Automatic DEM Generation From Satellite Image
Leong Keong KWOH, Soo Chin LIEW, Zhen XIONG
Centre for Remote Imaging, Sensing and Processing, National University of Singapore
Blk. SOC1 Level 2, Lower Kent Ridge Road, Singapore 119260
Tel: (65) 68746557 Fax: (65) 67757717
E-mail: crsklk@nus.edu.sg,
crslsc@nus.edu.sg,
crsxz@nus.edu.sg
ABSTRACT: This paper presents a method for automatic Digital Elevation Model (DEM)
generation from satellite stereo images. In particular, this paper focuses on DEM generation
from images acquired by a variety of satellite platforms. Images acquired by certain satellites,
such as SPOT4 and SPOT 5, are provided with ancillary data while some satellites, such as
EROS and IKONOS, do not provide ancillary data for their images. Our method of DEM
generation supports both of these two kinds of satellite images. The satellite sensor model is
built and then refined based on GCPs or tie points in the stereo pairs. This method is also able to
generate DEM from images acquired by mixed satellite platforms. In order to improve DEM’s
accuracy, an image contrast enhancement method is adopted during image preprocessing. The
gradient magnitude images are used in image matching. A relaxation procedure is used to
improve the stability of matching result. Accuracy analysis of the results will be presented.
1. Introduction
DEM is one of the most important data used for geo-spatial analysis. Unfortunately, DEMs of
sufficient point density are still not available for many parts of the earth, and when available
they do not always have sufficient accuracy. Since a DEM enables easy derivation of subsequent
information for various applications, elevation models have become an important part of
international research and development (R&D) programs related to geo-spatial data (Thierry
Toutin, Laurence Gray, 2000) .
So far, researchers have investigated various methods of generating a digital elevation model
(DEM) using remote sensing data. One of the methods is to use two images at a time for the
reconstruction of a three dimensional stereo model in which the altimetric information can be
extracted (Thierry Toutin, 1995). This method generally contains three basic steps: set up sensor
mathematical model to reflect the relationship between points on the ground and pixels on the
image, do image matching to get a disparity map, and finally calculate each point’s altitude.
During these three steps, stereo image matching is the most difficult. For one pixel on an image,
its correspondence point on the other image is generally searched in a window by image
matching. This window is called search window. If the size of search window is too big, the
matching procedure will be very slow and the probability of wrong matching will also be big.
But if the search window is too small, the probability of the correspondence point locating
outside of the search window will be big. So how to decide the size of search window is very
important. Generally we must look for a compromise. Firstly we should assure the
correspondence point is within the search window, later on we should reduce the size of the
search window as much as possible.
So far different matching method adopts different method to limit the size of search window.
VLL (vertical line locus) method (Zhang Zuxun and Zhang Jianqing, 1996) adopts a search
window that contains all the possible correspondence points. This window contains the vertical
line locus that the altitude range is from the minimum to the maximum. So actually it uses a
very large search window. Obviously this matching method is very slow. Because the size of
search window is too big, so the probability of wrong matching is very big. The DEM generated
with this method always has much noise caused by false matches.
Doug C’s hybrid matching model (Doug C. Brockelbank, 1991) uses feature based image
matching to extract some seed points firstly, and then uses a cubic polynomials to model the
horizontal and vertical disparity of every point. If the seed points distribute normally, the result
of this method is good. Otherwise, say in the mountain area, the fitting error of the polynomials
may be very big.
Zhang Zhenyou (Zhang Zhenyou and Shan Ying 2001) uses many tie points selected manually
to build up a Delaulay triangulation network, and then interpolate every point’s altitude. He uses
this method to realize three dimensional reconstruction.
The new method this paper presented is to use an image pyramid to limit search window to a
very small region. In order to improve the accuracy of image matching, we use a filter to do
local contrast enhancement and generate gradient magnitude image and after image matching,
we use a relaxation procedure to improve the stability of image matching.
