anbul 2004 
ınd MISR 
shop "High 
Hannover, 
he rational 
processing. 
5, Vol 67, 
ation from 
tical 3-D 
3 Zurich, 
ation from 
B3 (to be 
  
  
ACCURACY ANALYSIS FOR DEM AND ORTHOIMAGES DERIVED FROM SPOT HRS 
STEREO DATA WITHOUT USING GCP 
Peter Reinartz, Manfred Lehner, Rupert Müller, Manfred Schroeder 
German Aerospace Center (DLR), Remote Sensing Technology Institute 
PO Box 1116, D-82230 Wessling, Germany 
peter.reinartz@dlr.de 
Commission 1/2 
KEY WORDS: Stereoscopic data, image matching, orthoimages, DEM generation, DEM fusion, accuracy analysis 
ABSTRACT: 
ISPRS and CNES announced the HRS (High Resolution Stereo) Scientific Assessment Program during the ISPRS Commission I 
Symposium in Denver in November 2002. 9 test areas throughout the world have been selected for this program. One of the test sites 
is located in Bavaria, Germany, for which the PI comes from DLR. For a second region, which is situated in Catalonia — Barcelona 
and surroundings — DLR has the role of a Co-Investigator. The goal is to derive a DEM from the along-track stereo data of the SPOT 
HRS sensor and to assess the accuracy by comparison with ground control points and DEM data of superior quality. For the 
derivation of the DEM, the stereo processing software, developed at DLR for the MOMS-2P three line stereo camera is used. As a 
first step, the interior and exterior orientation of the camera, delivered as ancillary data (DORIS and ULS) are extracted. According 
to CNES these data should lead to an absolute orientation accuracy of about 30 m. No bundle block adjustment with ground control 
is used in the first step of the photogrammetric evaluation. A dense image matching, using very dense positions as kernel centers 
provides the parallaxes. The quality of the matching is controlled by forward and backward matching of the two stereo partners 
using the local least squares matching method. Forward intersection leads to points in object space which are then interpolated to a 
DEM of the region in a regular grid. Additionally, orthoimages are generated from the images of the two looking directions. The 
orthoimage and DEM accuracy is determined by using the ground control points and the available DEM data of superior accuracy 
(DEM derived from laser data and/or classical airborne photogrammetry). DEM filtering methods are applied and a comparison to 
SRTM-DEMSs is performed. It is shown that a fusion of the DEMs derived from optical and radar data leads to higher accuracies. In 
the second step ground control points are used for bundle adjustment to improve the exterior orientation and the absolute accuracy of 
the SPOT-DEM. 
1. INTRODUCTION 
The derivation of digital elevation models (DEM) from along 
track stereo data from space has up to now only been possible 
with the German MOMS-2P (Müller et al. 2001) and the 
Japanese/American ASTER sensor on TERRA (Toutin et al. 
2001). Both of them have lower resolution (15-18 meter pixel 
size) than the new HRS sensor on SPOT 5. HRS produces 
image stereo pairs with two optics looking forward and 
backward (£20 degrees) with respect to the nadir direction. The 
camera has a spatial resolution of 10 meter across track and 
along track, but a ground sampling distance of about 5 m along 
track for obtaining higher resolution of the parallaxes for the 
DEM generation. The swath of the HRS is 120 km (12000 CCD 
elements) and one acquisition sequence extends 600 km along 
track. 
After the ISPRS Commission I Symposium in Denver in 
November 2002, the HRS Scientific Assessment Program has 
been established. This program gives the user community the 
opportunity to test HRS data, which are usually not available, 
for generating DEM and for comparison with other DEM 
generation methods. Further it should provide CNES an 
international scientific performance assessment of the HRS 
which will be taken into account for future programs. For the 
investigations 9 test areas around the world with corresponding 
PIs and co-investigators have been selected. Only those areas 
have been selected, where the PIs could provide a sufficient 
data set of ground control points and a high precision DEM for 
comparison and accuracy checking of the derived HRS-DEM. 
The data which have been provided by SPOT IMAGE contain 
the following parts: 
e 2 sets of 8 bit panchromatic image data (size 12000 x 
12000 pixel = 120 km across x 60 km along track) of the 
Bavarian and Catalonian test area from two viewing 
directions in TIF format 
e in the Catalonia area as well HMA and HMB data: 5 mx 5 
m resolution panchromatic nadir looking channels of part 
of the test site (size 12000 x 12000 pixel = 60 km across x 
60 km along track ) 
e  XML-files containing all additional information regarding 
time synchronization, position (DORIS), attitude (star 
sensors and gyros), interior orientation 
e text files containing information on the delivered data. 
The data of Bavaria have been acquired on October 1" 2002 
with a sun elevation of 38? and nearly no clouds. The data of 
Catalonia have been acquired on October 15" 2002 with a sun 
elevation of 39? and no clouds. The radiometric quality of the 
Catalonia images is superior to the Bavarian imagery. 
The HRS-data have been received in July 2003. First results 
have been presented at the ISPRS Workshop High Resolution 
Mapping from Space, Hannover '03 (Reinartz et al. 2003).