shifted tie point areas 
  
  
Figure 8: Tie point areas before initialization 
  
  
  
  
  
Figure 10: Block DEM for 960 um pixel size 
with a multiple image overlap of up to 15 images. 
Ground control points (# 16) as well as check points (# 
16) were signalized as circular targets which had a size 
of 3 - 4 pixels in the image. The block served in first 
instance to test the practical applicability of GPS OTF- 
ambiguity solutions (Ackermann, 1996a). GPS 
observations were hence available for the exposure 
centers which were derived by the software package 
SKIP (Friess, Heuchel, 1992). 
Initial parameters for the kernel system were determined 
from the strip azimuths, the GPS antenna observations, 
and an average terrain height. The system created a 
large number of tie point areas per image because of the 
large multiple-image overlap. After the initialization the 
automatic block adjustment was carried out with the 
kernel system in self calibration mode, whose results are 
summarized in table 2. 
Let us firstly focus on the terms indicating the 
redundancy effect of the MATCH-AT approach. 
410 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
correct tie point areas 
  
Figure 9: Tie point areas after initialization 
  
     
         
      
     
Se S 
ZZ SAN CSS 
NENNEN ITE 
S EM TT, AS QS 
CERA S CSSS 
5S SS e SC 
SS SS 
RSS OS SOS 
Y EN S S S Neh shed SACS cS CD 
SES acies REST CO SOS Sn 
SSMO iOS REEC EE AUG 
3 IAS 
a 
  
  
  
Figure 11: Block DEM for 480 um pixel size 
Especially in the 30 um level, a huge number of 590 
points per image was matched and transferred. The 
corresponding value for 15 yum pixel size amounts to 490 
points per image which is practically the same order of 
magnitude. Such numbers are in general very 
characteristic for automatic aerial triangulation which are 
based on image processing techniques. They are the key 
to highly accurate and reliable results and can by far not 
be attained with conventional aerial triangulation. The 
next parameter to be discussed is the sigma naught 
value which represents the measuring precision of the 
feature-based matching technique. Its value of 0.3 pixel, 
for 30 jum pixel size, is very close to the theoretical 
expectation, whereas for 15 um pixel size it ends up at 
0.5 pixel. The corresponding values in terms of image 
units are 9.4 um and 7.5 um. Those values compare well 
with many cases of conventional aerial triangulation with 
non-signalized points. However, we have to keep in mind 
the large redundancy and the number of transferred tie 
points per tie point area, resp. Since the system provided 
on average 18.7 points per tie point area in the 30 um 
      
   
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
   
    
  
  
  
  
  
  
  
  
     
       
   
    
     
    
   
   
    
   
   
    
   
   
   
   
    
   
   
   
       
      
   
    
pyram 
assur 
to 9.4 
triangt 
the sa 
sigma 
point c 
terms 
and n 
precisi 
propa 
values 
table 
tilts ai 
omeg: 
corres 
better, 
coordi 
excell: 
and Y. 
pixel : 
slighti 
20.1 € 
empiri 
values 
it can 
triang! 
um pi 
result: 
is to b 
are nc 
the se 
reduc: 
points 
match 
signifi 
work \ 
3.2.2 | 
This b 
3 stri[ 
scale 
and c 
points 
GPS i 
GPS : 
The i 
initiali 
centel 
startir 
was 
sumir 
As fe 
numb 
numb 
per G 
case 
This c 
naugk 
value 
pixel 
8.8 pi