ORB-SLAM

ORB-SLAM is a family of visual SLAM systems that uses image features to estimate camera motion and build a sparse 3D landmark map. Tracking, local mapping, relocalization, and loop closure work together to maintain and refine that estimate.

ORB-SLAM3 supports monocular, stereo, and RGB-D cameras, as well as visual-inertial configurations. Its landmark map is distinct from the occupancy maps used for navigation and obstacle avoidance.

Topics for these notes include:

• ORB features, matching, and geometric verification.
• Camera pose estimation, keyframes, and bundle adjustment.
• Relocalization, loop closure, and map reuse.
• Camera calibration and transforming camera poses into robot-base poses.

Resources

• ORB-SLAM3 official implementation
• ORB-SLAM3 paper