Robust RGB-D SLAM Using Point and Line Features for Low Textured Scene

Area of Science:

• Computer Vision
• Robotics
• 3D Reconstruction

Background:

• RGB-D cameras offer cost-effective 3D data collection.
• Point features are standard for RGB-D Simultaneous Localization and Mapping (SLAM) but fail in low-textured scenes.
• Existing RGB-D SLAM systems struggle with robustness due to insufficient reliable point features.

Purpose of the Study:

• To develop a robust RGB-D SLAM system that enhances accuracy and reliability in challenging environments.
• To address the limitations of point-feature-based SLAM in scenes with sparse features.
• To improve 3D reconstruction quality by integrating complementary geometric constraints.

Main Methods:

• Proposed a novel RGB-D SLAM system fusing both point and line features for robust geometric constraints.
• Developed a comprehensive cost function combining 2D/3D line reprojection error and point reprojection error.
• Implemented a robust pose solver using Gauss-Newton optimization and Chi-Square testing for feature match filtering.
• Utilized a sliding-window framework for keyframe pose and feature optimization to correct camera pose drift.

Main Results:

• The proposed system demonstrates comparable or superior performance to state-of-the-art methods.
• Achieved high accuracy and robustness in 3D reconstruction, even in low-textured environments.
• Successfully fused point and line features to overcome the limitations of traditional point-based SLAM.

Conclusions:

• The fusion of point and line features significantly enhances the robustness of RGB-D SLAM.
• The proposed system provides a reliable solution for 3D reconstruction in challenging real-world scenarios.
• This approach offers a promising direction for improving the performance and applicability of RGB-D SLAM.