How uz-slamlab/orb_slam3 works
ORB-SLAM3 is a real-time visual, visual-inertial, and multi-map SLAM library for monocular, stereo, and RGB-D cameras, including pinhole and fisheye models. Its documented user workflow supplies calibrated camera or dataset observations, runs SLAM, and produces a trajectory; optional ROS examples provide another input route. The graph separates tracking, map construction, and place recognition from the internal geometry and optimization layers. Several runtime connections are inferred from the product description rather than sampled caller code, so those edges are dashed and uncited.
Main entry points: Tracking.cc Tracking · System.cc SLAM system · Frame.cc Frame processing · Optimizer.cc SLAM optimization
Components 24
SLAM runtime
- SLAM system
src/System.cc - Tracking
src/Tracking.cc - Configuration
src/Settings.cc - Visualization
src/Viewer.cc - ROS input
Examples_old/ROS/ORB_SLAM3/src/ros_stereo_inertial.cc
Mapping and recovery
- Multi-map atlas
src/Atlas.cc - Map state
src/Map.cc - Map points
src/MapPoint.cc - Keyframes
src/KeyFrame.cc - Local mapping
src/LocalMapping.cc - Loop and merge
src/LoopClosing.cc - Keyframe database
src/KeyFrameDatabase.cc
Vision and geometry
- Frame processing
src/Frame.cc - ORB features
src/ORBextractor.cc - Camera models
src/CameraModels - Two-view geometry
src/TwoViewReconstruction.cc - IMU preintegration
src/ImuTypes.cc
Core algorithms
- SLAM optimization
src/Optimizer.cc - ORB vocabulary
include/ORBVocabulary.h - DBoW2 vocabulary
Thirdparty/DBoW2/DBoW2/TemplatedVocabulary.h - Graph optimizer
Thirdparty/g2o/g2o/core/optimizable_graph.cpp - Lie geometry
Thirdparty/Sophus/sophus/se3.hpp
Other components
- Camera user
- Camera observations
Connections 23
- Camera user → Camera observations: provides images
README.md - Camera observations → SLAM system: runs SLAM
- SLAM system → Tracking: dispatches tracking
- SLAM system → Configuration: configures
- Tracking → Frame processing: processes frames
- Frame processing → ORB features: extracts features
- Frame processing → Camera models: uses projection
- Tracking → Two-view geometry: initializes geometry
- Tracking → IMU preintegration: integrates measurements
- Tracking → Keyframes: creates keyframes
- Keyframes → Map points: associates points
- Local mapping → Map state: updates map
- Local mapping → SLAM optimization: optimizes locally
- Tracking → Keyframe database: queries relocalization
- Loop and merge → Keyframe database: queries candidates
- Loop and merge → Multi-map atlas: merges maps
- Multi-map atlas → Map state: maintains maps
- SLAM optimization → Graph optimizer: builds graph
src/Optimizer.cc - Frame processing → ORB vocabulary: computes BoW
- ORB vocabulary → DBoW2 vocabulary: uses vocabulary
include/ORBVocabulary.h - SLAM optimization → Lie geometry: uses poses
- SLAM system → Visualization: provides visualization
- ROS input → SLAM system: submits sensor data
README.md