This project implements Zhengyou Zhang's "A Flexible New Technique for Camera Calibration" (2000).
Wilhelm Burger's "Zhang's Camera Calibration Algorithm: In-Depth Tutorial and Implementation" (2016) provided substantial aid to my understanding of details in the original paper.
python calibrate.py --data_dir=<data_directory_path>
data_dir- Path to the directory containing the control point file and the image point files. Each file must contain an even number of space-separated values of <x, y> points.
- Python - Tested on version 3.7.0
- Matplotlib - Tested on version 2.0.2
- NumPy - Tested on version 1.15.0
- OpenCV - Tested on version 3.4.1
- Scipy - Tested on version 0.19.1
Here is a comparison with Zhang's original results (available here):
| Implementation | Ground Truth | |
|---|---|---|
fx |
832.499 | 832.500 |
fy |
832.529 | 832.530 |
s |
0.204503 | 0.204494 |
px |
303.959 | 303.959 |
py |
206.585 | 206.585 |
q1 |
-0.228602 | -0.228601 |
q2 |
0.190355 | 0.190353 |
fx and fy are the intrinsic focal length parameters, s represents skew, and px and py are the principal point offsets. q1 and q2 are distortion model coefficients for the radial distortion model.
To visualize the calibration results, here are the calculated world- and camera-centric extrinsic parameters from the original dataset:
