International Archives of the
A
i 1+]
wig
X,y
x (max),y (max)
Figure 7. Finding the maximum point
2 2
z » ax^ * by, * ex dy texy* f (3)
For the area-based template matching, the locating accuracy
ranges from a half to one pixel, which is, of course, not accurate
enough for most photographic products.
3.8.2 Least Square Matching(LSM)
Precise measurements of all fiducial. marks were performed by
matching the extracted patches. The algorithm used is known as
least square image matching, which allows point measurements
with sub-pixel accuracy and is described in Gruen and
baltsavias(1988). In these investigations, the algorithm was
used in its unconstrained mode using an real template which
was read from the original file is used. The accuracy of the
matching algorithm is dependent on the used templates. Thus,
the generation of templates for each camera type is an essential
factor. The result from the template matching is definitely good
enough for a successful LSM processing (half or one pixel).
Here is the equation we used for the system (Eq.4).
go, y) e Af(x* psy * p,)*h (4
f (x. y) = The template
g(x, y) — The picture
(p... p.) = Transformation parameters
h. h, = Two radiometric parameters of scale and shift
respectively
The initial values = (2s ph j h, y = (0,0,1,0)
4. Tests and Results
Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
The Results of all three algorithms implemented can be seen at
three tables below. Input data for any algorithm is first, images
of RMK camera of Ziess Company of Germany second, camera
calibration file including position of fiducial marks im a
standard conditions of labratory. Output data includes final
report including positions of fiducial marks in pixel system,
residual errors of fiducial coordinates, RMSE(Root Mean
Square Error) and the pose of the photo on scanner seen on the
designed menu(Fig. 8). Images were scanned on two resolutions
of 30 and 15 micrometer by Intergraph PhotoScan TD. In all
tables? is the overall accuracy of the automatic interior
orientation process in micrometer.
Resoluti 2 3
esolution ad... e dx C min
30 22 6.1 4.2
Table 1. : The results of using CCF method for localization and
LSM for precise measurement
Resoluti 2
esolution € um e o C min
15 4.1 6.2 5.4
30 3.5 7.2 5.4
Table 2. : The results of using BCCF method for localization
and LSM for precise measurement
Resolution Gon 0 max a min
15 3.7 6.1 5
30 2.6 6.4 4.5
Table 3. : The results of using CCF method for localization and
ISF for precise measurement
m ET T Interior (irientation [Ir xi
File Edit. Window Help
photo is positive
A P RE CU INS Photos
«photo must first flip vertical before 90 CW rotate»
Other © © |
<photo must first 90 CW rotate before flip horizontal>
# Of FMs
Accurate OF Int. Orientation OR rm.s.e is: Eight <|
4.9856 micron<—>0.16619 pixels & Resolution=30 Trenstom. Type
Projéctive |
Residuals is :
1.70163 -7.71973 1 - Unit Tape
-4 60447 11.3474 2 [Men 7
0.566483 2.83487 3
-4 64675 2.489 4 Method
0.195806 1.7184 5 [um — rz]
5.95626 -7.64393 6 [crusM |
0.689184 -6.36711 7
0.141862 3.34106 8
Position Of Fiducial Marks At Pixel System :
546
3831
3830.5567
66.361806
7595.1196
66.174325
7595.0313
65.841339
7595.2876
7595
66.125719
3830.1485
3830.3708
7594.7648
67.081698
66.292629
7594.3524
yt
1
CO 7 0» Cl 49 C2 NJ
CAMERA TYPE
RMK [Zeiss )
Image Type
RAW
Figure 8. A sample of designed menu
Interr
5. Co
Acco
differ
GVC
One r
for th
cente!
fiduci
Dust
scann
decre:
may |
least
accur:
affect
patch:
6. Rei
Kerste
Of D
Engin
-pp. !
Lue. *
Proc. |
Schicl
Orien:
Color:
Heipk
Orient
Richai
Analy
Foerst
correl:
Remo!
Atkins
Visior
Rafael
Proces
1987.
Scott,
Proces
Thurg:
of Ph
Report
Ali A
Crosse
Confer
Ackeri
and po
Recorc
