2. CALIBRATION AND ORIENTATION
The system calibration of POM provides the complete
calculation of interior and exterior orientation of the RSC
sensors. The spatial position of the axis of the rotary
table within the machine coordinate system is also
determined.
2.1 Camera Calibration
The determination of the interior orientation of the
cameras can be performed by a suitable
photogrammetric network using a testfield with sufficient
point density. In this case the 3D coordinates of the
testfield must not be known.
The initial approach for camera calibration was as follow:
Each camera was calibrated in the laboratory (IPB)
outside the POM system. A suitable set of 7 images was
taken of a testfield which carried 35 spatially distributed
targets. After self-calibrating bundle adjustment the
parameters of interior orientation were obtained with
standard deviations less than 5um. The mean accuracy
of image measurement was about 1pm.
One shortcoming of this calibration procedure is that the
cameras have to be dismounted from the measuring
system if a new calibration is required. This situation
occurs if, for instance, either a complete system re-
calibration is necessary, a camera has to be replaced for
service purposes or if the environmental conditions have
been changed significantly. Therefore we have
developed a method for complete system calibration in
situ.
2.2 Complete System Calibration
The new method uses a calibration frame with 16
spatially distributed retro-reflective targets which is fixed
on the part-mounting system (Fig. 3). In conjunction with
8 additional points which are located on the mounting
base of the rotary table, a pyramid-shaped testfield with
a total of 24 targets has been built. This object covers a
large part of the measuring volume of the system.
Fig. 3. Testfield for complete system calibration
26
The approximate 3D coordinates of the control points
have to be known only to a few centimeters. In order to
improve the determination of camera parameters,
additional precisely known distances, in different
coordinate directions, have to be introduced [Wester-
Ebbinghaus 1985]. These distances serve also to
determine the global system scale.
The turntable is rotated to 8 different positions in angular
increment of approximately 45 degrees. In each position
the testfield is recorded by all cameras, leading to an
photogrammetric net of 24 images (see Fig. 4).
After bundle adjustment all parameters of interior
orientation are given with standard deviations better than
10pm, whereby the mean error (sigma 0) of adjustment
is equal to 1.5um. The unknown coordinates of the
testfield are calculated with an accuracy of < 0.04mm.
The coordinates of these points are used to fix the
machine coordinate system in which the exterior
orientation of the cameras and the rotary table will be
determined.
Using the calibrated values of interior orientation, as well
as the adjusted target coordinates in the zero position
(reference position) of the rotary table, a second bundle
adjustment is computed. In this step, independent
coordinates of target points are calculated for the other
seven positions. As a result, the exterior orientation of
the three cameras in the reference position are obtained.
In addition there are eight sets of 3D coordinates which
can then be transformed on to each other by a suitable
spatial coordinate transformation. The parameters of this
transformation define the position and orientation of the
axis of the rotary table within the machine coordinate
system.
a
reference
position
Fig. 4. Image network of system calibration
2.3 Re-Ori
Due to the
complete c
have to be
environme
mechanica
system ge
exterior or
periods. Tl
effort usinc
While the
operator :
blunder de
automatice
geometry
parameter:
2.4 Point N
The mea:
performed
element m
For a mea
taken of tl
process a
element L
and/or tab
camera w
images fro
For each «
in space
angie « o
located at
X'. With th
the object
camera in