control points
ers. In order to
a parameters,
, in different
duced [Wester-
serve also to
tions in angular
n each position
, leading to an
ig. 4).
ors of interior
ions better than
) of adjustment
rdinates of the
; of < 0.04mm.
1sed to fix the
1 the exterior
ry table will be
ntation, as well
e zero position
second bundle
0, independent
od for the other
r orientation of
jn are obtained.
ordinates which
er by a suitable
rameters of this
rientation of the
hine coordinate
pn
A
reference
position
X
»
v
alibration
2.3 Re-Orientation of the System
Due to the internal construction of the RSC sensors, a
complete calibration of the interior orientation does not
have to be done very often. In order to avoid effects from
environmental influences (temperature changes,
mechanical deformations) and to maintain a stable
system geometry, it is possible to determine only the
exterior orientation of the cameras within certain time
periods. This re-orientation can be performed with little
effort using only a few well-known control points.
While the complete system calibration initially requires
operator assistance (investigation of bundle results,
blunder detection etc), the re-orientation process runs
automatically. This procedure is used to check system
geometry and, if necessary, to improve orientation
parameters.
2.4 Point Measurement
The measurement of 3D points and elements is
performed by spatial intersection and object-based
element matching respectively [Loser & Luhmann 1992].
For a measurement a minimum of two images has to be
taken of the object element. With POM it is possible to
process an unlimited number of images for one object
. element using any combination of different camera
and/or table rotations. It is even possible to use only one
camera with two or more rotations in order to acquire
images from different spatial directions.
Fig. 5. Use of turn-table rotations
For each camera the system calculates a virtual position
in space K' (dashed) which depends on the rotation
angle œ of the rotary table. An object point P is then
located at P' and is projected into the image at position
x’. With this method the cameras apparently move round
the object, enabling 3D measurements with only one
camera in different rotations of the turntable.
The measured angle of the rotary table is corrected by
an internal correction functions which takes the adjusted
rotary positions of the second bundle adjustment into
account. For each rotation the system computes a set of
virtual camera positions with new exterior orientations
(Fig.5). With this method, an object point which is
measured in different rotations with one camera appears
exactly as it would if measured by the same camera,
virtually moved around the object (virtual position K').
This procedure offers great flexibility for the
photogrammetric measurement of complex shapes
where, for instance, occlusions require a large number of
different camera stations.
3. INVESTIGATION OF ACCURACY
In order to investigate system accuracy under practical
conditions, additional test measurements have been
carried out. On the one hand, the inner accuracy of the
sensors has been tested by a large number of repeated
measurements. On the other hand, calibrated scale bars
have been measured in different positions within the
measuring volume using different combinations of
sensors and rotary positions.
3.1 Repeated Measurements
The image coordinates of a retro-reflective target have
been measured with the target located in 1m
eccentrically from the rotary axis. In order to investigate
the stability of the RSC and the potential influence of the
CCD sensor movement, a total of 1100 measurements
have been done with each camera under the following
different conditions:
NM 300 measurements without sensor movement
SM 300 measurements, moving the CCD sensor to
an arbitrary position in all directions (x, y, focus)
and returning to the original position
RO 100 measurements, leaving the CCD sensor in
constant position but rotating the turntable by
about 90 degrees and returning to the original
position
SR 100 measurements, moving both sensor and
rotary table according to SM and RO
LV 300 measurement with constant sensor
positions, with temporary variation of light
intensity
Fig. 6 shows the measurement results. For each camera
an average of the 1100 observations has been
calculated, as well as the resulting deviation of each
measurement value. The illustrations show the maximum
differences between single measurement values and
average values under different measuring conditions,
tabulated separately in x and y. It can be proven that all
deviations are under O.9um in image space, which is
comparable to the internal accuracy of the RSC
according to calibration results. With respect to a mean
image scale of 1:30 an accuracy of repeated
measurements of less than 30um in object space is
achieved.
