FGO-SLAM++: Real-Time Geometry-Aware Gaussian SLAM With Continuous Opacity Field

Area of Science:

• Computer Vision
• Robotics
• 3D Reconstruction

Background:

• Existing Gaussian SLAM systems struggle with multi-modal input, often trading rendering quality for geometric accuracy.
• Efficient tracking and mapping across diverse sensor modalities remain a challenge in SLAM.

Purpose of the Study:

• To develop a real-time, geometry-aware Gaussian SLAM system supporting arbitrary video modalities.
• To overcome the limitations of existing systems by achieving high performance in tracking, rendering, and geometric accuracy simultaneously.

Main Methods:

• Explicit geometric feature extraction for scene structure and camera pose estimation.
• Gaussian-based ray tracing to construct and optimize a continuous opacity field.
• Marching tetrahedra for surface extraction, refined by a geometric constraint field, with global adjustment for loop closures.

Main Results:

• FGO-SLAM++ demonstrates superior performance in tracking accuracy, rendering quality, and geometric reconstruction.
• The system achieves real-time efficiency across various video modalities.
• Multi-view consistent map reconstruction is achieved through the continuous opacity field.

Conclusions:

• The proposed FGO-SLAM++ system successfully integrates geometry awareness and multi-modal support in real-time SLAM.
• It offers a significant advancement over existing Gaussian SLAM approaches by balancing rendering quality and geometric precision.
• This work paves the way for more versatile and robust SLAM applications.