which must be checked continually and corrected for
if not achieved. The corrections can be done as a self
calibration of the principle point in the bundle
adjustment.
The high speed cameras normally used have a
rotating prism and a continuously moving film. The
errors caused by this type of camera can be modelled
as a scale factor between the axis and separate radial
distortion coefficients for x and y.
Since the film is digitized with a scanner the
calibration must also be able to pick up the errors
caused by that procedure. The main type of error is
most probably a scaling error between the coordinate
axis and will be modelled as such.
In the single camera calibration the type of calibration
parameters can be any combination of the following:
- Principle distance
- Principle point
- Radial distortion
separate in x and y with a 3 ° polynomial
a uniform 5 ? polynomial
- Scale difference between x and y
The cameras can also be calibrated in the bundle
adjustment as additional parameters. Since there are
as many cameras as there are camera stations the
requirements for the ground truth is high even for
the self calibration, but since some of the calibration
parameters can be regarded as fix from previous
calculations the ability is still of use. It is also possible
to use other type of additional parameters such as
polynomial deformations.
43.1 Mathematical Formulation The calcu-
lations are based on the central projection model, i.e.
all image rays are assumed to pass through one and
the same point, the projection centre. This leads to
the use of the collinearity equations, which states that
the object point, the projection centre and the image
point lie on a straight line. If the rotation matrix, the
projection centre coordinates and the calibration
parameters are unknown the observation equations
will look like :
a11(X-X0) + a12(Y-YO) + a13(Z-Z0)
31(X-X0) + a32(Y-Y0) + a33(Z-Z0)
VX = (-X' + Xpp + corr) * lec
ca Le à A21(X-XO) + a22(Y-YO) + a23(Z-Z0)
VXS Cy + VPP + come Sce e GCXO) + a32(V-VO) +. 233(Z-Z0)
x, y the observed image coordinates
Xpp.Ypp the principal point
corr correction terms for the radial
distortion
SC scale difference in x and y
€ the principal distance
X Y,Z the object point coordinates
Xo,Yo,Zo the projection centre coordinates in
the object point coordinate system
811.233 the elements of the rotation matrix of
the image
The two equations are not linear in the unknowns.
To solve the equation system they are expanded in a
Taylor series where only the first degree terms are
used. The solution is iterated until stability.
43.2 Requirements for the test field The point
configuration for the test field is based on the
geometrical conditions for determining the
unknown parameters with as few points as possible
without loosing the quality and security in the
estimated parameters. The unknowns are the six
orientation parameters, principal distance, principal
point, radial correction terms (2) and a scale factor
between x and y giving a total of 12 unknowns.
The principal distance needs targets distributed
in the outer parts of the image in as large depth
difference as possible.
The principal point needs points in a similar
manner as the principal distance, but is
strengthen if 3D information is available in the
centre of the image as well.
The radial distortion is dependent on the
distribution of points over the whole image. It is
not dependent on any 3D information.
The scale difference have similar requirements
as the radial distortion.
The construction of a test field with this properties,
which at the same time is stable, foldable and easy to
move, is not a trivial matter. A three-folded test field
with locking devices to keep the stability is a model
which is considered.
43.3 Approximate values To be able to run the
calibration approximate values must be entered by
the operator for
- position (X,Y,Z in object coordinate system)