Image distortion calibration

Image undistortion example with OpenCV Python for TUM Visual-Inertial Dataset

Camera geometric model

(Reference: https://github.com/ethz-asl/kalibr/wiki/supported-models)

Two key parameters is been used for geometric calibration:

1. Intrinsic model
2. distortion model

Camera model

Omnidirectional

omnidirectional camera model (omni) (intrinsics vector: [xi fu fv pu pv])

The intrinsics vector contains all parameters for the model:

• fu, fv: focal-length
• pu, pv: principal point
• xi: mirror parameter (only omni)

  K = 
  fu   fu*cot(xi) pu
  0     fv        pv
  0     0         1

  Pinhole

  pinhole camera model (pinhole) (intrinsics vector: [fu fv pu pv])

  K = 
  fu    0       pu
  0     fv      pv
  0     0       1

Distortion model

equidistant (equi) (distortion_coeffs: [k1 k2 k3 k4])

Before calibration (pinhole-equi-512 model)

After calibration (pinhole-equi-512 model)

Matlab example

https://github.com/ethz-asl/kalibr/files/272054/Undistort.m.txt

imageD = imread('1.png');
% pinhole-equi-512
intri = [190.97847715128717, 190.9733070521226, 254.93170605935475, 256.8974428996504];
D = [0.0034823894022493434, 0.0007150348452162257, -0.0020532361418706202, 0.00020293673591811182];
fu = intri(1);
fv = intri(2);
pu = intri(3);
pv = intri(4);

K = [fu 0 pu
    0 fv pv
    0 0 1];
unImg = Undistort(imageD, D, K, 'equi');
imshow(unImg);