of the fiducial 
mum influence 
oint using the 
im to the pixel 
affic light: 
nel (22) 
| to inform the 
tions. 
  
  
ial mark 
ark. The black 
ducial mark. 
fiducial mark 
ensitivity anal- 
single fiducial 
an observation 
in error. Fidu- 
cs, is the one 
NMENT 
tween the In- 
nd Carl Zeiss, 
art of the Dig- 
HODIS (Pho- 
1). 
company uses 
ly work with a 
her automatic 
or autonomous processes like ARO (Automatic Relative Ori- 
entation) and PHODIS AT (Automatic Aerotriangulation), 
which are based on the results from the AIO. Therefore it is 
very important that the process works robust and reliable. 
The AIO has been in use for quite a while and been tested 
on different cameras (RMK TOP, RMK A, LMK and RC) 
respectively images with different resolutions and quality. The 
results confirm the high reliabilty of the approach and also the 
internal decesive measure for selfdiagnosis. 
The succcess rate with respect to the selfdiagnosis is summa- 
rized in the following table: 
  
  
  
  
  
  
  
  
  
  
Sensitivity analysis 
Reality correct | incorrect 
correct 49 | (I) 0 
incorrect (iy^ "g 3 
Table 2 
94% of the interior orientations have been correctly found to 
be good (green cases). In 6% the system has correctly de- 
tected a failed orientation (red or yellow cases). Erroneously 
detected as bad (type | error) and the most expensive type | 
error, a failed orientation classified as beeing correct, has not 
occured. 
The computational time for an automatic interior orientation 
including the orientation determination on an aerial image 
with 8 fiducial marks is about 20 sec on a SGI Indigo2. The 
accuracy of a single fiducial mark measurement is about one 
tenth of a pixel. The sigma nought of the transformation 
estimation, which is a quality measure of the estimation, is 
obout 0.2 pixel. This is 3m at pixel resolution of 15 um, and 
therefore comparable with a high precission manually mea- 
sured interior orientation. 
Finaly we could say, the AIO is an autonomous process, ready 
for use in production. 
REFERENCES 
[BAARDA W. 67] Statistical Concepts in Geodesy, Nether- 
lands Geod. Commision new series, Vol. 2, no. 4 Delft, 
1967 
[BAARDA W. 68] A testing procedure fore use in geodetic 
networks, Netherlands Geod. Commision new series, Vol. 
2, no. 5 Delft, 1968 
[BorLEs R. C., FrsHLER M. A. 81] Random Sample Con- 
sensus: A Paradigm for Model Fitting with Applications 
to Image Analysis and Automated Cartography, Comm. of 
ACM, Vol. 24, no. 6, 1981 pp 381 - 395 
[BRÜGELMANN R., FÓRSTNER W. 92] Noise Estimation 
for Color Edge Extraction, in FÓRSTNER/RUWIEDEL 
1992, pp. 90 - 106 
[HARALICK R. M., SHAPIRO L. G. 93] Computer and 
Robot Vision, Vol Il, Addison-Wesley, USA 1993, p 316 ff 
[FORSTNER W. 83] Reliability and Discernability of Ex- 
tended Gauss-Markov Models, DGK A98, Munich, 1983, 
pp 79 - 103 
[FORSTNER W. 92] Quality of Vision Algorithms, tutorial 
notes, Bonn, Institute of Photogrammetry, 1992, pp 30 
- 40 
751 
[FORSTNER W., RUWIEDEL ST. (Ed. 1992)] Robust Com- 
puter Vision — Quality of Vision Algorithms, Wichmann, 
Karlsruhe, 1992 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996