Add multi-view camera calibration: registerCameras and calibrateMultiview

[shimat]

Summary

Wraps the OpenCV 5 multi-view camera calibration functions — one of the headline new features of the calib module (from #1924).

API

• Cv2.RegisterCameras — registers a pair of cameras (the two may use different camera models, pinhole/fisheye), estimating their relative pose R/T plus the essential/fundamental matrices and per-view errors. Returns the overall RMS.
• Cv2.CalibrateMultiview — estimates per-camera intrinsics (Ks, distortions) and extrinsics (Rs, Ts, relative to camera 0) for a multi-camera rig from per-camera/per-frame detected pattern points, with a detection mask for partial observations. Returns the overall RMS.
• CameraModel enum (Pinhole / Fisheye).

Native bridge

• New calib_multiview.h. The jagged per-camera/per-frame imagePoints (vector<vector<Mat>>) is marshaled as a flat Mat[] plus a per-camera frame-count array; the InputOutputArrayOfArrays outputs use VectorOfMat.

Test

MultiviewCalibrationTest builds a synthetic 2-camera rig whose relative pose is identity rotation + a pure X translation (so camera 1's view is camera 0's view with the baseline applied):

• RegisterCameras recovers an X-dominant translation with low RMS.
• CalibrateMultiview returns per-camera Ks / distortions / Rs / Ts (2 each).

Rebuilt OpenCvSharpExtern locally; full Calib3D suite green (31 passed).

Note: calibrateMultiview requires single-channel CV_32F point matrices (object points NUM_POINTS x 3, image points NUM_POINTS x 2); a 3-/2-channel layout crashes the native algorithm. This is documented on the managed method.

Refs #1924.

🤖 Generated with Claude Code