FusionGS-SLAM: Multiple Sensors Fusion for Simultaneous Localization and Real-Time Photorealistic Mapping
Thanh-Danh Phan, Gon-Woo Kim
AI summary
Problem
Purely visual SLAM systems struggle with scale ambiguity and tracking instability in unbounded outdoor environments, while existing multi-sensor fusion methods for photorealistic mapping are either offline or too slow for real-time robotics applications.
Approach
The framework tightly fuses visual, LiDAR, and inertial data via factor graph optimization for robust odometry, then initializes and optimizes 3D Gaussian primitives using depth-invariant spherical projection and adaptive densification for real-time, high-fidelity rendering.
Key results
- Tightly-coupled visual-LiDAR-inertial localization framework for robust outdoor tracking
- Novel 3D Gaussian initialization via depth-invariant spherical projection fusing LiDAR geometry and visual features
- Adaptive keyframing and sliding-window optimization ensuring real-time photorealistic mapping
- Superior accuracy and rendering quality over state-of-the-art 3DGS SLAM on public and self-collected datasets
Why it matters
Provides a reliable, real-time mapping solution for autonomous robots and AR/VR systems operating in complex, unbounded outdoor environments where traditional SLAM fails.
Abstract
This work presents a FusionGS-SLAM, a robust framework for simultaneous localization and real-time photore- alistic mapping leveraging the power of sensor fusion techniques. To achieve this, the proposed method employs a tightly-coupled technique to effectively combine multiple factors from improved subsystems, thereby generating a robust odometry for the down- stream tasks. Moreover, a dense 3D Gaussian map is constructed by leveraging geometric information across sensor modalities, with real-time mapping strategies designed to enhance robustness and rendering quality in large-scale and challenging environments. Experimental evaluation of various challenging scenes, including the public and self-collected datasets, showcases the superior per- formance compared to the current state-of-the-art 3DGS SLAM.