Full text: XVIIIth Congress (Part B2)

  
camera to each tie area in the photograph with photo 
coordinates which are obtained from digitizing the paper 
print on a digitizer tablet. Each time when the CCD camera 
moves to a new position, it will first take the image of the 
reseau crosses, calculates the absolute reseau coordinates 
of each CCD pixel, than take the image from the film. But 
two serious problems are associated with this solution. 
First there is a long waiting time for the RS1 to switch on 
and off the two light sources and to calculate the absolute 
coordinates of each pixel. Second, due to the constant 
short period on and off switching of the light, the light 
intensity from patch to patch is uneven, which induces false 
gradient in the gray value along the connection border of 
patches. 
Details about the construction of the instrument and the 
calculation of the absolute pixel coordinates can be found in 
the literature and will not be given here any more. 
3. Digitizing the Tie Areas 
Prior to the scanning of the tie areas, their positions and 
approximate coordinates in the photograph have to be 
determined. This can be achieved by a digitizer tablet. First 
the tie areas are selected and marked on the paper prints 
manually. This procedure is very similar to the traditional 
procedure of aerial triangulation point measurement. The 
only difference is that here no actual point but a small area 
of approximate 2 to 4 mm diameter will be selected and 
marked, in which the operator thought it would be suitable 
for the tie points to locate. Corresponding conjugate tie 
areas in the overlapping neighboring photographs are 
marked under a mirror stereoscope. Conjugate tie areas in 
different photographs are given the same number, but are 
stored in different paths in the computer. The path name is 
adopted from the strip number and the photo number. By 
this naming method, the search of the corresponding tie 
areas in different photographs is simplified. The marked 
paper prints are then put on the digitizer tablet and digitized 
manually to get the approximate photo coordinates of each 
tie area. The photo coordinates are then transferred to the 
RS1 scanner for driving the CCD camera. 
In RS1 the positive film is used for scanning. First the 
fiducial marks will be digitized and matched to get their 
reseau coordinates for calculation of the transformation 
parameters between the reseau coordinate system and the 
photo coordinate system. Various matching methods can 
be selected for determining the center of the fiducial marks, 
like centroid of gray values, template matching, Foerstner 
operator etc., also manually measuring by the cursor is 
possible. After the transformation parameters are 
calculated, the CCD camera can be driven to each 
individual tie area according to the photo coordinates and 
that area will be digitized into pixels. Since the patch size 
which can be digitized at one time is very small and the 
marking of the corresponding tie areas in different 
photographs is relatively rough, we have to digitize for each 
tie area at least 4 connecting patches to form a larger 
square of doubled side length to ensure that there will be 
enough overlap between different photographs. 
The 4 patches are joint together through the reseau 
crosses. Since the pixel coordinates (row, column) of the 
reseau crosses are not integers, during the connection a 
resampling process is necessary. After joining, the central 
part encompassed by the four outer reseau crosses is cut 
out for subsequent matching. The reason for cutting out 
only the center part is to avoid geometric distortion which 
might exist in the border area of each patch, because the 
transformation is based on the four reseau crosses in the 
patch and what ever outside the area enclosed by the four 
crosses is less accurate than inside. The center part has 
the size of 512x680 pixels. 
4. Searching for Matching Windows in the Tie 
Area 
From all conjugate tie areas, one will be selected as the 
master area, the rests are the slave areas. The slave areas 
will be matched to the master area. Each time only one 
slave area will be matched to the master area. In each 
master tie area, a number of places should be selected as 
the potential locations of the TPs. The exact location of a 
TP is determined by the Foerstner operator. In order to 
reduce the searching time, the search for TPs is done in the 
image pyramid. For every tie area the image pyramid is 
built by simply averaging 2x2 pixels into one single pixel. 
From the original image upward, three higher levels will be 
built up. The highest level has only 64x85 pixels. 
In the highest level of the image pyramid of the master 
area, the Foerstner operator will be used to find the interest 
points (IP). All points which have the w value greater than 
0.5 times of the mean w in this tie area, and the q value 
larger than 0.5 will be marked as an IP. Maximum 20 IPs 
will be retained, the rests will be neglected. The w value is a 
measurement of the error ellipse of the matched position 
and the q value is a measurement of the roundness of the 
error ellipse (Foerstner 1987). Going from these IPs, their 
corresponding locations in the lower levels will be traced 
down the pyramid until to the lowest level (with the highest 
resolution as originally digitized). Since these locations 
represent interest points in the highest level, which do not 
necessarily represent interest points in the lowest level, 
new IPs will be searched again in the lowest level. But we 
don't have to do blind search over the whole area, only in 
the vicinity of these traced locations the new IPs will be 
searched. In this way, it is guaranteed that the IPs are 
separated with enough distance. The search is done by 
opening a window surrounding these locations and running 
the Foerstner operator again using the same criterion as 
before. This time, again several IPs will be found. But we 
need only one. Therefore the one with the maximal variance 
398 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996 
poi 
Aft 
det 
the 
sta 
the 
dor 
TP 
hig 
Sar 
finc 
The 
of t 
cor 
trac 
sea 
higl 
pos 
nec 
Wh 
is s 
Wh 
abe 
the 
ma 
and 
leve 
pos 
into 
pro 
the 
Bas 
pre 
a pr 
ther 
win 
con 
faile 
con 
find 
poir 
try t 
rest 
Alth 
kno 
COIT 
NCC 
was
	        
Waiting...

Note to user

Dear user,

In response to current developments in the web technology used by the Goobi viewer, the software no longer supports your browser.

Please use one of the following browsers to display this page correctly.

Thank you.