TRIPLET-MATCHING FOR DEM GENERATION WITH PRISM, ALOS
---- À Case Study using Air-Borne Three Line Scanner Data
Tianen CHEN ', Ryosuke SHIBASAKI *, Koichi TSUNO”*, and Kazuya MORITA
* Shibasaki Lab, Institute of Industrial Science, University of Tokyo
4-6-1 Komaba, Meguro-ku, Tokyo 153-8505, JAPAN
shiba@skl.iis.u-tokyo.ac.jp
" STARLABO Corporation
1-20-1, Sanshin BLDG, Shibuya, Shibuya-ku, Tokyo 150-0002, JAPAN
chentian@iis.u-tokyo.ac.jp
Commission ITI, WG I11/2
KEY WORDS: PRISM/ALOS, Three-Line Scanner, Pyramid Image, Triplet Image Matching, DEM Generation
ABSTRACT:
For natural landscape, area-based matching can be successfully applied because the elevation is undulating rather gently. On the
other hands, urban areas with the mixture of artificial structures such as buildings and natural landscape, where the elevation tends to
have steeper changes, conventional area-based matching may not be successful. Many computer vision papers demonstrated
matching methods based on features such as edge and regions extracted from images have much better performance for artificial
structures. 2.5 m ground resolution of PRISM, however, may not be enough to extract detailed features enough to represent artificial
structures. This may suggest the combination of area-based and feature-based matching approach can be more appropriate for ALOS
PRISM data. This paper presents the results of experiments on the combination of the two approaches using simulated PRISM image
generated from air-borne image sensor, TLS (Three Line Scanner).
1. INTRODUCTION
A number of space-borne scanner systems using solid-state
linear array of detectors have been successfully in operation for
stereoscopic mapping since SPOT-1 was launched in 1986.
Different from SPOT-1 and SPOT-2 launched in 1990, which
are collecting stereo images by side-looking push-broom mode
and have certain disadvantages, for example different
atmospheric conditions can prevail on different passes, thus
degrading the stereo images, MOMS (Modular Optoelectronic
Multispectral Stereo Scanner)-02/D2, MEOSS (Monocular
Electro-Optical Stereo Scanner), HRSC (High Resolution
Stereo Cameras) and IKONOS-1 of Spacelmaging collect
digital along-track stereoscopic imagery using a three-line
camera (Ebner ef al., 1991 and 1992; Fraser and Shao, 1996;
Frisch et al., 1998; Ebner et al., 1999; Kornus ef al., 2000).
PRISM (Figure 1) of ALOS to be launched in 2004 has three
radiometers for Forward, Nadir and Backward viewing to
collect along-track stereoscopic imagery at 2.5m ground
resolution on 691.65km orbit altitude above equator for
1:25,000 mapping. Each radiometer has field of view with over
70km. Observation width is switched to 70km and 35km
respectively in the Nadir viewing radiometer, and in the
Forward and Backward viewing radiometer, it is 35km. When
70km observation mode of the Nadir is selected, only the
observation data of the Backward radiometer can be acquired
simultaneously. To correct distortion of observation views due
to the Earth’s rotation, even if satellite dose not carry out the
yaw steering to correct it PRISM has an Earth rotation
correction function and obtains images by selecting
automatically the best image extraction position according to
the satellite position information provided from the satellite
system. Table 1 lists its main characters.
Forward
Backward
Figure 1. Illustration of the PRISM
This article deals with the extraction of height information from
PRISM simulated imagery generated from SI-100 which is an
airborne three-line-scanner developed by | STARLABO
Corporation jointly with University of Tokyo in 2000 (Chen ef
al., 2003). Table 2 lists the main characters of SI-100.
International Arc.
ML
Item
Observatic
Band
IFOV
FOV
Focal Lene
Scan Cyc]
Pointing Ar
MTF
S/N
Gain Settii
B/H
[ADB
Angle fro
nadir
Side angl
Ite
Observati
[FC
FO
Focal I
NT
Flight |
AD
Data
Ang le frc
This article deals
PRISM simulate
airborne — three-
Corporation joint
al., 2003). Table
As conversional
generation, imag
DEM generation
epipolar lines in
matching is unfe
feature points an
method has be«
jointly with Univ
100 photogramm
concept of the m
simulated images
comparing with a