International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
2.2.3. Autotriangulation (using a 5x5 pattern of vonGruber
point positions): In this project the same images and the same
parameters were used. The only difference was in the
initialization. Using a 5x5 pattern of vonGruber point positions
the tie points areas increase in number and density.
The comparison between the manual solution and the AAT
solution for the 1:3500 images showed that (a) the RMS value
of control points and the standard deviation are of the same
order and (b) a lower a-posteriori value for o, is observed in
AAT (Table 3). The feature based matching approach provides
very accurate and reliable exterior orientation parameters due to
its capability of generating many redundant tie points due to the
strong geometry configuration. The computational time was
about 3 minutes per image and the o, value was 0.30-0.35 pixel,
given that the system used had four Pentium Pro 200MHz
processors, with 128MB RAM and Ultra SCSI HDD. Today on
a modern system built around two Pentium IV Zeon 3.2GHz
processors, with 2GB RAM the time required would be less
than 1.5 minutes per image.
2.3 Images of scale 1:6000
The 1:6000 scale images are deployed in two strips with
absolute flying height of about 1300m. The approximate values
of the exterior orientation parameters were given with an
uncertainty of 100m.
2.4 Images of scale 1:8000
The 1:8000 scale images formed just one strip with absolute
flying height of about 1500m. The approximate values of the
exterior orientation parameters were given with an uncertainty
of 100m. In this set of images a second analysis step was
carried out in order to assess the accuracy of the obtained
orientation parameters independently. Using interactively
additive measured coordinates of GCPs as observations and the
obtained tie points coordinates as GCPs in the robust bundle
adjustment of the first analysis step as constant values, over-
determined least-square forward intersections were computed.
Obviously the solution gave better accuracy.
2.5 Analytical Aerial Triangulation
Finally, a comparison of the above results (AAT) with those
derived from an analytical solution of the same area and using
the same images has also been carried out. In the analytical
project several of attempts were made with the intention to
conclude on essential and thought provoking results. The point
measurements on the diapositives were carried out on an
analogue Zeiss Jena Technocart D, converted to analytical with
the ADAM conversion system. The GCPs used were the same
as those used in the Automatic Triangulation. The standard
deviation for the photo measurements was set to 10 pm, the
Omax 0f the interior orientation to 10um, the 6,4, of the relative
orientation to 10 jum and the 0,4, of the absolute orientation to
10pm. The bundle adjustment for the aerial triangulation was
performed with the BINGO-F software (Zervas & Papaharisi,
2000).
3. RESULTS AND ANALYSIS
3.1 Multi-ray tie points
In contrast to conventional aerial triangulation, ISAT’s
matching strategy determines tie point clusters instead of single
points. The measuring precision is directly influenced by the
preferred matching technique and can be maximized by the
LSM technique up to 0.1 or 0.2 pixels. The matching strategy
also leads to measuring fully transferred tie points. If this goal
cannot be reached, ISAT accepts only the best matches,
avoiding 2-fold points. Tie point clusters are well distributed
over the entire image with preference to the von Gruber
positions as the most effective positions for tie points.
Project Scales # of tie
points
1 Manual 1:3500 -
2 ISAT 1:3500 (5x5 von Gruber pattern) 1911
3 ISAT 1:3500 (3x3 von Gruber pattern) 891
4 ISAT 1:6000 869
5 ISAT 1:8000 818
6 ISAT combination 1:6000 — 1:8000 2199
7 ISAT combination 1:3500 — 1:8000
(it stops at the sixth pyramid level)
Project # of multi-ray tie points observations
2x 3x 4x 5x 6x | 7x 8x
1 = z 5 = 3 - .
2 247 | 349 39 26 14 - -
3 50 128 26 33 23 - -
4 109 | 134 41 17 - - -
5 148 | 174 - - - - -
6 207 |-200 | 125 70 Si 3 4 1
7 -
Table 2: Projects and multi-ray tie points after AAT
3.2 Analysis of results
The comparison between ISAT solution with 3x3 and 5x5 von
Gruber points respectively, using different number of patches,
shows similar accuracy. However, in the case of 5x5 pattern an
improvement in the residuals was observed. That was due to the
avoidance of gross errors. Consequently, in the 5x5 pattern the
blunder errors were avoided.
Project [A RMS (m) at the GCPs
pixels | pm X Y Z
1 5.2
2 0.35 4.9 0.043 0.078 0.047
3 0.35 4.9 0.044 0.073 0.055
4 0.44 6.1 0.049 0.077 0.079
5 0.31 4.3 0.009 0.011 0.018
6 0.54 7.6 0.064 0.136 0.148
g 4.23 59.2 | 0.017 0.029 0.017
Table 3: Results of bundle adjustment for GCPs
Inter
In T
(botl
arep
For :
is no
the
obtai
ordei
stabi
The.
resul
succi
mad:
wind
the r
inter;
impr
The
prov
diffe
diffe
Thus
to s
DEN