Small changes done by Autoliv Research

• Add devcontainer to be able to work with VSCode in a dev container: Rename .devcontainer/devcontainer_template.json to .devcontainer/devcontainer.json and don't forget to mount your data path.
• In the ROS2 preprocess node, we use the topic type instead of the topic name to differentiate Images to CompressedImages.

direct_visual_lidar_calibration

This package provides a toolbox for LiDAR-camera calibration that is:

• Generalizable: It can handle various LiDAR and camera projection models including spinning and non-repetitive scan LiDARs, and pinhole, fisheye, and omnidirectional projection cameras.
• Target-less: It does not require a calibration target but uses the environment structure and texture for calibration.
• Single-shot: At a minimum, only one pairing of a LiDAR point cloud and a camera image is required for calibration. Optionally, multiple LiDAR-camera data pairs can be used for improving the accuracy.
• Automatic: The calibration process is automatic and does not require an initial guess.
• Accurate and robust: It employs a pixel-level direct LiDAR-camera registration algorithm that is more robust and accurate compared to edge-based indirect LiDAR-camera registration.

Documentation: https://koide3.github.io/direct_visual_lidar_calibration/
Docker hub: koide3/direct_visual_lidar_calibration

Video

Dependencies

• ROS1/ROS2
• PCL
• OpenCV
• GTSAM
• Ceres
• Iridescence
• SuperGlue [optional]

Getting started

1. Installation / Docker images
2. Data collection
3. Calibration example
4. Program details

License

This package is released under the MIT license.

Publication

Koide et al., General, Single-shot, Target-less, and Automatic LiDAR-Camera Extrinsic Calibration Toolbox, ICRA2023, [PDF]

Contact

Kenji Koide, National Institute of Advanced Industrial Science and Technology (AIST), Japan