hul 2004
ons with
Le "GPS
se. Since
raphy in
'd not to
sion, for
eal flight
later. All
1ccessful
ping has
in that:
r direct
mber of
rrors and
al aerial
10uld be
pany has
mapping
era and
platform
> of the
ture that
from its
U-based
ographic
s/inertial
| GIS -
Verlag,
NS und
58.
tegrated
ie Image
1any, pp.
atischen
h Ebner
:hnische
3: First
mmetrie,
44.
delivery
'eedings
Aapping
GENERATION OF ORTHORECTIFICATED IMAGE PATCHES
FROM MULTIPLE IMAGES
Shih-Hong CHIO*
Dept. of Land Economics, National Chengchi University, Taipei City, 116, Taiwan, R.O.C.
chio0119@nceu.edu.tw
KEY WORDS: Photogrammetry, Automation, Fusion, Generation, Orientation, Triangulation, Orthorectification, Orthoimage,
Aerial
ABSTRACT:
Orthoimages can be used for several purposes. One of them is used them as control information, instead of artificial targets, to
determine the exterior orientation of new aerial images. For such a purpose, the image displacements on aerial photos, usually tilted
photographs, caused by topographic relief (including the relief of topographic objects) should be removed. In that case the correct
geometrical and radiometric information can be obtained from those orthoimages and provided for control information. That means
“true” orthoimages should be generated. However the generation of “true” orthoimahes needs the accurate height information of
terrain areas and the acquisition of accurate height information are always time-consuming and labour-intensive. Therefore this
paper will try to propose a method to efficientally generate the orthorectificated image patches as controls for aerotrianguation. This
proposed methodology will generate orthoimage patches from multiple aerial images without detailed terrain height information. A
concept of floating plane, similar to the concept of floating mark, will be presented in this paper to produce several candidate sets of
orthoimages for a terrain area at some height range. Each candidate set consists of several possible related orthoimage patches of a
terrain area from multiple images at some height. Subsequently, a procedure will be developed to decide an optimal set from those
candidate sets for this terrain area. A unique orthorectified image patch of this terrain area will be fused from this optimal set. The
relevant assessment of accuracy will be also discussed and conducted in this paper. Finally, experiments prove the good geometric
accuracy of proposed methodology.
1. INTRODUCTION
The procedure of orthorectification will remove the image
displacements on tilted aerial photos caused by topographic
relief and generate the orthoimages. Orthoimages will not only
have the same property as the maps, but also have the detailed
image information. Therefore, they are essential for urban plan,
national plan, resource investigation, environmental monitoring
and so on. Thereby, they play an important role for geographic
information systems (GISs). Because of the same property as
the maps, orthoimages can be also used for the extraction of
topographic information. In addition, 3D Landscape simulation
will be realized together with digital terrain model by using the
relevant computer hardware and software. Moreover,
orthoimages were also used as control information in these
years, instead of artificial targets, for new aerial images [ Hóhle,
1999,2001;Paszotta, 2000; Shan, 1999;Yang, 2003] Those
studies select the existing orthoimage patch and DTM for
automatic exterior orientation. But for better quality of control
information, the "true" orthimage should be produced. The
process of orthoimage is called as orthorectification. During
orthorectification process, the collinearity condition is utilized
to remove the image displacements when the camera exterior
orientation and terrain relief information, including the height
of topographic objects, should be known accurately. That
means “true” orthoimages should be generated from the
accurate height information of terrain area. However the
acquisition of accurate height information is always time-
consuming and labour-intensive. Hence, this paper is going to
propose an approach to efficientally generating the “true”
orthorectificated image patches as controls in aerotrianguation
from multiple aerial images with known exterior orientation,
but without detailed height information of a terrain area. A
concept of floating plane, similar to the concept of floating
mark, will be presented in this paper to produce several
875
candidate sets of orthoimage patches of a terrain area at
different height after the approximation height information of
this terrain area is given. Each candidate set consists of several
possible related orthoimage patches at some height for this
terrain area. Then a procedure to decide an optimal set from
those candidate sets for this terrain area will be also developed.
At last, a unique orthorectified image patch for this terrain arca
will be fused from this optimal set of orthoimage patches. The
relevant assessment of accuracy will be also discussed and
conducted in this paper. Unlike the tests of Hóhle, [1999, 2001 ],
Paszotta [2000], Shan [1999], and Yang [2003] existing
orthoimage and height data are used and the ground coordinates
(XY) with an interpolated height value (Z) are used together
with the corresponding image coordinates in an orientation
program. We hope the proposed orthorectification procedure
will be much more efficient to establish the more accurate
control information for automatic determination of exterior
orientation of new aerial images.
2. METHODOLOGY
In this study, a method to generate a unique orthorectified
image patch of a terrain area from multiple acrial images with
known exterior orientation will be proposed. The basic
assumption is that this terrain area is composed of horizontal
plane and the approximation of height information about this
terrain area is provided. That is, only the approximation of
height information is used for the orthorectification. No precise
height information must be measured by the operator.
After the approximation of height information of a terrain area
is provided, this corresponding 3-D space is subdivided into
several equal-spacing horizontal planes according to the
desirable height accuracy, e.g. 20cm, and the possible range of
height. Each horizontal plane will be further divided into the
