ORB-SLAM2: An Open-Source SLAM System for Monocular, Stereo, and RGB-D Cameras
By Raul Mur-Artal, Juan D. Tardós. 2016
6,462 citations recorded by the source corpus.
We present ORB-SLAM2, a complete simultaneous localization and mapping (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 central processing units 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 with 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.
Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Aerospace Engineering
- Vladlen Koltun
- Daniel Cremers
- Andrew J. Davison
- Davide Scaramuzza
- Jia Deng
- Juan D. Tardós
- Luca Carlone
- Jitendra Malik
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Jitendra Malik
- Dacheng Tao
- Trevor Darrell
- Ming-Hsuan Yang
- Abhinav Kumar Gupta
- Xiaogang Wang
- Andrea Vedaldi
Citations in the corpus
- Continuous Scene Representations for Embodied AI (2022, background): “Classical mapping and localization work in the robotics literature [16, 19, 35] relies on low-level geometric features and does not encode semantic information.”
- Beauty and the Beast: Optimal Methods Meet Learning for Drone Racing (2018, methodology): “Traditional approaches to autonomous MAV navigation build on visual inertial odometry (VIO) [5], [1], [13], [24] or simultaneous localization and mapping (SLAM) [21], [23], which are used to provide a pose estimate of the drone relative to an internal metric map [14], [4].”
- RLBench: The Robot Learning Benchmark & Learning Environment (2019, neutral): “However, if we would like a manipulation system to make use of a SLAM map, it is not currently clear what the best way to represent this map is: whether it be sparse [54], [55], dense [56], [57], or semidense [58].”
- Habitat: A Platform for Embodied AI Research (2019, methodology): “We use the classic agent by Mishkin et al. [19] which leverages the ORB-SLAM2 [20] localization pipeline, with the same parameters as reported in the original work.”
- Zero-Shot Visual Imitation (2018, methodology): “However, at sub-sampling rate of five, SIFT-based feature matching approaches did not work and ORBSLAM2 (Mur-Artal & Tardós, 2017) failed to generate a map, whereas our method was successful.”
- Learning to Explore using Active Neural SLAM (2020, methodology): “2005; Hartley and Zisserman, 2003; Fuentes-Pacheco et al., 2015), by fusing information from multiple views of the environment. While sparse reconstruction can be done well with monocular RGB images (Mur-Artal and Tardós, 2017), dense mapping is inefficient (Newcombe et al., 2011) or requires special”
- DeepV2D: Video to Depth with Differentiable Structure from Motion (2018, methodology): “NYU does not have ground truth camera pose data, but we are able to generate good estimates applying RGB-D SLAM [36].”
- A Review of Visual-LiDAR Fusion based Simultaneous Localization and Mapping (2020, background): “Such research works generated semi-dense maps representing the environment, but direct SLAM approaches are time consuming and often require GPU based processing.”
- DynaSLAM II: Tightly-Coupled Multi-Object Tracking and SLAM (2020, background): “DynaSLAM II builds on the popular ORB-SLAM2 [1].”
- Differentiable SLAM-net: Learning Particle SLAM for Visual Navigation (2021, methodology): “We adopt the popular ORB-SLAM [49; 48] as a classic baseline.”
- SuperDepth: Self-Supervised, Super-Resolved Monocular Depth Estimation (2018, neutral): “[39], [13]) and limit our quantitative comparison to similar self-supervised learning based methods.”
- Fusion++: Volumetric Object-Level SLAM (2018, methodology): “They use a separate SLAM system, ORBSLAM2 [22], whereas in our system the discovered object instances are tightly integrated into the SLAM system itself.”
- ReFusion: 3D Reconstruction in Dynamic Environments for RGB-D Cameras Exploiting Residuals (2019, background): “[1] combine a geometric approach with a deep learning segmentation to enable removal of dynamics in an ORB-SLAM2 system [8].”
- Deep Visual Teach and Repeat on Path Networks (2018, background): “This process is fused in SLAM, which has has been successfully applied in numerous applications, including indoor navigation and autonomous driving [32, 22, 17, 16].”
- Stereo DSO: Large-Scale Direct Sparse Visual Odometry with Stereo Cameras (2017, methodology): “More recently, ORB-SLAM [21, 22] introduced an efficient visual SLAM solution based on ORB features and innovations such as map reuse.”