How hkust-aerial-robotics/vins-fusion works
Project type and purpose
• VINS-Fusion is a ROS-based, optimization-driven multi-sensor (camera + IMU + optional GPS) state estimator for drones, cars, AR/VR.
• It comprises several ROS packages:
• vins_estimator (core VIO),
• loop_fusion (visual loop closure),
• global_fusion (GPS fusion),
• camera_models (calibration utilities),
• config (device YAMLs),
• docker (containerization).
Main components
a. Sensors & data sources
– Cameras (mono/stereo)
– IMU
– (Optional) GPS
– Rosbag recordings
b. Core nodes
– vins_node (in vins_estimator)
– loop_fusion_node
– global_fusion_node
c. Algorithm modules
– Feature Tracker
– Estimator (factor graph construction)
– Ceres Solver (non-linear optimizer)
– DBoW2 (loop detection)
– GeographicLib (global optimization)
d. Support
– Config YAML files
– camera_models library
– Docker container for environment consistency
– RViz for visualization
Relationships & data flow
• Sensor drivers / rosbag → publish raw image & IMU topics
• vins_node subscribes → feature tracking → factor graph → optimization → publishes pose
• loop_fusion_node subscribes to poses → builds pose graph with DBoW2 → optimizes → publishes corrected poses
• global_fusion_node optionally fuses GPS + VIO poses → publishes global poses
• All pose streams → RViz for display
Architectural patterns & design principles
• Modular ROS node structure (single-responsibility nodes)
• Publisher/Subscriber data flow with clear topic boundaries
• Library reuse: camera_models for calibration, Ceres for optimization, DBoW2 for loop closure, GeographicLib for geodesy
• Configuration-driven behavior via YAMLs and ROS launch files
• Containerized build/runtime via Docker