Pattern-based calibration and radial distortion
| This documentation was generated with the assistance of AI. Please report any inaccuracies. |
Self-calibration recovers intrinsics with no calibration object, from
natural-scene correspondences alone. When a physical calibration pattern is available (a checkerboard, a
circle grid, a QR-coded grid), irurueta-ar also provides the more classical alternative: photograph the
pattern from several poses and solve directly for each camera’s intrinsics from known-to-observed point
correspondences.
Calibration patterns
com.irurueta.ar.calibration.Pattern2D is the abstract base for a set of ideal, known 2D points; concrete
patterns are QRPattern2D (a QR-code-like marker grid) and CirclesPattern2D (a grid of circle centers), both
exposing getIdealPoints().
Camera calibrators
Given a Pattern2D and one CameraCalibratorSample per photographed view (the pattern’s detected image
points for that view), com.irurueta.ar.calibration.CameraCalibrator estimates the homography between the
ideal pattern and each view, then the image of the absolute conic per view, following the same
DIAC-to-intrinsics relation described in Self-calibration but now with the "world
conic" being the known pattern instead of the (unknown) absolute conic at infinity. Two concrete strategies are
provided: AlternatingCameraCalibrator (alternates between estimating intrinsics and radial distortion) and
ErrorOptimizationCameraCalibrator (jointly refines both by minimizing reprojection error). Both are created
through the CameraCalibrator.create(…) factory, selected via CameraCalibratorMethod.
Radial distortion
Lens radial distortion is estimated separately from the pinhole intrinsics, following the same two-tier
robust/non-robust pattern: LMSERadialDistortionEstimator /
WeightedRadialDistortionEstimator fit distortion coefficients from distorted/undistorted point pairs, and
{RANSAC,LMedS,MSAC,PROSAC,PROMedS}RadialDistortionRobustEstimator add outlier rejection when those pairs come
from noisy detections. Maps to com.irurueta.ar.calibration.RadialDistortion /
com.irurueta.ar.calibration.estimators.RadialDistortionEstimator.
Example
Pattern2D pattern = new QRPattern2D();
var calibrator = CameraCalibrator.create(pattern, samples); // one sample per photographed view
calibrator.calibrate();
PinholeCameraIntrinsicParameters intrinsics = calibrator.getEstimatedIntrinsicParameters();
RadialDistortion distortion = calibrator.getDistortion();References
Irurueta, PhD thesis §2 (camera models, DLT camera estimation)
Hartley & Zisserman, 2003 §7 (computation of the camera
matrix)