anbul 2004 
  
  
  
  
   
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004 
  
  
  
  
  
  
  
  
Table 7. Accuracy analysis based on orthogonal distances. In each comparison the absolute maximum distance, the average distance, 
the standard deviation and the RMSE are shown in meters. 
  
  
  
  
  
  
  
  
  
  
  
  
  
DEM PROCEDURE 1 PROCEDURE 2 
Max Average Standard RMSE Max Average Standard RMSE 
distance distance deviation distance distance deviation 
1 19.60 1.6 1.5 22 18.7 2.2 1.7 2.8 
2 37.99 1.8 1.8 2.6 37.5 27 2.0 3.4 
3 22.33 1.4 1.3 1.9 21.4 27 1.8 32 
4 19.74 1.5 1.4 2.1 20.0 2.2 1.6 27 
5-] 26.25 6.3 4.3 7.6 26.3 6.4 4.4 7.8 
5-2 73.60 6.8 5.8 8.9 70.1 6.0 5.0 7.8 
  
  
  
  
  
  
8. CONCLUSIONS 
In this report the methodology applied and the results obtained 
during the ISPRS-CNES Initiative about DEM generation from 
SPOTS-HRS are described. 
Our Institute was involved as Co-Investigator in the HRS-SAP 
Initiative through the processing of the dataset number 9, 
located in Bavaria (Germany). 
All the algorithms used to process the data and generate the 
DSMs have been developed at our Institute. Using the 
information contained in the image metafile and a suitable 
number of GCPs, the images have been oriented according to 
two different approaches, based on a rigorous sensor model for 
CCD linear array sensors with along-track stereo capability 
(Procedure 1) and on Rational Polynomial Functions 
(Procedure 2). More than eight million image points have been 
measured in the stereopair with the modified Multi Photo 
Geomuiically Constraint matching algorithm designed for 
pushbroom imagery. Using the two orientations estimated by 
rovedures 1 and 2, two distinct DSMs of tie tull area (120km 
x 60km) have been generated and compared to the reference 
DEMs. For the quality control, a 2.5D (calculation of height 
differences) and 3D analysis (normal distance between one 
reference surface and the measured DEM) have been used. 
Also, the areas covered by trees have been manually removed in 
order to provide a congruent analysis in order to judge the 
influence of trees. 
From both the 2.5D and 3D quality analysis it resulted an 
average error between the generated and the reference DSMs of 
around 1-2 pixels (2.5D analysis) and up to slightly more than 1 
pixel (3D analysis), depending on the terrain type. The best 
results were achieved in smooth and flat areas, while in 
mountain areas some blunders even exceeded 100 meters. The 
differences between the DSMs obtained by the two different 
methods of orientation were less than a forth of a pixel. 
In conclusion, the work carried out at ETH confirmed within 
the HRS-SAP Initiative the high potential of SPOTS-HRS 
scenes for DEM generation.