How calibration works¶
Calib GEOM estimates how your camera maps points in the real world to positions in an image. For multi-camera setups, it also estimates the cameras' relative positions and orientations.
You supply images of a target with known geometry. Calib GEOM detects its features, finds an initial estimate, and then adjusts the calibration to make the predicted feature positions agree with the detected positions. This final adjustment is called optimization.
The difference between a predicted and detected position is the reprojection error, measured in pixels. A low error is useful, but it is only one part of Analysis.
Choosing a camera model¶
A camera model describes the lens projection and distortion. Choose a model suited to your optics and to the software that will use the calibration.
| Model | Typical reason to choose it |
|---|---|
| OpenCV | Direct integration with OpenCV's standard camera model. |
| Matlab | Direct integration with MATLAB's compatible camera model. |
| OpenCV Fisheye | Fisheye calibration for use with OpenCV's fisheye tools. |
| Double Sphere, Enhanced Unified, FOV, OCamCalib | Alternative models for wide-angle or omnidirectional optics. |
| Brown-Conrady, Thin Prism, Halcon Division | Applications requiring these particular distortion models. |
| B-Spline Warp, B-Spline Ray (Pro) | Flexible modeling when simpler models leave systematic errors. |
Start with a suitable, relatively simple model. A more flexible model may fit the calibration images better but needs more varied observations to remain reliable. Compare coverage, residual patterns, and Test RPE before adding complexity. See Optimization.
Different models are not interchangeable, even if some parameters have the same names. If your application does not support the chosen model, consider exporting distortion maps or unprojection maps.
What makes a useful dataset?¶
- Features cover the usable image area, including edges and corners.
- Target positions include different tilts, distances, and image locations.
- Images are sharp and well exposed, with little glare or motion blur.
- Focus, aperture, camera mounting, and target geometry remain stable.
Repeating almost identical images adds less information than adding complementary views.
To learn more about planning a capture, read Calibration Best Practices.
Multiple cameras and targets¶
Calib GEOM can jointly calibrate multiple cameras and multiple targets. This is useful when one target cannot cover the full field of view or when cameras see different parts of a larger setup.
The observations must connect the cameras and targets through shared views. Targets used together as a rig must remain fixed relative to one another throughout acquisition. ChAruCo, Kalibr, and ArUco targets support this workflow.
Understanding uncertainty¶
Optional uncertainty estimates show how well the observations constrain the calibration parameters and what that means for projection or triangulation. They help identify weakly supported results, but do not account for every source of error, such as an inaccurate target or movement during capture.
Use them together with residual plots, Test RPE, and measurements representative of your application. See Analysis.