International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
Object space points are simulated and back-projected into
stereopair using the rigorous perspective projection model.
Scanner IOP are simulated similar to those of IKONOS (e.g.,
c=10m). EOP are simulated to have almost 100% stereo
coverage by changing the scanner pitch angles along the flying
direction, Table 1. It is assumed that the scanner’s trajectory and
orientation comply with the constant-velocity-constant-attitude
EOP model.
Table 4. These values are computed based on the total twenty-
five points shown in Figure 3.
PTP transformation is also tested, based on the available
scanner roll angle. The effect of performing a PTP
transformation in terms of reduction of the Z-component of the
errors can be seen. It can then be concluded that omitting such a
correction results in large errors in the derived height values.
Table 1. EOP used for the simulation
Twenty object space points covering an area of 11km by 11km
are simulated with an average elevation of zero and a height
variation of +1000m. Figure 3 shows the distribution of the
object points and the footprint of the left and right scenes.
Among the object points, sixteen points are used as GCP shown
as red triangles, while nine points are used as check points
shown as green circles.
— Let Scene FootPrnt
—— Right Scene FoutPrint
4000 £000 4000 2000 2000 4000 6000 8000
Figure 3. Object space points together with scene footprints
Using scene EOP in addition to IOP and average elevation,
scene parallel projection parameters are derived, as explained in
Section 4 (see Table 2).
Scene X; | Yo | Z | AX | AY | AZ | Roll |Pitch JÁzimuth Meanyy+Stdyy,m Mean +Stdz,m
Km | Km | Km |Km/s|Km/s|Km/s| v^ | ? 2 No PTP 0.582 + 3.427 0.944 + 0.853
Left |-288| 60 1680 | 7 0 0 «5 {2251 0 PTP 0.590 + 3.466 0.023 + 0.194
Right] 277 | -60 | 680 | 7 0 0 5 1225 0 Table 4. Mean error and standard deviation values of the
directly estimated object space points with and
without Perspective-To-Parallel (PTP) correction
For the same synthetic data, the 2-D Affine parameters are
indirectly estimated using GCP. Three sets of results were
obtained: without PTP correction; with PTP correction using the
available roll angle (the true roll angle); and with PTP
correction using the estimated roll angle (Equations 8). The
estimated 2-D Affine parameters and the estimated roll angles
are listed in Table 5. Among the 25 object points, 16 points
(seen as red triangles in Figure 3) are used as GCP in the
parameter estimation. The other nine points, shown as green
circles, are used as check points.
Left scene Right scene
PTP- PIP
No PTP PIE. estimated|No PTP PrP. estimated
true y true y
v V
o: 3.47 2.61 2.60 3.95 2.63 2.58
pixels
y? 0.00 -5.00 -4.63 0.00 5.00 6.09
A, 1.35e-5 | 1.35e-5 | 1.35e-5 |1.35e-5| 1.35e-5 | 1.35e-5
A, |4.88e-7 | 4.88e-7 | 4.88e-7 |4.88e-7| 4.88e-7 | 4.88e-7
A, |5.58e-6 | 5.58e-6 | 5.58e-6 _|-5.58e-6/-5.58e-6| -5.58e-6
A, |3.26e-4|3.26e-4 | 3.26e-4 |-3.26e-4|-3.26e-4| -3.26e-4
Table 2. Derived scene parallel projection parameters
Using the derived scene parallel projection parameters, 2-D
Affine parameters are derived, as explained in Section 3.3.1 (see
Table 3).
Scene| A; A» A; 4, [Ast Ae A; |Ag
Left | 1.35e-5|4.89e-7|5.59e-6 3.26e-4| 0 | 1.35¢-5|-1.18e-6| 0
Right| 1.35e-5 |4.89e-7 -5.59e-6|-3.26e-4| 0 |1.35e-5|1.18e-6| 0
Table 3. Derived 2-D Affine parameters
Some attention must be given to the quality of the derived 2-D
Affine parameters as well as PTP transformation. Space
intersection is chosen as a means of examining their quality in
the object space. In this case, the 2-D Affine parameters for
both left and right scenes, along with left and right scene
coordinates, are used. Therefore, for each scene point, two sets
of 2-D Affine equations are written. For each pair of tie points,
four equations can be written containing 3 unknowns, the object
coordinates of the point. Thus, the object coordinates can be
solved for, in a least-squares adjustment. The errors can then be
computed in the object space between the estimated coordinates
and the original coordinates used in the simulation. The mean
and standard deviation of the errors are computed, as shown in
Scene, L M aigla|am |dem| 5 As |4.95e-9|4.81e-9 | 4.82e-9 |4.54e-9 | 4.81e-9 | 4.87e-9
Left |-3.83e-1|8.05e-2| -5 | 0 | 0 |3.26e-4| O |1.35e-5 As |1.35e-5|1.35e-5 | 1.35e-5 | 1.35e-5) 1.35e-5 | 1.3565
Right| 3.83e-1 |-8.05e-2| 5 0 | O |-3.26e-4| 0 |1.35e-5 A; -1.20e-6|-1.19e-6| -1.19e-6 | 1.18e-6| 1.17e-6 | 1.17e-6
As | 1.62e-5 |-7.80e-6| -6.03e-6 |-3.50e-5|-1.13e-5| -6.06e-6
Table 5. Indirectly estimated 2-D Affine parameters and roll
angles using GCP
As seen in Table 5, the square root of the estimated variance
component, g,, is the smallest by using the estimated roll
angles for PTP correction. On the other hand, omitting PTP
correction results in the largest estimated variance component.
In addition, the estimated roll angles (as indicated in Equation
8) differ from the true roll angles. The difference is attributed to
the assumption made in Section 3.4 of a flat terrain. On the
other hand, it can be seen that using the estimated roll angles
gives better results, in terms of the smallest variance
component. In order to evaluate the quality of the estimated 2-D
Affine parameters, a space intersection is performed based on
the estimated parameters and the coordinates of the points in
both left and right scenes, as explained earlier. The estimated
object coordinates are then compared to the true values and the
errors associated with the points in Tables 6.
International Ai
No |
PTP |
PTP esti
; No]
i. PIP
PTP esti
Table 6. Mean
estin
As shown in Te
object coordinat
by using the inc
using the true rt
for the check p
bias values. A
suitability of inc
the direct metho
6. CONCI
In this paper, |
scenes such as
model is that i
acquired in very
with the rigorou
to be altered in
parallel project
together with
transformation I
GCP are availab
data are availat
parameters from
focussing on P'
developed to est
Scene parameters
mathematical mc
the estimated rol
the object space
addition, indire
smaller error va
derived using the
Future work wil
IKONOS and SP
(such as linear a:
to the parallel -
normalized scen
based on the para
Abdel-Aziz, Y.
Transformation
Space Coordir
Proceedings of
Photogrammetry,
Urbana, Illinois, |
Dowman, I., and
Functions for |
Archives of Photo
254-266.