Author: Lochman
Year: 2021
- Calibration toolbox for omni cameras
- Rough calculation of a back projection model first (pixels -> rays), then refine the camera model
- Corner correction step
- The initial estimate is done with the Division model
Preliminaries:
- Micusik equation: for a scene point
$X$ and an image point$u$ : $$ \gamma g(A u) = P X $$ - A : maps from image coord to sensor coord (with scale, center of image and pixel ratio)
- g : map from retinal plane to ray direction (non linear)
$g(u) = g(u, v, \psi(r(u)) )$ - P : Camera matrix (K * T)
- Radial fundamental matrix:
$0 = u^T F_r x$
Initial estimate:
- solving the radial fundamental matrix to recover the center of projection and the camera pose
- correction of corners with epipolar constraint and fundamental matrix refinement => proved to increase accuracy on simulated data
- solving intrinsics of division model of degree 4 (empirical value from simulated data)
- model to model regression to get initial estimate of the wanted camera model