ORB_SLAM3 is a visual-inertial SLAM library designed for real-time simultaneous localization and mapping. It provides a framework for tracking camera movement and building 3D maps of environments using monocular, stereo, or RGB-D cameras combined with inertial sensors.
The system features a multi-map fusion engine capable of merging separate spatial sessions into a single seamless representation of an environment. It includes specialized processing for wide-angle and fisheye lenses to expand the visual field of view for spatial tracking.
The library covers a broad range of spatial intelligence capabilities, including visual odometry, loop closure, and map optimization. It supports the integration of inertial data to estimate scale and velocity, as well as integration with the Robot Operating System for robotic spatial mapping and localization.
The project includes tools for execution performance analysis to measure processing time and system latency.