EREO
MARS)
elds. However,
other hand, the
lity of both 3D
: and the model
llins in 1968. It
mate is exactly
reomate is used
vith the current
; and thus leads
| be achieved in
d by a mosaic
coordinates on
'e and a mosaic
| but also offers
model without
seamless stereo
ilar with that in
the sum of two
h images of the
gether with the
) surface of the
ig stereo model
users (Blachut,
rjakoski (1990)
aken as a map
thophoto pair
in Wijk, 1970).
o pair have also
k, 1979). It is
ith the current
tereo models is
‘the photo pair,
n photo pair.
to generate the
gorous and thus
nent in highly
iccuracy. of 3D
| low relief.
more than one
will enable us to
ereo orthophoto
‘7001: Li etal,
the concept of
is formed by à
e. mosaics of à
whole block of aerial photographs), with the lineage (image
coordinates on original photograph and the orientation
parameters of the original photograph) of each pixel on both
mosaic orthoimage and a mosaic stereomate recorded. Such a
measurable seamless stereo model not only provides seamless
3D landscape environment but also offers the rigorous and thus
accurate 3D measurement of any object and feature visible in
the measurable seamless stereo model without explicit
orientation procedure.
Following this introduction is a review section, which examines
the limitations of existing solutions, then the concept of
measurable seamless stereo model is introduced and the
principle described. Then the procedures for measurable
seamless stereo model and algorithm of accurate measurement
are presented. An experimental testing is also reported at last.
2. A CRITICAL ANALYSIS OF EXISTING METHOD
The stereo orthophoto pair was introduced in 1960s (Collins,
1968; Blachut, and van Wijk, 1970). A stereo orthophoto pair is
formed by an orthoimage and a stereomate (Collins, 1968). The
stereomate is usually produced from the neighboring images
(left or right) within a flight strip by artificially introducing
horizontal parallaxes, but it can also be produced from the same
image. If it is produced from the same image, the objects not
included in the DTM (like buildings, etc.) will appear lying on
the terrain when the orthoimage and stereomate are viewed
stereoscopically.
The horizontal parallax is introduced usually as a linear function
of the height. The formula is as follows:
Poe,
Where h is the terrain height; k is an optional factor. k = tan (a),
and a is the angle of oblique projection. Usually œ is chosen as
tan (a) = B/H, namely the base-height ratio.
For a more rigorous agreement the original parallaxes with the
introduced parallaxes, the projection angle a should vary with
the terrain height (Wang, 1990), so the parallax is introduced as
a nonlinear function of the height (Wang, 2001). The
mathematical formula is as follows:
P Bh Q)
mH
Where:
B is the base line of stereo model
H is the flying height.
h is the terrain height
In addition, a special parallax function -- a logarithmic function
has also been proposed by Collins (1970) as follow:
Where:
B is the base line of stereo model
H is the flying height.
h is the terrain height
JD measurement in a stereo orthophoto pair has been discussed
by Collins (1969), Kraus et al. (1979) and van Wijk (1979). The
traditional measurement method is that the height can be
directly derived from the parallax without the need for DTM
and the planimetric coordinates can be acquired from the
orthophoto. The height measurement accuracy is better than the
accuracy of DTM. The accuracy of the height measurement
from stereo orthophoto pair is two or three times more accurate
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
than the DTM which was used for their generation (Kraus et al.,
1979, Kraus 1984; van Wijk, 1979). The is because the errors in
the orthophoto and its stereomate caused by DTM errors will
have the same sign and therefore be partly cancelled out in the
computation of parallax.
The traditional measurement method is very simple but not
rigorous. The accuracy of measurement on the stereo
orthophoto pair is not able to reach the level when measuring
the stereo model formed by the original images. When the
terrain is flat, good results are achievable, however in
mountainous areas, with the decrease of orthoimage accuracy,
the errors may become too big to be acceptable for
high-accuracy applications. Table 1 illustrates the variation of
measurement results with terrain type. (The data sets are
described in section 5).
Table 1. The 3D measurement error of the stereo orthophoto
pair model in different terrain types
Landform Error X (m) Y (m) Z (m)
MAX] 1.72 1.70 1.83
Flat Are: |
hide RMS | 092 | 091 | 0.99
Hill Mountainous | |MAX| 4. 44 4. 38 7.45
Area RMS 2.83 2.77 3.94
To summarize, the existing method is based on photo pair and
the 3D measurement is a simple computation. That cannot meet
with the requirements of a large area in practical applications.
Wang (2001) and Li et al. (2002) have ever discussed an
approach for generating a seamless stereo model, however, it is
very complex and inconvenient in practical applications.
3. THE PRINCIPLE OF *MEASURABLE SEAMLESS
STEREO MODEL"
For a photographic block, many stereo models can be formed by
stereo pairs. Figure 1 illustrates a block consisting of three strips,
with six stereo models in each strip.
Strip I
Strip II
Strip IH
[ —— ] The stereo extent of a photo-pair model
7 The overlapping area among adjacent photo-pair
CALE wi
stereo models along flying line
The overlapping area among adjacent photo-pair
stereo models across flying line
Seam lines among adjacent photo-pair stereo models
The valid mosaic polygon of the photo-pair
stereo model
Figure 1. The arrangements of photo-pair models
In this figure, the overlapping area between two adjacent images
is called the stereo extent; and a line in the overlapping areas
between two adjacent stereo models is called a seam line. The
polygon area formed by the seam lines of every stereo model is
called valid mosaic polygon of the model.
The mosaic of othoimages of all the valid mosaic polygons in
