ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual–Inertial, and Multimap SLAM

Abstract

This paper presents ORB-SLAM3, the first system able to perform visual,\nvisual-inertial and multi-map SLAM with monocular, stereo and RGB-D cameras,\nusing pin-hole and fisheye lens models. The first main novelty is a\nfeature-based tightly-integrated visual-inertial SLAM system that fully relies\non Maximum-a-Posteriori (MAP) estimation, even during the IMU initialization\nphase. The result is a system that operates robustly in real-time, in small and\nlarge, indoor and outdoor environments, and is 2 to 5 times more accurate than\nprevious approaches. The second main novelty is a multiple map system that\nrelies on a new place recognition method with improved recall. Thanks to it,\nORB-SLAM3 is able to survive to long periods of poor visual information: when\nit gets lost, it starts a new map that will be seamlessly merged with previous\nmaps when revisiting mapped areas. Compared with visual odometry systems that\nonly use information from the last few seconds, ORB-SLAM3 is the first system\nable to reuse in all the algorithm stages all previous information. This allows\nto include in bundle adjustment co-visible keyframes, that provide high\nparallax observations boosting accuracy, even if they are widely separated in\ntime or if they come from a previous mapping session. Our experiments show\nthat, in all sensor configurations, ORB-SLAM3 is as robust as the best systems\navailable in the literature, and significantly more accurate. Notably, our\nstereo-inertial SLAM achieves an average accuracy of 3.6 cm on the EuRoC drone\nand 9 mm under quick hand-held motions in the room of TUM-VI dataset, a setting\nrepresentative of AR/VR scenarios. For the benefit of the community we make\npublic the source code.\n

Affiliated Institutions

• Universidad de Zaragoza ES

Related Publications