« . Measure manually all remaining tie/control points on
one image. Auto-measurement will transfer all
points to the other image.
e If auto measurement fails, remeasure manually the
point on the second image.
b) Measuring Subsequent Models in a Strip
e Set proper auto-measurement flags on.
« Measure manually center points of stereo pair.
e Transfer automatically points already measured
down the center of the left image to the trailing edge
of the second image.
e Measure manually all remaining tie/control points on
one image. Auto-measurement will transfer all
points to the other image.
e If auto measurement fails, remeasure manually the
point on the second image.
e Repeat this process for all remaining models in the
strip.
Strips were tied together in a 4 or 6 image mode, viewing
2 or 3 images from each strip. On-line bundle
adjustment was performed in relative mode to check the
quality of the measured tie points. Blunder detection was
used for on-line checking of model and strip connections.
3.3 Bundle Block Adjustment
Bundle block adjustment was performed on the Texas
project using all photogrammetry measurements and 9
full control points with the assumed standard deviation of
1 cm. Seventy-one signalized check points were used
for empirical accuracy estimation. The estimated
precision of the image coordinates (G,) was 3.4 um, and
the empirical accuracy indicators of the check points
were about 1 cm. in XY and 2.9 cm. in Z (Table 2).
The MATCH-AT program was invoked for the automatic
triangulation of the Texas project using the same number
of control and check points. Necessary parameters and
approximate locations for tie point areas were created by
the ISDM product. MATCH-AT was run on the Silicon
Graphics Indigo R4000 workstation using both 15 um
and 30 um pixel size imagery. The number of tie points,
especially at 30 um, was quite large. This gave high
redundancy and hence improved accuracy and reliability.
Most of the tie point areas were located on a runway of
the Texas project with a low contrast. Therefore, the
feature-based matching technique provided less matched
points for the 15 um data set than the 30 um imagery
(Table 2). MATCH-AT automatically eliminated these
erroneous observations which occurred due to poor
texture. The estimated precision of the automatic digital
triangulation was 8.8 for 30 um and 6 um for 15 um
imagery, respectively. These O, values were very close
to the theoretical values obtained in the conventional
aerial triangulation using natural tie points (Ackermann,
Tsingas , 1994).
The theoretical results were excellent, especially for the
30 um case. The standard deviations of object points
were equivalent to 1.1 O for XY and 2.5 O, for Z. Also,
the standard deviations of the exterior orientation
parameters were excellent compared to the ISDM
triangulation results. The main reason for such a high
accuracy is the large redundancy of the feature-based
494
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
automatic aerial triangulation. The mean standard
deviations of the object point coordinates (theoretical
values) corresponded quite well with the RMS values at
check points.
Very good results for the theoretical accuracy and the
empirical accuracy were also obtained with the 15 um
case. lt is remarkable that for 15 um no gain in accuracy
is attained, although the ©, value is better. The
reduced redundancy of the 15 um case is due to the low
texture which causes decrease in matched and
transferred points. A least squares matching might
improve the precision and the theoretical accuracy.
Again, the mean standard deviations corresponded
reasonably well with the RMS values.
MATCH-AT Results | ISDM Results
Pixel Size (um) 30 15 15
Matched Point Area 15.0 10.9 1
No. of Points/Image 490 120 15
O, (um) 8.76 6.0 3.4
O, (pixel) 0.29 0.4 0.23
Empirical Accuracy
No. Of Check
Points
XY 71 71 71
Z 71 71 71
RMS X (cm) 1.2 1.3 0.8
RMS Y (cm) 1.0 1.1 1.3
RMS Z (cm) 5.1 2.9 2.9
Theoretical Accuracy
Sigma X (cm) 2.9 2.4 1:3
Sigma Y (cm) 2.7 2.9 1.2
Sigma Z (cm) 6.5 5.1 2.4
Exterior Orientation Parameters
Sigma X (cm) 4.4 4.7 3:2
Sigma Y (cm) 2.4 4.2 3.6
Sigma Z (cm) 3.2 3.0 2.2
Sigma o (mgon) 6.5 7.1 54
Sage © (ngon 7.6 6.9 3.6
Sigma x (moon 24 3.0 2.2
Table 2. Texas Project Bundle Adjustment Results
Concluding, it is to be stated that the MATCH-AT results
for the 30 um case are better than those which could
have been attained with ISDM using only a 30 pm pixel
size. In a theoretical scenario, the estimated precision of
the semi-automatic digital triangulation using the 30 uum
pixel size would be two times larger than the
corresponding value of the 15 um data set. This is, the
©, value would amount to about 6.8 um. Also, the
results of ISDM mainly refer to the signalized points.
Therefore, it is expected that the ISDM results using non-
signalized points would end up in a O, value of about 6
to 8 um. Thus, from that point of view, the MATCH-AT
results of the 30 um case are equivalent or even better
than those obtainable with conventional aerial
triangulation using non-signalized points. The main
reason that MATCH-AT did not fully reach the excellent
results of ISDM with 15 pm pixel size was the drastic
loss of the redundancy due to the poor texture. A
remedy for such cases is to use a least squares
matching method, which gives best measurement
precisi
AT sys
The E
Maryle
the st:
were |
from
photoc
the M/
Sizes ©
time c
digital
ISPRS
The c
digital
minute
minute
signal
autom
nature
Graph
image
perfor
Image
4
The a
on th
Featul
match
The
match
simila
conve
match
exteri
gener
config
This f
a ven
This [
and 1
creatii
used
of ra
contrc
the tie
accur
exist
DTM
appro
photo
build-
The c
navig
a co
locati
In the
integr
levels
adjus
