Multi-Feature Nonlinear Optimization Motion Estimation Based on RGB-D and Inertial Fusion

Area of Science:

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

Background:

• Traditional visual SLAM methods often struggle in low-textured environments due to reliance solely on point features.
• Geometric information from line segments can enhance environmental feature representation for SLAM.

Purpose of the Study:

• To develop a high-precision RGB-D visual-inertial SLAM system for indoor robot motion estimation.
• To improve SLAM robustness and accuracy in low-textured scenes by incorporating line features alongside point features.

Main Methods:

• A tightly coupled RGB-D visual-inertial SLAM system integrating point and line features.
• Fast initialization using RGB-D camera data for real-time performance.
• A novel backend nonlinear optimization framework minimizing IMU residuals and re-projection errors of points and lines within a sliding window.

Main Results:

• The proposed system demonstrates enhanced robustness in low-textured environments.
• Experimental results on public datasets show superior accuracy in trajectory and pose estimation compared to state-of-the-art visual SLAM systems.

Conclusions:

• Integrating both point and line features significantly improves the performance of RGB-D visual-inertial SLAM.
• The system offers a robust and accurate solution for indoor robot localization and mapping, particularly in challenging environments.