Down
Wonkyu Park
THE DEVELOPMENT OF AN ACCURATE DEM EXTRACTION STRATEGY FOR SATELLITE
IMAGE PAIRS USING EPIPOLARITY OF LINEAR PUSHBROOM SENSORS AND
INTELLIGENT INTERPOLATION SCHEME
"Department of Computer Science, Satellite Technology ‚Research Center, eatem of Electrical and Electronical
Engineering, Korea Advanced Institute of Science and Technology
(hytoiy |wparkltjkimlsbkim) @ satrec.kaist.ac.kr
Working Group
KEY WORDS: Digital Elevation Model, Stereo Matching, Interpolation, Epipolarity, SPOT.
ABSTRACT
This paper describes a strategy to extract accurate Digital Elevation Models (DEMs) from satellite image pairs. In
general, extracting DEMs consists of camera modeling, stereo matching, editing and interpolation. Because of the
unique geometry and radiometric characteristics of satellite image pairs, methods developed for aerial or perspective
image pairs cannot be applied. The proposed strategy employs a stereo matching algorithm taking the epipolarity of
linear pushbroom sensors and scene geometry into account and an optimal interpolation scheme for satellite images.
Using 100m-resolution Digital Terrain Elevation Data (DTED) by NIMA, USA and 60m-resolution DEM generated
from digitized contours produced by National Geography Institute, Korea, we assess the performance of our strategy in
comparison with other commercial software packages on two 60kmx 60km SPOT panchromatic stereo image pairs.
Based on results, DEMs from our strategy have 25.5m RMS (Root Mean Squares) errors for Boryung area, Korea, and
33.6m for Seoul area, Korea, while DEMs from PCI have 44.7m and 61.1m, respectively and DEMs from Intergraph
have 50.8m for Boryung area.
1 INTRODUCTION
The generation of Digital Elevation Models (DEMs) from remotely sensed images is an important task for various
applications such as three-dimensional Geographic Information System (GIS), cell planning, environment monitoring,
virtual reality and so on. DEMs from satellite images have several advantages over aerial images: 1) A scene covers
large and restricted area. 2) Satellite images are a digital data so that the automation for extracting DEMs can be
archived. 3) In recently, a number of remote sensing satellites have been launched and it is becoming easier to get in
hands. Despite of these advantages, generating DEMs from satellite images suffers from shortcomings — accuracy,
coverage and computation time (Lee et al., 2000).
Extracting DEMs involves the following steps: camera modeling, stereo matching, editing and interpolation. Camera
modeling is a step estimating the sensor position and orientation parameters during the seconds required to capture
images. Stereo matching finds corresponding points in stereo image pairs (Lee ert al, 2000). The elevation data
generated by camera models and corresponding points can be erroneous. In automatic editing, these errors are corrected
using a statistical model (Kim ef al., 1999). After these steps, the elevation data do not provide a complete spatial
coverage. A complete coverage may be obtained by interpolating scattered elevation data.
Ever since computers become available, many researchers have tried to generate DEMs automatically from satellite
images. Brockelbank and Tam (1991) compares the techniques to extract accurate elevation information from SPOT
stereo satellite images. Tateishi and Akutsu (1992) explains a method to produce a relative DEMs from SPOT image
pairs without GCPs. AI-Rousan ef al. (1997) tests and validates DEMs extraction module in PCI EASE/PACE system
commonly used. However, in our opinion, most of these methods do not consider geometric and radiometric
characteristics of satellite images or adapt the model for aerial or perspective images.
In this paper, we propose a DEM extraction strategy considering the geometric and radiometric characteristics of
satellite images. A stereo matching algorithm based on epipolarity and scene geometry is used (Lee et al, 2000) and an
optimal interpolation scheme for satellite images are selected after various interpolation schemes are studied (Kim er
al., 1999), Adapting these two methods, robust and accurate extraction of DEMs can be achieved from satellite images.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 705