RY 
Digital Elevation 
packages. It can 
|, using the PCI 
ips. Experiments 
ages, in order to 
been done of the 
de, estimated by 
n 100 meters. A 
was of 9 meters 
omatic filling of 
according to user 
roducts. 
iction techniques 
|, 2003). Time 
n conditions are 
ic matching. The 
3B (backwards). 
hree bands (1, 2 
imaging system 
f 4100 by 4200 
ed cartographic 
entation is well 
ed in order to 
| that express the 
se coordinates. 
POT or ASTER, 
ally some orbital 
ribe the sensor 
11, pitch, yaw) at 
vital perturbation 
ird, position and 
'he sensor model 
nt variations are 
- Kratky. (1987), 
r Toutin (1994). 
x from satellite 
nodels. 
an estimate the 
geo-location of 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
optical satellite images is usually in the order of 100 meters. For 
example, in the case of SPOT-4, the DORIS (Doppler 
Orbitography and Radiopositioning Integrated by Satellite) 
positioning system provides positions with an accuracy of 1 
meter or better (Spotimage, 2002). Therefore the geo-location 
uncertainty is caused by the absolute attitude. Anyway, a 
significant reduction is possible in the number of GCPs since 
only attitude parameters are needed. 
Alternative orientation methods are also possible. Dial and 
Grodecki (2002) describes the orientation of Ikonos imagery 
(also a linear array sensor) based on adjustments in image 
space. 
Commercial packages that deal with satellite images 
implement, in general, physical sensor models. That is the case 
of PCI OrthoEngine, used in the work described in this paper. 
However, the user does not have control on the parameters 
considered in the orientation or on the initial approximations. 
1.3 DEM generation with commercial software 
PCI Orthoengine was used in this work for the image 
orientation and DEM extraction. The process is very 
straightforward and can be carried out by unexperienced users. 
It consists of the following steps (PCI, 2000): 
1. Define the coordinate system. Usually, cartographic systems, 
such as UTM, are used. A local geodetic datum, different from 
WGSS84, can also be chosen. Users can enter their own datum 
by means of the 3 or 7 parameter transformations. 
2. Chose the sensor model to use. A global option can deal with 
different sensors including optical and SAR. Images are input 
preferably in their standard distribution format. This allows for 
the use of the initial orientation parameters, provided in the 
ancillary data. 
3. Enter the GCP data and apply the sensor model. A residual 
report is output. 
4. Generate epipolar images: in this step images are transformed 
so that parallax effect exists only in x direction. In the case of 
ASTER this step is essentially a rotation of 90 degrees since 
that, due to the across track stereo mode, disparities occur in y 
direction. 
5. Automatic DEM generation: the user defines the pair of 
epipolar images to use and the process is automatic. The only 
options respect to DEM spacing and detail and if holes are 
filled automatically by interpolation. No options exist explicitly 
related to the matching process, such as window size or 
acceptable correlation value. 
6. DEM geocoding: the initial DEM is in the epipolar image 
space. This step consists in projecting from image to map space. 
There is an option to fill gaps, by inerpolation, of holes 
resulting from the resampling process. 
DEM generation from ASTER is a well documented process. 
Results obtained by several authors refer a vertical accuracy of 
10 meters or better, especially in terrain with low vegetation. 
This study intends to assess, making use of a standard 
commercial software, how efficient can be the DEM extraction, 
the vertical accuracy that can be achieved and how small can be 
the number of ground control points. 
2. EXPERIMENTS OF DEM GENERATION 
2. Data used 
Some images of a region in north Portugal, near the city of 
Porto, were provided by the Japanese Space Agency. One of 
these images was used for the experiments of DEM generation. 
Figure 1 shows a map of the region, where bands 3N and 3B are 
represented. The region has heights ranging from sea level to 
1000 meters. 
A DEM produced by the Portuguese Army mapping service 
was used in order to assess the accuracy of heights derived 
automatically from the images. This DEM, with a spacing of 8 
m, was produced by photogrammetric means and is known to 
have an accuracy better than 2 m. The rectangle of the DEM is 
shown on the map of figure 1. 
| 
E: 
z 
| 
42°N 
  
| ATLANTIC | 
OCEAN 
    
PORTUGAL 
  
  
9°W 8°W 
Figure 1. Location of the 3N and 3B images (dashed polygons) 
and DEM (small grey rectangle). 
The DEM area (160 km2) has heights ranging from 30 to 600 
meters. Although the area is small relative to the image area, its 
relief is well representative of the whole area. 
2.2 Radiometric corrections 
Aster images of the level 1A show a stripping effect as can be 
seen in figure 2. Image (a) shows a sample of 100 by 100 pixels 
extracted from band 3N. 
In order to avoid possible problems in the stereo correlation this 
effect was corrected. Instead of using a filter, a different 
approach was implemented. The even and odd columns of the 
3N image were extracted and the histograms of these new 
images were calculated. The two histograms, represented in 
figure 3, are very similar and differ only by a systematic shift. 
  
Re - (b 
Figure 2. Sample of band 3N of the ASTER image: (a) original 
image, showing stripping, and (b) after correction. 
This shift was found to be of 7 units of greyscale value and it 
was added to the pixels of the even lines. Figure 2(b) shows the 
corrected image. In the 3B band the stripping effect was less 
significant and a correction of only 2 units was applied. 
However, experiments of matching images with the radiometric 
correction did not give any improvement in the matching 
success or in better accuracy, relatively to the results obtained 
with the original images.