|
|
|
Camera calibration is the recovery of the intrinsic parameters of a camera.
The standard off-line calibration procedure is to acquire images of an object
with known Euclidean structure and compute the camera matrix that minimizes the
error between the image of the object and its reprojection, as represented in the
animation on the left. An interesting example of
this method is the use of
surfaces of revolution for camera calibration,
developed by Wong, Mendonça and Cipolla, 2000
(download paper,
BibTeX entry). This class of techniques
gives good results, but knowledge of the Euclidean structure of a scene may not
always be available. Moreover, if the intrinsic parameters change,
the camera has to be recalibrated.
|
||
|
To overcome these difficulties, the idea of camera self-calibration was developed. Self-calibration methods can be used when no Euclidean information is available, and, if properly designed, can also cope with varying intrinsic parameters. Knowledge of the camera motion and the intrinsic parameters allows for the Euclidean reconstruction of the sceene. However, if the intrinsic parameters are wrong, the reconstruction will suffer from a projective distortion even if the motion is correct, as in the images shown below. The need for good Euclidean reconstructions is a good enough motivation for the pursue of accurate camera calibration. |
||
| Euclidean reconstruction | Projective reconstruction I | Projective reconstruction II |
|---|---|---|
|
|
|
|
Image produced by Kenneth Wong |
Image produced by Kenneth Wong |
Image produced by Kenneth Wong |
Self-calibration from the Huang-Faugeras' Constraints| Previous | Home | Next |
|
This page is eternally under construction. Last modified: September 13, 2000. |