GL-VSLAM: A General Lightweight Visual SLAM Approach for RGB-D and Stereo Cameras.

Area of Science:

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

Background:

• Feature-based indirect SLAM offers robustness but suffers from slow feature extraction and descriptor computation.
• Direct SLAM methods can be faster but may lack robustness compared to feature-based approaches.

Purpose of the Study:

• To develop a general, lightweight visual SLAM (GL-VSLAM) system for RGB-D and stereo cameras.
• To improve the real-time performance of SLAM systems without sacrificing accuracy.

Main Methods:

• Utilizes sparse optical flow matching with a uniform motion model for keypoint correspondences, bypassing descriptor-based matching.
• Employs a two-stage coarse-to-fine pose estimation strategy: RANSAC PnP followed by reprojection error minimization.
• Extracts keypoints and descriptors from keyframes for backend optimization and loop closure detection.

Main Results:

• Achieves high real-time performance due to efficient sparse optical flow matching.
• Demonstrates comparable positioning accuracy to state-of-the-art methods.
• Shows up to twice the efficiency of ORB-SLAM2 in experimental evaluations.

Conclusions:

• GL-VSLAM offers a significant improvement in efficiency for visual SLAM systems.
• The proposed method provides a robust and accurate alternative for real-time applications using RGB-D and stereo cameras.