ORB-SLAM3

A classic open-source visual SLAM system from the University of Zaragoza, supporting monocular, stereo, RGB-D, and IMU input.

ORB-SLAM3 is an open-source SLAM (simultaneous localization and mapping) library from Juan D. Tardós and José M. M. Montiel's team at the University of Zaragoza in Spain, with the paper published in IEEE Transactions on Robotics in 2021. It uses ORB feature points (a corner feature that's fast to compute) for tracking and mapping, supports monocular, stereo, and RGB-D cameras with either pinhole or fisheye lenses, and can be tightly coupled with an IMU. Its multi-map system, called Atlas, lets it start a new map whenever tracking is lost and automatically merges maps when the robot returns to a previously mapped area. The paper reports an average error of 3.6 cm for stereo-plus-IMU on the EuRoC drone dataset. The code is open-sourced under GPLv3 and is often used as a visual SLAM baseline and teaching tool; it builds a sparse feature-point map that isn't directly usable for obstacle avoidance, and like other feature-based methods it can still lose tracking in low-texture scenes or under drastic lighting changes.

ExampleRunning ORB-SLAM3's stereo-plus-IMU or RGB-D mode on a mobile robot equipped with an IMU-carrying depth camera (such as a RealSense D435i) gives a real-time camera trajectory, used as the pose source for navigation or data collection.

Also called

ORB_SLAM3, ORB-SLAM Family

Related

Simultaneous Localization and Mapping · Visual SLAM · Visual-Inertial Odometry · Feature Points · Loop Closure Detection · Bundle Adjustment

Sources

ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual-Inertial and Multi-Map SLAM (IEEE T-RO 2021)
GitHub: UZ-SLAMLab/ORB_SLAM3

As of

2021-12