International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV. Part B2. Istanbul 2004
2. AEROTRIANGULATION
2.1 Test data
To achieve reliable conclusions, high accuracy reference data is
required. For one area of investigation, a triangulation was
measured on an analytical plotter Wild S9 using 26 aerial
images from 8 different strips. The acquired image coordinates
of the control and tie points, measured according to the von
Gruber distribution, then served as input data for the bundle
adjustment package BUN. As a result, a o of 9.65 microns
could be achieved for the processed subblock as global
accuracy. Compared to the values we achieved for the two
blocks of Palpa imagery (scale 1:7500), o» — 13.3 microns and
oo = 9.5 microns, the global accuracy of 9.65 microns can be
considered as a good result.
The second test area consists of two trips with 26 and 28
images, respectively.
2.2 Z/l Image Station
The workflow of automatical point measurements in Image
Station Digital Mensuration (ISDM) is divided in two main
parts: the relative orientation, where points are measured in
stereo models, and multiphoto orientation, where points in more
than two images can be measured. It is also possible to start
with multiphoto orientation directly. The control points have to
be measured manually in at least one image using the absolute
orientation menu item, and can then also be transferred to other
images using multiphoto orientation.
Automated measurement of tie points for relative orientation
requires initially manual measurement of two points in each
stereo model before the automated process can be started
successfully. The automated transfer of points between images
of different strips is possible, and can be used to measure the tie
points acquired during the relative orientation in corresponding
images of the neighbouring strips. For the automated matching
of both, relative orientation and multiphoto orientation, a patch
size of 11x11 pixels was used, combined with a number of 5
points maximum to be measured at each of the given 3 positions
per image (6 per model). These positions follow the Gruber
distribution, but can also be altered by the user to any other
distribution. The correlation threshold, the value each
measurement has to achieve to be accepted, was chosen to 0.95.
After performing the automated measurement with these
parameters in relative orientation mode, it could be observed
that at each position at least onc point could be measured,
usually more. Except for one model, the average parallaxes of
the models were smaller than 10 microns. Afterwards, using the
multiphoto orientation with automated point transfer, bundle
adjustment could not be performed successfully due to the large
parallaxes especially for points in the overlapping areas of strips
measured in more than two images. The same effect occurred in
a second effort with manually measured tie points for strip
connection. To clarify the reason for this exactly, some more
investigations have to be done. Better results were achieved
triangulating the first two strips of 54 images of the Nasca
block, where manually measured tic points were used for strip
connection. Figure 2 shows the distribution of the automatically
measured tie points. In this project, points at almost every given
position could be matched although especially strip 2 contains
predominantly desertous areas, while some images contain
agricultural crop land, which also can be classified as lowly
textured. The position where matching failed significantly is
marked in figure 2. As a global accuracy, 21.1 microns could be
achieved after bundle adjustment, which is less than scan pixel
accuracy.
Figure 2: Distribution of automatically measured tie points
3 y
using strip 1 and 2 (54 images)
2.3 Virtuozo 3.1
The test concerning aerotriangulation was performed using
Virtuozo 3.1. At IGP, a license including aerial triangulation is
available only for this version. In Virtuozo 3.1, no manual
measurements are required for the initialisation of the stereo
models, but for tie points between the strips. In Virtuozo, like in
Image Station, the user can influence the distribution of the tie
point positions to be measured. Unlike in Image Station,
Virtuozo supports Gruber and similar distributions and offers
the user to choose the number of points to be measured at each
position. Further parameters like mentioned above can not be
changed. The result of the automated tie point extraction in
Virtuozo (see figure) shows, that at several positions no tie
points could be extracted. Performing bundle adjustment in
PATB, a a, of 6.97 microns was achieved. This value can not
be compared directly to the result obtained from the analytical
plotter because it is too optimistic due to the fact, that only
points of high quality are included into the bundle adjustment,
the weak points are completely ignored at the positions where
gaps occurred (figure 3). Thus, the calculated o» does not
consider instabilities of relative orientations resulting from
missing tie points. In figure 3, red ellipses point up the areas
where a significant accumulation of gaps in the tie point
distribution appeared, compared to the manually measured
points.
For the second subset of Nasca imagery, tie point extraction
succeeded only for strip | although manual measurements for
strip connection had been accomplished in each of the
overlapping images. The mode for measurements of tie points
for strip connection does not provide a zoom function, therefore
the manual measurements were probably not exact enough for
successful point matching.
For this reason, no directly comparable results could be
achieved for automatic aerotriangulation, the presented results
can only be taken as preliminary results of qualitative character.
Inter
Figu
3.1
In tl
accu
gene
analy
XM/
paral
point
of 2
exce
with
The
direc
cala
From
error
Potei
spati
distri
grids
exaır
The
pred
mode
whic
For
Stati
plotte
contr
used
marg
Geor
is
The
Auto