This paper presents ORB-SLAM, a feature-based monocular SLAM system that operates in real time in various environments. The system is robust to motion clutter, allows wide baseline loop closing and relocalization, and includes full automatic initialization. ORB-SLAM uses the same features for all SLAM tasks: tracking, mapping, relocalization, and loop closing. A survival of the fittest strategy selects points and keyframes for a compact, trackable map that grows only when the scene changes. The system is evaluated on 27 sequences from popular datasets, achieving unprecedented performance compared to other monocular SLAM approaches. The source code is made public for community use. The system is built on previous work, including place recognition, scale-aware loop closing, and covisibility information for large-scale operation. ORB-SLAM uses ORB features, which are fast and invariant to viewpoint and illumination. It operates in real time in large environments using a covisibility graph, focusing tracking and mapping on a local area. Real-time loop closing is based on a pose graph optimization called the Essential Graph. Real-time relocalization is robust to viewpoint and illumination changes. An automatic initialization procedure creates initial maps for planar and non-planar scenes. A survival of the fittest approach selects keyframes and map points, improving tracking robustness and lifelong operation. The system is evaluated on indoor and outdoor sequences, including hand-held, car, and robot sequences. ORB-SLAM achieves better camera localization accuracy than direct methods, which optimize over pixel intensities instead of feature reprojection errors. The loop closing and relocalization methods are based on previous work, with improvements in initialization, the Essential Graph, and overall performance. The system is compared with other SLAM approaches, including PTAM, LSD-SLAM, and RGBD-SLAM, showing superior performance in terms of accuracy, robustness, and scalability. ORB-SLAM is the most complete and reliable solution for monocular SLAM, with public source code and demonstration videos available.This paper presents ORB-SLAM, a feature-based monocular SLAM system that operates in real time in various environments. The system is robust to motion clutter, allows wide baseline loop closing and relocalization, and includes full automatic initialization. ORB-SLAM uses the same features for all SLAM tasks: tracking, mapping, relocalization, and loop closing. A survival of the fittest strategy selects points and keyframes for a compact, trackable map that grows only when the scene changes. The system is evaluated on 27 sequences from popular datasets, achieving unprecedented performance compared to other monocular SLAM approaches. The source code is made public for community use. The system is built on previous work, including place recognition, scale-aware loop closing, and covisibility information for large-scale operation. ORB-SLAM uses ORB features, which are fast and invariant to viewpoint and illumination. It operates in real time in large environments using a covisibility graph, focusing tracking and mapping on a local area. Real-time loop closing is based on a pose graph optimization called the Essential Graph. Real-time relocalization is robust to viewpoint and illumination changes. An automatic initialization procedure creates initial maps for planar and non-planar scenes. A survival of the fittest approach selects keyframes and map points, improving tracking robustness and lifelong operation. The system is evaluated on indoor and outdoor sequences, including hand-held, car, and robot sequences. ORB-SLAM achieves better camera localization accuracy than direct methods, which optimize over pixel intensities instead of feature reprojection errors. The loop closing and relocalization methods are based on previous work, with improvements in initialization, the Essential Graph, and overall performance. The system is compared with other SLAM approaches, including PTAM, LSD-SLAM, and RGBD-SLAM, showing superior performance in terms of accuracy, robustness, and scalability. ORB-SLAM is the most complete and reliable solution for monocular SLAM, with public source code and demonstration videos available.