International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
The mother process of a specific project is a shell script running 
on a server. It includes the access to the raw data, the 
initialization of all client processes and the final product 
generation. Project parameters are defined within a set-up file 
the content of which can be adapted to the specific project 
requirements, e.g. the definition of datum shift parameters to the 
requested geodetic datum. Client processes are also shell scripts 
containing the single programs for image matching, ortho-image 
generation etc. 
5. PROCESSING SYSTEM EFFICIENCY 
Table 1 gives an impression of the net processing time for an 
exemplary HRSC-AX project with the following parameters 
(loading a 50 CPU-cluster to approx. 5096): 
Project size: 500 km? (20 km x 25 km) 
Flight altitude: 4,000 m above ground 
Ground resolution: 17 cm/pixel 
Swath width: 1,800 m 
Side lap: 50 96 
No. flight strips: 20 
Strip length: 25 km 
A product set consisting of 20cm/pixel ortho-image mosaics in 
true-color, color-infrared, black/white, and black/white infrared, 
as well as 1m-gridded DSM is assumed. 
  
Generation of coarse DSM within 1 day 
(5 m-grid) 
Generation of final DSM 4 - 7 days 
(1 m-grid) 
Generation of all ortho-image mosaics 2 - 3 days 
(20cm / pixel) 
  
  
Table 1. HRSC Processing efficiency 
Thus, even including common time add-ons (for data handling, 
controls, data transfer, tiling, etc.), project areas of hundreds of 
square kilometers can be processed at high resolution within 
few weeks. Note also that, due to parallel processing, the 
number of image strips of a project has only a secondary effect 
on the overall processing time (as long as enough clients are 
available). Only project-wide processing steps (such as 
mosaicking, DSM generation) are directly extended. 
6. SYSTEM ACCURACY, ROBUSTNESS, AND 
COMPATIBILITY 
Within most HRSC projects, investigations of the accuracy of 
the data processing and products confirmed the high quality of 
all data used for direct geo-referencing (interior and exterior 
orientation, measurement/matching of image coordinates). From 
flight altitudes of 3,000-4,000m, the relative accuracy for 
forward ray intersection of up to 5 stereo observations within an 
image strip was found to be 10-15 cm for selected points 
measured manually, and 15-20 cm for area-based image 
matching results. Strip-to-strip discrepancies typically range 
from 15 to 20 cm. The overall mean absolute point accuracy is 
about 15-25 cm (3D RMS, I sigma). Compared to these values 
the point accuracy at steep edges (e.g. buildings, bridges, etc.) is 
decreased when these discontinuous objects can not be 
modelled with sufficient accuracy by stereo observations 
because of dissimilarity of the stereo image data. 
  
Figure 5. Robustness of geometric correction 
(left: raw data, right: rectified data) 
The robustness of the processing system is clearly illustrated by 
an extreme data set (Figure 5), acquired when the stabilized 
platform, to which HRSC was mounted to, bumped during 
flight because of an excessive stalling angle of the aircraft. The 
frequency of this effect of a few Hz causes abnormal distortions 
(far away from the nominal case) in the raw image data. 
Nevertheless, the permanent and precise high-frequency data 
provided by the inertial measurement unit enabled a nearly 
perfect reconstruction of the scene automatically without any 
operator interaction. 
  
Figure 6. Subsets of ortho-images and digital surface models 
derived by DLR's processing system for multi-line scanners 
top: HRSC-AX, Potsdam, Germany, flight altitude 4,000m, 
ground resolution 17 cm 
bottom: ADS40, Nimes, France, flight altitude 800m, 
ground resolution 8 cm 
410 
Inter! 
  
The 
HR 
ade 
bas 
orth 
pro 
The 
the 
Not 
diff 
  
Subs 
appli 
follo