking and mapping performance with a notable tracking speed of up to 15 Hz.

2. SamSLAM

   

   SamSLAM: A Visual SLAM based on Segment anything model for Dynamic Environment

   Xianhao Chen, Tengyue Wang, Haonan Mai, and Liangjing Yang^†

   In The International Conference on Robotics, Control and Automation (ICRCA), Jan 2024

   Abs

   As autonomous robotic technology advances, Simultaneous Localization and Mapping (SLAM) plays a crucial role in improving the navigation capabilities of intelligent robots, thanks to its precision in positional data and detailed environmental perception. Traditional SLAM methods, however, assume a static environment—a limitation that diminishes their effectiveness in dynamic real-world settings, which often feature unpredictable changes and moving objects. To address these challenges, this paper introduces a new approach called SamSLAM, which incorporates the Segment Anything Model (SAM). SamSLAM enhances SLAM’s adaptability by integrating SAM’s precise object boundary segmentation with spatial geometric data from image sequences. This integration enables the detection of dynamic elements within the environment through changes in geometry between consecutive images, identified using techniques such as Random Sample Consensus (RANSAC) and epipolar geometry. The effectiveness of SamSLAM is validated through tests conducted on two well-known RGB-D datasets within the SLAM community, as well as in real-world environments. The results show that SamSLAM significantly outperforms traditional methods in dynamic settings.