B3. Istanbul 2004
Ax
Ay (
0
nts of interest, P;
rresponding point,
object and scene
>ction vector - i.e.,
and image points,
rd equation in (1).
odel is derived by
‘essed as:
(2)
on parameters, and
no et al., 1999). It
are useful if the
other hand, if GCP
becomes easier to
ar and non-linear
ivigation data are
as to note that a
mes 6-parameters
easily seen if we
ing Z as a linear
it subsection deals
» handling scenes
rmation
ers conform to the
n lines. Therefore,
nodel, Perspective
ne coordinates are
scene coordinates
m to the parallel
(3)
line according to
;pectively.
a flat terrain. In
scanner roll angle,
om the navigation
roll angle together
GCP. Combining
TP transformation
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
xe4Xv-45Y 4.744,
AX + AY + A,7 + A, (4)
y=
pend EF Vau E A
The parameters in Equations 3 - i.e., 4; to Ag and y - can be
estimated using a minimum of five GCP. This model can be
considered as a modified version of the parallel projection
model, by taking into consideration the necessary PTP
transformation. The next section deals with the utilization of the
parallel projection parameters for epipolar resampling of linear
array scanner scenes.
4. EPIPOLAR RESAMPLING OF LINEAR ARRAY
SCANNER SCENES
The epipolar line equation, adopting the parallel projection
model, can be expressed as (Morgan, 2004; Ono et al., 1999):
Gx+G,y+G,x+G,y'=1 (5)
where (x, y) and (x, v) are the coordinates of corresponding
points in the left and right scenes, respectively. Four parameters,
G, to G4, express the epipolar line equation. These parameters
can be derived based on the 2-D Affine parameters or using
point correspondences. In the latter case, at least four
corresponding points have to be available.
From Equation 5, it can be easily seen that epipolar lines are
straight in the parallel projection model. Furthermore, epipolar
lines, in any of the scenes, are parallel to each other, Figure 5.
Therefore, the epipolar line parameters can be utilized to
transform the original scenes in such a way to eliminate the y-
parallax values. One can consider rotating the scenes, each with
different angle, to get the epipolar lines coinciding with the
scene rows. Afterwards, scale and shift has to be applied to
obtain corresponding epipolar lines along the same row. Such
transformation can be expressed as:
I, recu)
5 2 —
Jar L À fo
Figure 5. Epipolar lines in the parallel projection
i ect = ; 0
[: |I es (0) sin(0)] x v. A
JX,] S|-sin(0) cos(9)|| y > (6)
S: À all . ' ' 0
es E (0) sin(0')I x AS
vd Losin) costo) y] [o
Where:
(Xo Fa) are the left scene coordinates after the
transformation;
(C, y) are the right scene coordinates after the
transformation;
0,0 are the rotation angles of the left and right scenes,
respectively;
-4y/2, 4y/2 are the shift values of the left and right scenes,
respectively; and
1/8, § are the scale factors of the left and right scenes,
respectively.
It is important to note that Equation 5 contains only four
parameters. Therefore, no more than four parameters can be
introduced in Equations 6. Eliminating y-parallax (i.e., equating
y 4 and y,) will result in an equation similar to Equation 5, but
containing the four parameters 6, 0', Ay, S. By comparing these
equations, one can derive the following equalities:
G
9 =arctan! - —
5
O'= arctan | — S (7)
4
s= | Gasin(0) _ [ G,cos(8)
G, sin (6") G, cos(0")
AY E sin (0) _ cos (6) _ S.sin (0) | S.cos(9")
S oco, IPIS z. G,
In summary, to eliminate y-parallax values, one has to derive
the epipolar line parameters, G, to G,. Afterwards, the
parameters (0, 0”, Ay, S) can be computed according to
Equations 7. Finally, these parameters are used to rotate, scale
and shift the scene coordinates according to Equations 6.
Although this procedure is successful in eliminating the y-
parallax, it is not sufficient to produce useful stereopair. The
problem is that the resulting x-parallax values are not in linear
relationship with the height values (Morgan, 2004) By
analyzing the transformations that were performed (rotation,
shift and scale), it can be proven that they are planar. In other
words, these transformations do not change the planes
associated with the scenes. Recall that the scene planes are
expressed by the rotation angles (c, 9) in Equation 1.
To solve this problem, scenes have to be projected onto a
common plane, similar to what have been discussed in frame
images in Section 2.1. To select the orientation of this plane, /he
normalization plane, it is important to remember the main
objective — that is to have linear relationship between x-parallax
and height values. Figure 6 shows two points (B and D) of same
elevation and the parallel projection vectors (L, M, N) and (L”,
M', N^) of the left and right scenes, respectively. It can be seen
in this figure that only a horizontal plane (shown as dashed line)
will maintain similar x-parallax values of points B and D.
Therefore, scenes have to be projected onto a horizontal plane
to maintain linear relationship between x-parallax and height.
N
d = ;
N Original left scene
N
\h ee Original right scene Zz
No \
\
\ Alternative planes Nd
\p, \d EY c. sacs
(L,M,N) N
^
N EN (LL MNY)
f \ MN)
Us Rs \ À
\ 7 X Sud
EN T He
ET
Figure 6. Selection of the normalization plane
Following this finding, the direction of epipolar lines within this
plane can be determined. The parallel projection vectors,
together with an object point, form a plane in space - epipolar
plane. The direction of the epipolar lines can then be