tanbul 2004
ial rotations
we the same
ons used in
ed matrices
715; are the
unit vector
Ay, Ay, Az)
bject space
ients of the
t coordinate
rm is valid
ransformed
ondition is
e equation.
individual
(5)
(6)
(7)
the rotation
its 7x-z, by
ion can be
(8)
(9)
ale factors,
; 8 and 9
IM.
lel used by
the use of
rdinates in
meters). A
could be
Ls), which
bservations
im amount,
bove, unit
e and thus
up of lines
s in object
om the test
and GCPs
ise of the
init. vector
veen those
; b) large
lis finding
concerning
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
Several solutions can be applied to overcome this problem.
First, the translation coefficients may be recovered if the shift
between the origins of the two coordinate systems is known. In
this case, the six coefficients representing the scale and rotation
are recovered by the aid of the GCLs and then the translation
coefficients (b; and hg) will be determined from the shift
between the origins. This is a special case, which occurs only
when local image and object coordinate systems are used.
Second, the six coefficients relating the scale and rotation
transformation coefficients are recovered as in the first case and
a single GCP could be used to define the translation
coefficients.
2.1.2 Using the 3D affine LBTM: Similar to the collinearity
equations in photogrammetry, the 3D affine LBTM can be
applied to various operations such as space resection for
calculating the model coefficients (image parameters), space
intersection for determining the location of a 3D point, and
image rectification. Recovering the model coefficients for
images by the LBTM leads to one of the followings processes:
either working on 2D image rectification (in case of single
image transformation) or 3D geo-positioning determination (in
case of using stereo pair images). For image rectification, points
in image space could be transferred directly to an object plane
by assuming an average elevation of terrain for the whole area
covered by the image and by using the recovered model
coefficients and image coordinates of the points. For the same
purpose, a Digital Elevation Model (DEM) could also be used
together with the previous information to enhance the results,
especially when the model is applied to undulated terrain.
For 3D geo-positioning determination, stereo images should be
used. More specifically, a group of GCLs should be used to
recover the stereo image parameters individually and then 3D
point coordinates could be calculated. Each point in the area
covered by the stereo images will raise four equations to
recover the three unknowns (X,Y,Z), which means one
redundancy is available to check the quality of the results.
2.2 The 2D LBTM
Two-dimensional transformations are required in many
applications of photogrammetry and remote sensing. We often
need to convert between coordinates in two different plane
systems having different origins, orientations, and possibly
scales. More importance is added to the need for a 2D
transformation when terrain information is either not necessary
(as in case of flat terrain or low accuracy requirements) or not
available (as in case of remote, unmapped areas).
Similar to the way in which point coordinates in the 3D affine
LBTM are replaced with unit vectors, the 2D affine LBTM can
be rewritten as follows:
a SC ACC EC (10)
E ^ + 11
AS Cry FC EC (11)
where (a,, a,) are unit vector components of a line segment in
the image coordinate system, (Ax , Ay) are planimetric unit
vector components of the conjugate line segment in the object
coordinate system, and C,,...,C; are the model coefficients.
In addition, the 2D conformal LBTM can be presented for
unique scale transformation as follows:
853
0, C4 —C 4. 4C, (12)
a=CA +64 +C (13)
4
where (a, a,) and (Ay, Ay) are unit vector components of the
line segment in the image and object coordinate system
respectively, and C,,...,C, are the model coefficients.
As was demonstrated in the derivation section of the 3D affine
LBTM, line unit vectors are not a unique representation of
lines. Therefore, coefficients presenting the scale and rotation
can be recovered with the use of GCLs (as was explained in
section 2.1.1), and calculating the translation coefficients will
require one additional GCP. A minimum of three GCLs is
sufficient for the determination of the 2D affine LBTM
coefficients, and a minimum of two GCLs is enough to
calculate the 2D conformal LBTM coefficients. Comparing the
3D LBTM to the 2D LBTM, it can be said that the latter can be
applied to rectify single images instead of stereo images
without needing any relief related information. As with the 3D
LBTM, synthetic as well as real data have been used to verify
the 2D LBTM model and the results are presented in the
following section.
3. EXPERIMENTAL RESULTS AND ANALYSIS
3.1 Synthetic Data
The effects of various factors such as differences in terrain
elevation, inclination angle, length, number, and distribution of
GCLs were tested for the developed model. Image coordinates
of four sets (S1, S2, S3, S4) of synthetic stereo image data were
derived from the actual orientation parameters of a stereo pair
of Ikonos imagery and synthetic object coordinates of 48 well-
distributed points. The sets were derived so that both S1 and S3
represented undulating terrain with height variations of about
100 m, while S2 and S4 represented flat terrain with height
variations of less than 15 meters.
Two different configurations of 12 GCLs were established from
half of the object points and the remaining object points were
used as checkpoints. One configuration consisted of long, 200 —
500 meter, lines with random orientations and the other of
short, 100 — 200 meter, lines predominately aligned to the
diagonals of the image. S1 and S2 used the long lines, while S3
and S4 used the short lines. The two configurations of GCLs
and checkpoint data are shown in Figures 2 and 3 respectively.
Extensive sets of experiments were performed, but only a few
representative cases are reported here due to the space
limitations. Tables 1 and 2 present the results of the 3D and 2D
affine LBTM respectively. Different groups of GCLs (from 4 to
12 GCLs) plus one additional GCP were used, and the results
are summarized in terms of RMS errors of the 24 independent
checkpoints in the X and Y directions. The RMS errors in the Z
direction (in case of using the 3D affine LBTM) are not
considered here as this study focuses on the rectification of the
satellite imagery. A detailed discussion of the 3D geo-
positioning determination will be presented in a future
publication.
From the results obtained, it is obvious that the LBTM works
significantly well for image rectification. The investigation
shows that, except for the translation coefficients which are
calculated by the aid of the additional GCP, coefficients