Camera Calibration

Two different approaches to camera calibration - one using OpenCV's built-in functions and another using SVD (Direct Linear Transform). Both scripts take 3D-2D point correspondences and compute the camera projection matrix.

Setup

Just install the dependencies:

pip3 install -r requirements.txt

You'll need numpy and opencv-python (though only the OpenCV script actually uses opencv).

Input Files

The scripts expect two files:

• 3D.txt - 3D world coordinates (X, Y, Z per line)
• 2D.txt - corresponding 2D image coordinates (u, v per line)

First line of each file should be the point count, then the coordinates follow.

Running the Scripts

OpenCV method:

python3 opencv_calibrate.py

Uses OpenCV's calibrateCamera which handles distortion and gives you rotation/translation vectors. Starts with a reasonable initial guess for the intrinsic matrix (focal length ~500, principal point at image center).

SVD method:

python3 svd_calibrate.py

Pure numpy implementation using DLT. Builds the system of equations and solves via SVD. Simpler approach but doesn't handle distortion - just gives you the projection matrix directly.

Both print out the 3x4 projection matrix and the average reprojection error so you can compare results.

Sample Output (OpenCV vs SVD Calibration)