ORB-SLAM2: an Open-Source SLAM System for Monocular, Stereo and RGB-D Cameras

Raul Mur-Artal,Juan D. Tardos

DOI: https://doi.org/10.1109/TRO.2017.2705103

2017-06-19

Abstract:We present ORB-SLAM2 a complete SLAM system for monocular, stereo and RGB-D cameras, including map reuse, loop closing and relocalization capabilities. The system works in real-time on standard CPUs in a wide variety of environments from small hand-held indoors sequences, to drones flying in industrial environments and cars driving around a city. Our back-end based on bundle adjustment with monocular and stereo observations allows for accurate trajectory estimation with metric scale. Our system includes a lightweight localization mode that leverages visual odometry tracks for unmapped regions and matches to map points that allow for zero-drift localization. The evaluation on 29 popular public sequences shows that our method achieves state-of-the-art accuracy, being in most cases the most accurate SLAM solution. We publish the source code, not only for the benefit of the SLAM community, but with the aim of being an out-of-the-box SLAM solution for researchers in other fields.

Robotics,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper aims to solve several key problems in Simultaneous Localization and Mapping (SLAM), especially for SLAM systems of monocular, binocular and RGB - D cameras. Specifically, the paper proposes a complete SLAM system named ORB - SLAM2, which has the following features:

1. **Multi - sensor support**: ORB - SLAM2 supports monocular, binocular and RGB - D cameras and can run on different sensor types, providing a flexible solution.
2. **Map reuse**: The system can reuse previously constructed maps, which is very important for long - running robotic tasks.
3. **Loop closure detection**: The system has the ability of loop closure detection. It can detect when the robot returns to a known area and correct the accumulated localization errors.
4. **Relocalization**: When tracking fails (for example, due to occlusion or violent movement), the system can relocalize the camera.
5. **Real - time performance**: ORB - SLAM2 can achieve real - time operation on a standard CPU and is suitable for a variety of environments, from indoor handheld devices to drones flying in industrial environments, and cars driving in cities.
6. **High - precision trajectory estimation**: By using the Bundle Adjustment (BA) technique, the system can achieve high - precision trajectory estimation, especially in binocular and RGB - D modes.
7. **Lightweight localization mode**: The system includes a lightweight localization mode. It uses the visual odometry trajectory to localize in unmapped areas and achieves zero - drift localization by matching map points.

In general, the main goal of this paper is to develop an efficient, accurate and robust SLAM system that can achieve real - time localization and mapping in multiple sensors and environments.