ositions
n of the
aneous
onds to
| shown
icy are
on
st field
n order
> points
t
As the off-line systems Large Format Camera and
Rolleiflex 6008 metric are to be included in the
comparison, the use of retro-reflecting material is
imperative. The projection of these points, which do not
have an ideally defined circumference, through the
object lens onto analogue film material works like a low-
pass filter. However, irregularities in the retro-signals are
detectable in the case of direct projection of the points
onto the CCD-sensor of the RSC, and this reduces the
accuracy of point determination. This effect can be
largely corrected by filtering around the measured points
with a 3 x 3 low-pass filter. The same point measuring
algorithm is used with all three photogrammetric
systems. It is based on the convolution of the image
function with a gradient operator. It could be shown in
various applications that this algorithm is suitable for
accuracy of up to 3/100 of the image elements
[Bósemann et al. 1990]. The proposition that the ideal
centre of the object point and of the measured centre
can be equated holds true for the initial approximation,
and is sufficient for the evaluation required here.
4. INTERPRETING THE RESULTS
Besides consideration of accuracy and reliability,
derived from excess mesurements a direct comparison
of several measuring systems allows an insight into the
accuracy of a system.
The essential results of the bundle adjustment are
summarized in table 1. Initially, all measured directions
and image coordinates, together with the respective
apriori standard deviations are adjusted together within
the so-called free net. The object point coordinates of
this adjustment, considered as datum points, are the
basis for the individual adjustments. After individual
bundle adjustments some deviations on the datum
points exceed the respectiv three-fold standard
deviations. The fact that some difference vectors lie
outside the three-fold standard deviation indicates that
the standard deviation calculated in the free net is a
value which must be handled with deliberation.
Furthermore, systematic tendencies in the difference
vectors between combined adjustment and the
individual adjustments are considerable. Thus the
vectors near the edge of the test field are larger than
those in the centre.
On account of its large image format, the LFC attained
the highest accuracy at the testfield. The results of the
theodolite measurements do not quite reach this level,
as, with the combination of the two theodolite system
used here, the configuration of image acquisition
indicated in Fig. 5 could not be met exactly.
Furthermore, only 6 theodolite positions were taken into
account here, as opposed to 10 recording stations on
the photogrammetric systems. Evaluation of mid-size
images produces, compared to the RSC, which has the
same format, object coordinates with inferior standard
deviations on the object, in spite of an approximately
equal image measurement accuracy. While the
Rolleiflex 6008 images could be tilted up to 35 gon due
to the depth of field problems, the test field for RSC
images could be tilted up to 75 gon to the test field level
exploiting the focus function.
residuals at image image relative mean standard mean deviations
measuring- the image format scale measurement deviations of the at the datum points
system coordinats (RMS) accuracy object points (RMS) (RMS)
in the image
[um] [mm] [um] [um]
X y X y z X y z
combined
bundle - - - - - 2.7 2.8 5.4 - - -
adjustment
SPACE 0.38* 0.40* - - - 5.9 6.4 12.2 3.2 4.8 12.0
Large Format 1.13 1.21 230 x 230 1:7 1:200 000 2.8 2.7 5.9 3.2 4.2 5.7
Camera (LFC)
Rolleiflex 6008 0.91 0.97 52 x 52 1:30 1:55 000 8.5 7.9 14.8 17.1 18.4 19.2
metric
Réseau - Scan- | 0.86 0.78 52 x 52 1:30 1:60 000 7.1 6.5 94 10.3 13.7 15.1
ning Camera
* measurement accuracy of directions [mgon]
Table 1: Results of combined and individual bundle adjustments
