'. Istanbul 2004
ies are less than
profile the ERS
s in DEM, even
} ranges from |
DATIONS
jtation of ERS
5m- and. i10.m
derately rough
nt of reflectivity
ailable sources
jtation of data
SAT.
‚eter waveforms
ion : an expert
ir Environment,
f ERS-1 derived
points. Space at
414 Vol.I, Mars
appariement des
es d’eau libre et
2.2002
formes d’onde
Rapport interne
ations of surface
al of Glaciology,
| de l’altimétrie
ntróle terrestre —
l'exploitation de
tensity of a radar
tial measurement
nsity. Journal of
AUTOMATIC DETECTION OF SHADOW POINTS
IN DIGITAL IMAGES FOR AUTOMATIC TRIANGULATION
Yandong Wang, Mostafa Madani
Z/] Imaging Corporation, 230 Business Park Blvd., Madison, AL 35757, USA
ywang(@ziimaging.com, msmadani@ziimaging.com
Commission III, WG III/1
KEY WORDS: digital, extraction, image, photogrammetry, triangulation.
ABSTRACT
This paper presents a novel method for automatic detection of shadow points in digital images in order to improve the reliability of
automatic triangulation. The new method is based on the model of shadow points, which contains the radiometric and geometric
properties of shadow in the image. The first is that there is a distinct contrast along the edge of a shadow in the image. The second is
that the surface around a shadow 1s not a smooth surface and the points in the shadow area have smaller elevation than the objects
casting the shadow, such as trees or buildings. The proposed method works in two steps: it first detects points with high contrast by
image processing, and then finds points where the terrain slope changes sharply in its surrounding area by using a local digital
surface model (DSM) which is generated by image matching. The proposed approach has been implemented within Z/I Imaging's
automatic triangulation product — ISAT. The tests on different projects show that the proposed method can detect shadow points
effectively and thus improve the reliability of the triangulation results.
1. INTRODUCTION
Shadow is a common phenomenon in digital images. Since the
information in shadow area may be lost or is very difficult to be
extracted, it often causes problems in many photogrammetric
applications such as automatic extraction of buildings and
roads, generation of orthophotos and automatic triangulation. In
automatic triangulation, tie/pass points of images are extracted
automatically by image matching and the exterior orientation
(EO) parameters of images are computed by using the extracted
image points. Thus, the quality of the obtained EO parameters
of images largely depends on the quality of the automatically
extracted tie/pass points and their distribution in the block. The
existence of shadow points reduces the accuracy and reliability
of the derived EO parameters of the images, especially when
the flight line is long, photos are taken at a large sun azimuth,
or images are taken in a bad weather condition, e.g., a windy
day. Therefore, how to detect and remove shadow points in the
image automatically is an important issue in automatic
triangulation. There are different methods developed for
detecting shadow points. Some of them use the sun azimuth of
the image and DSM to compute the location of shadows in the
image (Amhar et al, 1998; Rau et al, 2002). They work well
when the sun azimuth and DSM in the image area are available.
However, DSM is sometimes unavailable for some
applications, for example, in automatic triangulation. Other
approaches detect shadow area by simply performing
thresholding operation (Gong et al, 2002). Since shadows
usually have large intensity values in the image, most shadow
points can be detected by thresholding operation of the image.
But, some non-shadow objects may also have large intensity
values in the image and thus be detected incorrectly. In this
paper, a new method based on the model of shadow points is
presented. The model contains two distinct properties of a
shadow in the image. The first is that there is a distinct contrast
along the edge of a shadow in the image. The second is that the
surface around a shadow is not a smooth surface and the points
in the shadow area have smaller elevation than the objects
casting the shadow such as trees or buildings. The proposed
method works in two steps: it first detects points with high
contrast and then find points where the terrain slope changes
sharply in its surrounding area by using a local DSM. It is
assumed that terrain has a sharp change in slope when high
objects such as trees and buildings occur. The slope change of
terrain can be calculated by using a local DSM generated by
image matching during the generation of tie/pass points. The
proposed approach has been implemented within Z/I Imaging's
automatic triangulation product — ISAT. In the following
sections, the process of automatic generation of tie/pass points
and detection of shadow points will be described and some
examples of shadow point detection will be presented.
2. AUTOMATIC GENERATION OF IMAGE TIE AND
PASS POINTS
In automatic triangulation, tie/pass points for computing EO
parameters of images are generated automatically by image
matching. Different matching techniques have been developed
for finding corresponding image points in overlapping images
in the last two decades. Feature-based matching (FBM) and
area-based matching such as least-squares matching (LSM) are
two widely used matching methods. Both of them have their
advantages and disadvantages. LSM has very high matching
accuracy, but needs good approximation of the position of the
point to be matched while FBM doesn't need good approximate
location of the corresponding point and has relatively low
matching accuracy. To reduce mismatch points, and thus to
increase the reliability of triangulation results, a combination of
LBM and LSM is usually used in the automatic generation of
tie/pass points since they can compensate to each other. In Zl
Imaging's ISAT, the tie/pass points in the image are generated
by using LBM and LSM from coarse to fine (Tang et al, 1997;
Madani et al, 2001). The following diagram shows a general
procedure of the automatic generation of tie/pass points in Z/I
Imaging's ISAT.
