Camera calibration is a foundational process in computer vision, robotics, and 3D reconstruction. It determines the camera's intrinsic properties—including focal length, principal point, and lens distortion parameters—enabling precise correction of image geometry and conversion between pixel cooordinates and real-world measurements.

Preparation

1. Calibration Pattern

A checkerboard pattern is common used due to its distinct corners, which are easily detectable. Record the following pattern details:

• Number of horizontal inner corners (e.g., 9).
• Number of vertical inner corners (e.g., 6).
• Actual physical size of a single square (e.g., 0.025 meters).

Note: The count refers to inner corners (the intersection points between squares), not the total number of squares. A 10x7 checkerboard yields 9x6 inner corners.

2. Image Capture Guidelines

• Capture 15–30 images of the pattern from various distances and orientations.
• Ensure the pattern occupies a significant portion of each image and appears near the image edges.
• Avoid blur, glare, or partial occlusion of the pattern.
• Maintain consistent lighting conditions throughout the capture session.

Calibration Procedure in MATLAB

Step 1: Launch the Calibration Tool

In MATLAB, navigate to the Apps tab, select Image Processing and Computer Vision, and open the Camera Calibrator app.

Step 2: Load Calibration Images

Click Add Images and select the folder containing your pattern images. A dialog will prompt you to enter the pattern specifications:

• Number of squares in [width/height]: Enter the count of inner corners for each dimension.
• Square size: Enter the physical side length of one square. Using meters (m) is recommended for consistency with other sensors (e.g., LiDAR, IMU).

The tool will automatically detect the patttern in each image. Images where detection fails will be flagged and can be excluded.

Step 3: Configure Calibration Settings (Optional)

Click Calibration Settings to adjust parameters:

• Radial Distortion: Model lens curvature effects. 2 coefficients is suitable for standard lenses; 3 coefficients is for wide-angle or fisheye lenses.
• Skew: Typically enabled for modern digital cameras where pixel axes are orthogonal.
• Tangential Distortion: Recommended for most cameras to account for minor lens-sensor misalignment.

Step 4: Perform Calibration

Click Calibrate to compute the camera's intrinsic matrix and distortion coefficients. The tool reports the mean reprojection error. A value below 0.5 pixels is generally acceptable. Higher errors indicate poor image quality or insufficient viewpoint coverage.

Step 5: Review and Validate Results

• Inspect the Reprojection Errors plot to identify outlier images.
• Use the Show Undistorted button to preview the corrected images.
• Generate a calibration report (PDF) for documentation.

Step 6: Export Calibration Parameters

Click Export Camera Parameters to save the results as a cameraParameters object in the MATLAB workspace. Key properties of this object include:

• IntrinsicMatrix: Contains focal lengths (fx, fy) and principal point coordinates (cx, cy).
• RadialDistortion: Coefficients for radial distortion correction.
• TangentialDistortion: Coefficients for tangential distortion correction.

You can access these parameters programmatically:

% Access the intrinsic matrix
K = cameraParams.IntrinsicMatrix;
% Access distortion coefficients
radialCoeffs = cameraParams.RadialDistortion;
tangentialCoeffs = cameraParams.TangentialDistortion;

Troubleshooting Common Issues

Issue	Likely Cause	Recommended Action
Pattern detection fails	Poor lighting, motion blur, extreme viewing angle	Recapture images with clear, fronto-parallel views of the pattern.
High reprojection error	Insufficient number of images or poor viewpoint distribution	Capture more images, ensuring the pattern appears in all image quadrants.
Inconsistent calibration results	Incorrect physical square size entered	Verify the actual dimensions of the checkerboard squares.
Severe cropping after undistortion	Excessive lens distortion or very wide field of view	Consider using a fisheye lens model via the fisheyeCalibrator app.