TOPOGRAPHIC MAPPING WITH SPOT IN POLAR REGIONS 
Frank Pattyn 
Centre for Cartography and Geographic Information Systems 
Geografisch Instituut 
Vrije Universiteit Brussel 
Pleinlaan 2, B-1050 Brussel, Belgium 
LS.PR.S. Commission IV 
ABSTRACT: 
A method is presented for the automatic extraction of topography from stereoscopic SPOT images in polar areas. In 
these regions conventional algorithms fail due to SPOT imaging characteristics and terrain texture. The main 
difficulties are the large difference in orientation between two images of a stereocouple, the shortage of reference 
points and the occurrence of high reflectance zones such as snow and ice. The stereo matching procedure is split up 
in two parts : (i) a geometric correction method in order to register both images in a reference system of quasi- 
epipolar geometry, using few ground reference points and (ii) an area-based matching technique, based on cross- 
correlation. In order to reduce the time needed to match corresponding points a multiresolution approach is 
applied. In the matching procedure also heuristics are incorporated, identifying texture elements such as 'saturated 
snow’, ‘snow’ and. ‘shadow’. Matching accuracy over snow areas is further improved by applying selective image 
enhancement techniques. 
The methodology was tested on a stereopair of the Ser Rondane Mountains, Antarctica. Planimetric accuracy lies 
within the region of 10 to 20m, which is rather low due to the difficulty by which reference points can be identified 
in the image. The highest matching accuracy was found on medium slopes of rock outcrops and on the ice surface. 
On the steep slopes and some snow patches accuracy was significantly lower. 
KEY WORDS : Image Matching, Photogrammetry, SPOT, Stereoscopic 
1. INTRODUCTION liable edges on ice and snow surfaces, so that an area- 
based matching technique is in order. At high latitudes 
Since the launch of the first SPOT satellite in 1986, dig- the prevalent low sun angle accounts for large shad- 
ital topographic mapping from space gained more in- ows, further enhanced in mountainous areas. During 
terest as compared to analogue cartography. It became the winter months images cannot be taken, because of 
possible to extract automatically and very accurately the polar night. The optimal satellite scanning period 
surface topography so that maps at scales up to 1:50,000 is therefore restricted to the months June and July in 
could be produced. The high speed and low cost of the northern and December and January in the south- 
production facilitates large scale mapping, not only in ern hemisphere. Due to SPOT imaging characteristics, 
developed and industrial zones, but also in develop- a large difference in orientation exists between two 
ing countries or such remote places on earth, as polar images of a stereopair. This difference increases with 
regions. Conventional mapping in the Antarctic, for both increasing viewing angle and latitude, so that at 
instance, is very expensive and requires several years high latitudes epipolar lines are not parallel to the 
of of field work (ground truth collection) and a lot of scanline direction. For instance at latitude 70°N, im- 
air surveys, the latter totally dependent on the rapid ages scanned with a viewing angle of -27° and +27° re- 
changing weather conditions. Furthermore, due to the spectively account for a difference in orientation of 
high reflectance of the snow and ice surface it remains 24°. 
very difficult to extract conjugate points with analogue 
equipment. In this paper, a method is presented in order to derive 
surface topography from stereoscopic SPOT images, 
Compared to civilised areas at middle latitudes, digital adapted in such a way to overcome above mentioned 
topographic mapping from space in polar regions is difficulties. Starting from raw SPOT HRV Level 1A 
hampered by terrain texture and the satellite's imaging images, satellite auxiliary data and reference points are 
characteristics (SPOT HRV). Very few reference points processed together in order to resample the images 
are available and it remains difficult to locate the posi- into a geodetic raster, thus leaving only relief displace- 
tions of these reference points accurately in both im- ments unaltered (fig.1). Afterwards, matching is per- 
ages of the stereocouple, because no man-made fea- formed in order to collect conjugate points from both 
tures, such as crossroads and buildings are encoun- image planes. The obtained parallax values are con- 
tered. High reflectance zones such as snow and ice verted into elevations through spatial intersection. By 
show a rather small greylevel variance. On some introducing a priori knowledge concerning surface 
snow surfaces oriented towards the sun, SPOT detec- texture and applying selective image enhancement 
tors are saturated (Digital Number=255) and corre- techniques, the number of ‘mismatches’ is greatly re- 
sponding points cannot be matched. In view of the duced. 
small greylevel variance it is not possible to extract re- 
472