如何将MATLAB立体标定结果转换为OpenCV可用的.npz文件？

Great question—converting MATLAB stereo calibration results to a .npz file (the format typically used by OpenCV/Python scripts) is a common workflow, and it’s straightforward with a two-step process: exporting from MATLAB, then converting to .npz with Python. Here’s how to do it properly:

Step 1: Export Stereo Calibration Parameters from MATLAB #

After running MATLAB's stereo calibration, you’ll have a stereoParams struct containing all the critical data. First, extract and organize these parameters to match the structure expected by your target script (usually aligned with OpenCV’s output format), then save them to a .mat file.

% Assume you've completed stereo calibration and have the stereoParams struct
calib_data = struct();

% Extract left camera parameters
calib_data.cameraMatrix1 = stereoParams.CameraParameters1.IntrinsicMatrix;
% Combine radial and tangential distortion coefficients (matches OpenCV's order: k1, k2, p1, p2, k3)
calib_data.distCoeffs1 = [stereoParams.CameraParameters1.RadialDistortion, ...
                          stereoParams.CameraParameters1.TangentialDistortion];

% Extract right camera parameters
calib_data.cameraMatrix2 = stereoParams.CameraParameters2.IntrinsicMatrix;
calib_data.distCoeffs2 = [stereoParams.CameraParameters2.RadialDistortion, ...
                          stereoParams.CameraParameters2.TangentialDistortion];

% Extract stereo transformation parameters
calib_data.R = stereoParams.RotationOfCamera2;  % Rotation matrix from left to right camera
calib_data.T = stereoParams.TranslationOfCamera2;  % Translation vector from left to right camera

% Optional: Add reprojection error if your script needs it
calib_data.reproj_error = stereoParams.ReprojectionError;

% Save the structured data to a .mat file
save('matlab_stereo_calib.mat', '-struct', 'calib_data');

Step 2: Convert .mat to .npz Using Python #

Since MATLAB doesn’t natively support .npz files, we’ll use Python (with scipy and numpy) to read the .mat file and save it in the required format. This ensures compatibility with scripts expecting .npz inputs.

import scipy.io
import numpy as np

# Load the MATLAB .mat file
mat_calib = scipy.io.loadmat('matlab_stereo_calib.mat')

# Extract parameters and adjust formats to match OpenCV conventions
cameraMatrix1 = mat_calib['cameraMatrix1']
distCoeffs1 = mat_calib['distCoeffs1'].flatten()  # Convert to 1D array (OpenCV's preferred format)
cameraMatrix2 = mat_calib['cameraMatrix2']
distCoeffs2 = mat_calib['distCoeffs2'].flatten()
R = mat_calib['R']
T = mat_calib['T'].flatten()  # Convert translation vector to 1D
reproj_error = mat_calib['reproj_error'].item()  # Extract scalar value

# Save as .npz file with keys your script expects
np.savez('file.npz',
         cameraMatrix1=cameraMatrix1,
         distCoeffs1=distCoeffs1,
         cameraMatrix2=cameraMatrix2,
         distCoeffs2=distCoeffs2,
         R=R,
         T=T,
         reproj_error=reproj_error)

Key Compatibility Notes #

• Match Key Names: Double-check the key names used in your target script. If it uses M1 instead of cameraMatrix1, for example, adjust the keys in the np.savez() call accordingly.
• Distortion Coefficient Order: MATLAB’s radial distortion is [k1, k2, k3] (if 3 coefficients are used) and tangential is [p1, p2]. Ensure this matches the order your script expects (OpenCV uses [k1, k2, p1, p2, k3, k4, k5, k6] by default).
• Matrix Storage: MATLAB uses column-major order, while numpy uses row-major, but for 2D matrices like camera intrinsics and rotation matrices, this won’t affect functionality as long as the dimensions are correct.

内容的提问来源于stack exchange，提问作者gaetano ma