DEM EVALUATION GENERATED FROM HRS SPOTS DATA 
M. Tsakiri-Strati ', O. Georgoula?, P. Patias? 
g 
‘ > 9 ; 3 
' Associated Prof., “Assistant Prof., "Professor 
Department of Cadastre, Photogrammetry and Cartography, Aristotles University of Thessaloniki, Greece 
martsak(deng.topo.auth.gr 
Commission I 
KEY WORDS: SPOTS, HRS data, auto-matching, triangulation, DEM, Evaluation 
ABSTRACT: 
The purpose of this project is to test High Resolution Stereoscopic (HRS) data of SPOTS. This projeet is done in the framework of 
the “HRS study Team” international test, supported by ISPRS-CNES. The Digital Elevation Model (DEM) was derived from the 
HRS data and was evaluated by using a 3D reference data set. The study site for HRS data is located in Manosque, France. It is a 
countryside landscape, with fields and villages. The height relief is medium and in some areas considerably intensive. The HRS data 
was acquired on 14" August 2002. A number of different DEM generation experiments were carried out, in order to examine the 
influence of different processing parameters on the final accuracy of the DEM. Thus: 1) The same date acquisition of the stereo-pair 
reduces the radiometric variations of images; nevertheless, the fact that the images are captured by two different sensors has led us to 
apply a histogram matching technique for image pre-processing, in some experiments. 2) The georeferencing and the stereo model 
set-up of the images were established in four different ways: by using only the metadata; by using the metadata and tie points that 
have been generated automatically; by using the metadata, tie points that have been generated automatically and some tie points that 
have been derived manually; The same as the second solution but in the triangulation process 6 GCP's (XYZ coordinates) have been 
used. For the extraction of the GCP's an additional procedure has been performed, which consisted basically of the production of an 
orthoimage based on the reference DEM. The study of the accuracy statistic results of 8 different derived DEM,s showed that the 
elevation accuracy was better, about 0-5 m, in flat and even areas than in mountainous and steep areas. 
1. INTRODUCTION 
Automatic generation of a Digital Elevation Model (DEM) is a 
very important issue for many applications as map generation, 
three-dimensional GIS, environmental monitoring, geo-spatial 
analysis etc (Kim, et al, 200; Vadon, 2003). In the past, the 
automatic generation of a DEM from aerial images has received 
a lot of attention among photogrammetrists. Now Digital 
Photogrammetry offers all the necessary algorithms and 
methods for the automatic production of accurate DEM from 
aerial photos. On the other side satellite images have several 
advantages over aerial photos: it is easier to acquire images 
periodically, in digital format and to cover large geographic 
extents etc. But satellite images have many disadvantages over 
aerial photos as well, like unknown or improper understandable 
camera geometry, inappropriate geometry of reception, low 
resolution etc. However, automatic DEM generation has 
become an important part of international research in the last 10 
years. Thus, new algorithms have been developed, the 
performance of which have been assessed and reported in the 
bibliography (Singh, et al, 1998; Toutin, 1998; Zhen, et al, 
2001; Toutin, et al, 2001; Toutin, 2001; Lee, et al, 2003). 
On May 2002, the Earth Observation satellite Spot5 from 
CNES, was successfully placed in orbit. The basic 
characteristics of this just 2 years - old satellite are well known. 
The major advantages, of SPOTS system, that influence the 
productivity and the accuracy of a DEM are (Boissin, et al, 
2002; Breton, et al, 2002; Lasselin, et al, 2003; Léger, 2003; 
Valorge; 2003): 
e The resolution: a sampling interval at 5m in panchromatic 
mode. 
e The new instrument High Resolution Stereoscopic (HRS) 
conceived especially to do stereoscopy along side satellite 
track. It acquires simultaneously stereo-pairs in quasi real 
time, within an interval of 90 seconds. 
e The possibility to know the location of the satellite, with 
absolute precision better than 15 meters, without any ground 
control point (GCP). 
This project was done in the framework of the "HRS study 
Team" supported by ISPRS-CNES. The purpose of this Study 
Team is to test the capability of HRS data for the production of 
digital elevation models (DEMs). The Department of Cadastre, 
Photogrammetry and Cartography of the Aristotles University 
of Thessaloniki in Greece participate in the Initiative as Co- 
Investigator. 
2. DEM PRODUCTION 
2.1 Data 
The study site is located in Manosque, France. It is a 
countryside landscape, with fields and villages. The height 
relief is medium and in some areas considerably intense. The 
HRS SPOT-S5 panchromatic stereo-pair covers a region about 
120 Km x 60 Km and it was acquired on 14" August 2002 (Fig. 
1). The processing level of the scenes is 1A (original images 
with radiometric corrections). Metadata files for each scene 
were available. An aerial reference DEM for the accuracy 
assessment was available as well. The reference data set 
describes an area of roughly 6 Km x 65 Km. 
   
   
   
    
   
   
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
   
    
   
   
   
    
   
   
   
   
   
   
   
    
    
   
    
    
    
   
  
   
  
Inte. 
Thi: 
COV! 
me 
nor 
Pho 
the 
Fic 
2.2 
In 1 
havı 
diff 
acci 
are | 
The 
the 
pro 
faci 
that 
to : 
prox