inbul 2004 
stereo-pair 
tions was 
ie 6 GCPs 
n. In these 
m) and the 
ation. This 
lace of the 
generated 
reference 
6Km. The 
2 
RMS 
(m) 
14.8 
14.8 
en 
N 
14.8 
  
  
  
extracted 
regards to 
he area of 
reference 
  
DEM is presented while in Fig. 4b the corresponding region of 
DEMI is given as well. Figures 5a and 5b show a part of the 
above DEMs. 
  
Figure 3. A part of he produced DTMI 
  
  
Figure 4b. The corresponding to reference DEM, region of 
DEMI 
  
Fig 5a. A part of the reference DEM 
  
Fig. 5b. The corresponding region of DEMI 
The statistical analysis of results (table 2), concerning the 
produced DEMs, showed that all the solutions are acceptable 
for flat and even regions, while in the highly mountainous areas 
the precision is small. For example, in the basic solution (a), 
while the value of the min and max error are calculated at - 
[19m and 166m, the value of the mean absolute error is 7m. 
467 
    
     
      
   
   
   
  
  
  
  
  
  
   
   
   
  
  
  
  
   
  
   
  
  
   
  
  
  
   
  
  
  
   
  
  
  
  
  
  
   
  
  
  
  
  
  
   
  
  
   
  
   
    
    
   
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part BI. Istanbul 2004 
The DEMI was also checked manually by overlapping the two 
DEMs. The DEM quality certainly depends on terrain type and 
terrain roughness. Thus in plain areas the achieved accuracy 
was 0-5 m while the maximum height differences have been 
occurred in mountainous and steep areas. In Figure 6 is 
presented a part of the difference image between DEMI and 
DEM reference, for the same area. 
e 
  
Figure 6. A part of the difference image between the DEMI and 
the reference DEM 
Negative errors in elevation are indicated in black and that 
means that the DEM] has a smaller altitude than DEM. On the 
contrary, positive errors are indicated in white. This occurs in 
the areas where the produced DEMI has higher altitude than 
DEM. Areas with a difference of 0-5m are indicated in grey. 
3. CONCLUSIONS 
DEM generation from satellite data set is a fact since new- 
launched satellite images have been used to provide 
stereoscopic images. This paper reports experiments carried out 
for automated extraction of a digital elevation model (DEM) 
from HRS data of SPOT-5 satellite, based only on the Metadata 
provided by the sensor. Neither GCPs nor any kind of digital 
map were available. The absence of Ground Control Points led 
us to test, except the basic solution, other procedures. These 
tests concern  “semiautomatic” solutions in which a 
triangulation technique with automatic or semiautomatic 
methods of determination the tie points has been applied before 
the generation of DEMs. The results indicate that the accuracy 
of the produced DEMs was not ameliorated. 
The other factors that influence DEM's quality are image 
quality and terrain type and roughness. The pre-processing of 
the images with histogram matching it does not appear to 
influence the results, since the radiometry of this stereo-pair 
was good. In all the solutions the matching was done with the 
same high accuracy. Thus, the localization of the 94% of points 
was characterized as excellent. 
Finally, the terrain type certainly has an effect on the DEM 
quality. Thus in plain areas the achieved accuracy was 0-5m 
while the maximum height differences have occurred in 
mountainous and steep areas. 
4. REFERENCES 
Boissin, B., Baudoin, A., Begni, G., Fontannaz, D., and Munier, 
P., 2002. A new generation satellite: SPOT 5 in orbit. MEDIAS 
NEWSLETTER, Toulouse, France, N°13, CNES, pp. 74-77. 
Breton, E., Bouillon A., Gachet R. and Delussy, F., 2002. PRE- 
FLIGHT AND IN-FLIGHT GEOMETRIC CALIBRATION OF 
SPOTS HRG AND HRS IMAGES. In: Conference