Skip to content

Latest commit

 

History

History
31 lines (24 loc) · 1.5 KB

accuracyFisheye.md

File metadata and controls

31 lines (24 loc) · 1.5 KB

Accuracy of fish-eye lens models

Author: Hughes

Year: 2010

Notes:

  • Radial barrel distortion noticeable on fish eye lenses => resolution decrease on peripherical areas of the image
  • odd-order polynomial models cannot sufficiently compensate for radial distortion => alternative models studied here
  • assumption : lens displacement curve (?) is linear near the foveal and is monotically increasing
  • Displacement curve = describes the radial dusplacement of points from pinhole to fisheye plane given any radial value
  • pinhole model preserves straightlines $r_u = f tan(\theta) $

projection models:

  • equidistant $ r_d = f \theta $
  • equisolid: length of the chord on the unit sphere $ r_d = 2f sin(\frac{\theta}{2} )$
  • orthographic $ r_d = f sin(\theta) $ but suffer from strong distortion
  • stereographic $ r_d = 2f tan(\frac{\theta}{2})$ the center of reprojection is the oposite of the tangential point
  • PFET (polynomial Fisheye transform) $ r_d = P(r_u)$ 5th order is good
  • FET (fish eye transform) $r_d = s ln(1+\lambda r_u)$
  • FOV model $r_d = \frac{1}{\omega} arctan \left ( 2 r_u tan (\frac{\omega}{2})\right )$ with $\omega$ the FOV of the camera
  • radial distortion can model deviation from real model $\Delta r_d = P(r_u)$

Results:

  • Most of fish eye lenses are supposed to fit with equidistant or equisolid projection model
  • Init f = 2 / FOV and 0 for the radial distortion parameters

Commentaire:

Décrit précisément les modèles de lentille, top! Pas hyper bien compris la technique de calibration