art B3. Istanbul 2004
vector have to be
24,4, + A} + A; (4)
uted as expressed in
ite parameters (73, 75,
(5)
's can be computed as
nputing intermediate
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
d. 25; d.
“sin pcos X + —U cos vg + —sin x
; s 5 (6)
Q = arcsin re =
sin” x + (sin p cos x + U cos e)
Readers interested in the derivation of the above formulas can
refer to (Morgan, 2004). So far, the relationships between the
non-linear and linear forms of the parallel projection model are
established. It is important to remember that the linear model is
preferred in cases where GCP are available, while the non-linear
model is preferred if scanner navigation data are available.
Section 4 deals with the latter case. But first, a pre-requisite
transformation prior to handling the scenes according to the
parallel projection model is discussed in the next subsection.
3.4 Perspective To Parallel (PTP) Transformation
Original scenes captured by linear array scanners comply with
the rigorous perspective geometry along the scan lines.
Therefore, before dealing with the parallel projection model,
Perspective To Parallel (PTP) transformation of the scene
coordinates are required. Such a transformation alters the scene
coordinates along the scan lines in such a way to make them
comply with parallel projection (Ono et al, 1999). The
mathematical model for this transformation requires the
knowledge of the scanner roll angle, v, and is expressed as:
|
Mey i
1-2 tan (v)
ré:
where:
C is the scanner principal distance; and
y,y are the coordinates along the scan line according to
parallel and perspective geometry, respectively.
Therefore, any coordinate along the scanning direction can be
transformed from its perspective geometry, y, to another
coordinate according to parallel projection, y’. One has to note
that the mathematical model in Equation 7 assumes a flat
terrain. Morgan (2004) used GCP for estimating the scanner roll
angle together with the parameters associated with the parallel
projection. This model is expressed as:
oe
I
ELXT AYA Z+ A,
AX cr Y udo 44. (8)
j, nv) tt. diy dn
c
In the above model, measured coordinates from the scenes (x, y)
are used directly to estimate the 2-D Affine parameters together
with the scanner roll angle. The next section deals with an
alternative approach for obtaining the scene parallel projection
parameters — that is when scanner navigation data are available.
BETWEEN THE
PARALLEL
4 THE RELATIONSHIP
NAVIGATION DATA AND
PROJECTION PARAMETERS
Referring to the discussion in Section 3.1, space scanners can be
assumed to travel with constant attitude (c,, @,, &;) and constant
velocity V,(AX, AY, A7) during the scene capture, starting from
(Xo, Yo, Zo) as the exposure station of the first image. The
parallel projection vector can be computed as:
A — ^ ^ C
[z M N]= Ks Fos sd (9)
where 7,11 to 7,45 are the elements of the scanner rotation matrix.
l'he scene rotation matrix A in Equations 1, may differ from the
scanner rotation matrix and it can be computed as:
Rss. yz] (10)
where x, y and z are unit vectors along the scene coordinate
axes with respect to the object coordinate system. They can be
computed as follows:
T
V eir T» Fu]
LAN (11)
V, xy]
X=yxz
The scale value of the scene can be computed as:
where:
(Xo, Yo, Zo.) are the components of the exposure station at
the middle of the scene;
Zo is the average elevation value; and
€ is the scanner principal distance.
Finally the two scene shifts can be computed as:
s(r,U mU )X,, dr s(r4U = M. T. s(r, U E M.
s(n; V mo M. + Lo F ln Y, + 8. F-r 7
(13)
Ax
Ay
where U and V can be computed from Equations 3. Derivation
of this transformation is not listed in this paper due to its
complexity. Interested reader can refer to (Morgan, 2004) for
detailed derivation. So far, the derivation of scene parallel
projection parameters, Equation 1, using the scanner navigation
data is established. The next section deals with experimental
results to verify the derived mathematical models and
transformations using synthetic data.
5. EXPERIMENTS
Synthetic data are generated and tested to achieve the following
objectives:
e Verifying the transformation from navigation data to
scene parallel projection parameters (as discussed in
Section 4);
e Verifying the transformation between the non-linear
and linear forms of the parallel projection parameters
(as discussed in Section 3.3);
e Analyzing the effect of PTP transformation, based on
the true/estimated roll angles, on the achieved
accuracy; and
e Comparing direct and indirect estimation of the
parallel projection parameters.