Camera Calibration for 3D Reconstruction and View Transformation - 3D Modeling and Animation: Synthesis and Analysis Techniques for the Human Body - page 109

Other techniques

In addition to traditional optimization methods, some other techniques, including
the Bayesian probability framework, genetic algorithms, and artificial neural
nets, have also been considered in the camera calibration system (Redert, 2000).
In the area of structure from motion, n ( n ) cameras are calibrated one after

another based on an extension of the Kalman filter (Jebara, Azarbayejani &
Pentland, 1999).

> 1

Other Calibration-Related Investigations

Besides algorithms for calibrating cameras, other related aspects have been
investigated to refine the performance of the camera calibration system. For
example, several suggestions on the design of the calibration object and the data
acquisition for increasing the calibration performance were given in Tu &
Dubuisson (1992).

Feature extraction issue

For passive camera calibration, the calibration object should be carefully
designed to ease the extraction of feature points. Several types of calibration
object patterns have been used, e.g., a circular pattern in Heikkilä (2000) and a
checkerboard pattern in Bouguet (2002). Here sub-pixel feature extraction
(Devernay, 1995) is always necessary.
In most cases, features are extracted before the camera parameters are
calculated. If feature coordinates are not accurately located, the camera
calibration results will not be accurate, either. Unfortunately, when there are
distortions in the imaging process, feature extraction will always suffer from
systematic errors (Heikkilä, 2000). To circumvent this problem, the calibration
can be carried out directly, based on the intermediate characterization of image
features, such as maxima of the intensity gradient or zero crossings of the
Laplacian. The same idea was also applied to distortion estimation in Ahmed &
Farag (2001). Alternatively, the whole process, including the feature extraction
and the camera calibration, can be carried out iteratively to yield better accuracy.
Within each iteration, the feature extraction is performed after the distortion
correction with the distortion coefficients obtained from the calibration (Devernay
& Faugeras, 2000).