| Platform | CI Status |
|---|---|
| Ubuntu (Focal) |  |
| Lint (Clang-Format) |  |
| Lint (CMake-Format) |  |
A loosely connected bundle of tools for calibrating cameras to (industrial) robots and vice-versa with a focus on being easy to integrate into more complex applications.
This work is based on the work of Dr. Chris Lewis @drchrislewis (see industrial_calibration) and modifications made by Geoffrey Chiou in IC2.
The focus and philosophy of this library is provide to provide a suite of stand-alone "optimization functions" that take as input structure full of well-documented arguments and produce an answer that is your calibration. How you collect data and what you do with the answer is left to you.
This library is meant to be used within the ROS ecosystem, but the core optimizations depends on only Ceres-Solver and its dependencies. Follow the link for instructions to install Ceres. Afterwards, clone this package into your workspace and build with catkin.
Other support libraries use additional tools:
- The built-in "target finder" in
rct_image_toolsuses OpenCV. - The data collection helpers in
rct_ros_toolsuses TF and various image transport mechanisms.
- Please read the calibration primer for a description of terminology and conventions used in this library.
- Please visit the documentation for the calibration you wish to solve in
rct_optimizations. See below for a discussion of supported problems. - If you need to do intrinsic calibration of your sensor, then please use ROS' built in method from here. Alternatively, use OpenCV's calibrateCamera function on which the ROS utility is built.
- After setting up your "experiment", perform any necessary data collection: moving robots, taking images, etc. Be sure to get a good diversity of images; skew is important!
- Create an instance of your problem's
Problemdata structure and fill it out. - Call
rct_optimizations::optimize(Problem&)and get aResultback. - Check that the result converged and then apply it to your workcell: either manually modify the URDF, publish the TF frame, etc. Consider using something like (shameless plug)
mutable transform publisher.
rct_optimizations: Contains core libraries for performing the "math" of optimizations, usually using Ceres. This is where you'll find your library calls.rct_image_tools: Builds on OpenCV to provide tools for finding targets in images. This lets you extract "correspondences" easily for supported target types.rct_examples: This package contains both simple examples and offline processing tools for a variety of common calibrations. Look here for experimental setups and examples.rct_ros_tools: (Experimental) Provides some "helper tools" for data collection and loading of data through ROS parameters/YAML files.
See the readme of rct_examples for more information on the most popular and tested calibrations. There's even some nice pictures.
Currently supported calibrations:
- Extrinsic calibration of a 2D camera on robot wrist
- Extrinsic calibration of a 3D camera on robot wrist
- Extrinsic calibration of one or more static cameras in the workcell
Semi-Experimental (these have testing or robustness that needs to be improved):
- Solve PnP for one or more cameras
- Intrinsic calibration
The source code for these calibrations can be found in rct_optimizations.
For example, see rct_example's camera-on-wrist calibration.
- Replicate the OpenCV intrinsic calibration
- Implement Qt GUI based on IC2 to speed calibration process in simple scenarios.
- Provide examples of qualifiying the accuracy of a calibrated system
- Investigate robot kinematic calibration (ala Mike Ferguson' package)
- Provide tools explicitly for robot tool calibration (for the moment, please see my super simple one at tool_point_calibration
