PGMF-VINS: Perpendicular-Based 3D Gaussian-Uniform Mixture Filter.

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Simultaneous Localization and Mapping (SLAM)

Background:

• Visual-Inertial SLAM (VI-SLAM) is crucial for robotics, autonomous driving, AR, and VR.
• Existing VI-SLAM research often prioritizes localization over mapping robustness.
• Map point quality and stability are critical for reliable VI-SLAM performance.

Purpose of the Study:

• To enhance the robustness and quality of the mapping task in VI-SLAM.
• To introduce a novel map-point convergence strategy for improved map point estimation.
• To ensure state-of-the-art localization accuracy while improving map quality.

Main Methods:

• Proposed a map-point convergence strategy estimating position, uncertainty, and stability (SoM).
• Developed the Perpendicular-based 3D Gaussian-uniform Mixture Filter (PGMF).
• Utilized perpendicular-based triangulation and a 3D Gaussian-uniform mixture filter within the strategy.

Main Results:

• The proposed method significantly improves overall map quality.
• Achieved state-of-the-art localization accuracy.
• Demonstrated an average increase of 0.1471 in Ratio of Valid Map points (RVM) compared to VINS-mono.
• Showcased a 68.8% reduction in variance for map points.

Conclusions:

• The developed map-point convergence strategy effectively enhances VI-SLAM mapping robustness.
• PGMF offers a significant improvement in map point validity and stability.
• The method provides a robust solution for applications requiring high-quality maps and accurate localization.