fthe
It is
| the
tizer
hing
y by
hing
d by
nine
ned,
| for
tthe
ding
| the
ut in
stem
ause
man
tem,
arch
yout
can
from
nt of
ke in
!see
TPs
the
icult.
d be
n of
jonal
from
atch
xels.
ry to
digitize the whole photograph and since the blind, fully
automated system is not easy to develop, we decided to
use this RS1 to develop a kind of semi-automatic TP
measuring system. We call it the semi-automatic system,
because the area which should be digitized is determined
by the operator not by the computer. That is, conjugate
areas in the overlapping photographs are searched by the
human operator. Each time only a small area is digitized by
the CCD camera. This area will be called the tie area,
because in it the aerial triangulation tie points should
locate. Since the conjugate tie areas in different
photographs are selected and digitized by human operator,
the difficult blind searching of them in the whole block or in
a whole photograph is avoided.
After these areas have been digitized, TPs will be defined in
them and transferred between them. The methods used in
our system for defining and matching the TPs are mainly
adopted from the Helava's proposal for the DCCS. The
greatest change in our system is that instead of using one
fixed window size, we use several different sizes for the
Least Squares Matching window in order to increase the
reliability and the accuracy of the matched results. Another
change is that we match each time only one pair of tie
areas. There will be no multiple image matching like in the
DCCS.
With this semi-automatic point measuring system, we
combine the point transfer and the point measurement in
one procedure. Both the speed and the accuracy of the
aerial triangulation point measurement are increased.
Since only small tie areas are digitized, the data amount is
very small. There is no problem concerning the storing and
the managing of the image patches. An ordinary PC486 is
used for this system.
After the tie area is digitized, image pyramids will be built
up. From all the conjugate tie areas, one will be chosen as
the master tie area and the rests as slave tie areas. In the
highest level (with the lowest resolution) of the pyramid of
the master tie area image, the Foerstner interest operator is
used to find the places where the windows for defining the
TPs should be. A number of such places should be chosen,
in case that some would fail in the Subsequent matching.
These places will be traced down through the pyramid to
the lowest level (with the highest resolution). The actual
TPs will finally be found there.
While applying the Least Squares Matching technique, we
have noted that the reliability is a problem. If the initial
approximate position for starting the matching is not
accurate enough, the LSM method could converge to a
false place and there is no information from the matching
itself to judge the reliability. Also the size of the matching
window is a problem. There is no such thing as the best
window size. Different window sizes converge to different
positions, the difference can some times be larger than 0.2
397
pixel. Therefore we have developed a strategy to cope with
these problems. Instead of one fixed window size, we use
several different sizes for the LSM and by analyzing their
normalized cross correlation coefficients (NCC) after the
LSM matching, the best result is selected. When no reliable
one could be found, that point will be abandoned. If there
are not enough reliable points in a tie area, the system will
switch automatically to the manual selection of the target
and the search windows in the master and the slave tie
areas respectively.
2. System Configuration
The whole system is built up on the basis of the Rollei RS1
scanner running on a COMPAQUE PC. The RS1 is
designed for digitizing the photographic film with very high
geometric accuracy at low costs. Which is achieved by
using a reseau plate (Luhmann and Wester-Ebbinhaus
1986). The main parts ofthe RS1 area CCD camera anda
glass picture carrier with 2mm spacing reseau crosses all
over its entire surface. The CCD camera has an array of
512x512 pixels and locates above the picture carrier. The
pixel size in the film is approximately 8umx 6um. Therefore
the total area of the film which can be digitized at one time
is approximately only 4mm x 3mm. But the CCD camera
can be moved to any place over the picture carrier by step
motors, so that the whole picture can be digitized patch-
wise. Since step motor is not very accurate, Luhmann and
Wester-Ebbinhaus developed this cheap solution of using.
reseau plate to obtain the high precision absolute reseau
coordinates for each pixel.
Two different light sources are built in the RS1 which can be
Switched automatically in turn. When the bottom light is
switched on, the light goes through the film into the CCD,
and the film can be digitized. When the side light is
switched on, only the reseau crosses are imaged into the
CCD. In this case, the pixel coordinates of the reseau cross
center can be obtained after an automatic matching of the
cross image to the standard template. Then the
transformation parameters from the pixel coordinates to
the reseau coordinates can be calculated. Since the
absolute coordinates of the reseau crosses are known, the
absolute coordinates of each pixel can be calculated by the
transformation parameters.
The distance between the reseau crosses is 2mm and the
CCD digitizes an area of 4mm x 3mm, so that it is
guaranteed that there will always be at least 4 reseau
crosses used for calculating the transformation
parameters. The area encompassed by the 4 crosses has
approximately 240x350 pixels. In order to reduce the error
caused by extrapolation, only the pixels within the four
crosses will be used for the TP measurement.
The CCD camera can be driven by the reseau coordinates
or by the photo coordinates. In our case we drive the CCD
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996