In order to efficiently conduct the data management in PHODIS 
AT, a special relational data base (RDB) is used. This adapted 
RDB allows to add any additional information, that may be 
required later on, to the existing data. 
3 Block measurement 
Block measurement is subdivided in three parts: Fully auto- 
matic tie point determination, their checking and semiauto- 
matic control and new point measurement. For the measure- 
ment itself no additional information is required. Neither 
overlaps nor any tolerances have to be entered. 
The names of the tie points are assigned automatically. The 
operator only defines a start value and an interval for the new 
points and the tie points. A name may be numerical and/or 
alphanumerical. Automatic name assignment ensures that a 
point name is not assigned repeatedly to different points. 
3.1 Fully Automatic Tie Point Determination 
Tie points are conjugate points, serving to connect neigbouring 
images together. The determination of tie points in images 
does not need to recognize any specific features. Therefore, a 
full automation of this procedure is possible. A corse-to-fine 
image matching approach combining the feature-based and 
area-based techniques supports the fully automatic tie point 
determination in PHODIS AT. The approach is an extention of 
the one, which is successfully used for automatic relative orien- 
tation in PHODIS ST (Tang/Heipke, 1996; Tang et al., 1996). 
The fully automatic tie point determination in PHODIS AT is 
done in two steps. The first step is called "block formation", 
serving to connect images of the whole block together on 
higher pyramid levels with lower image resolution. The second 
step is called "point tracking", aiming at achieving as high a 
measuring accuracy of each tie point in images as possible 
through the rest of pyramid levels. 
Feature-based matching (FBM) is used to determine conjugate 
points in image pairs in the block formation. Using an interest 
operator, point features are first extracted from images inde- 
pendently and then matched according to certain geometric and 
radiometric criteria. For a model formed by an image pair in a 
strip, the matched point pairs are used as observations in a 
robust bundle adjustment, in which relative orientation pa- 
rameters and object coordinates of the points are determined 
and outliers are also eliminated. The orientation parameters 
and the point object coordinates are then forwarded to the next 
lower pyramid level in order to repeat the matching procedure. 
For a model formed by two images from two neighbouring 
strips, due to the possibly very limited overlap, a robust affine 
transformation is performed to eliminate outliers in the 
matching. A manifold tie point can be obtained by finding out 
the same feature in the common image of the neighbouring 
models. The final result of this step is a list of tie points in the 
whole block determined at a so-called intermediate pyramid 
level. The intermediate level is a pyramid level in which the tie 
point determination can still be carried out fast enough and 
from which enough tie points can be generated for a reliable 
point tracking. 
A tie point in the list generated in the step of block formation 
consists of a point name, number of tie images and a list of 
these tie images including the point measurements at the in- 
termediate level. In order to precisely measure the image co- 
ordinates of the point in the tie images, a least squares match- 
ing (LSM) (eg. Ackermann, 1983) is performed pair by pair 
through the rest of pyramid levels down to the original image 
resolution at the step of point tracking. Around a given point 
pair at the intermediate level, a reference and a search window 
are defined. Six affine parameters and two radiometric ones 
are calculated between the two windows in an iterative way. 
For a convergent result of calculation, the cross correlation 
coefficient between the two windows is then computed. If the 
coefficient is larger than a threshold, the match is declared as 
successful. The interest operator is used again in the reference 
window to find a proper point for transfering to the next lower 
pyramid level. This point is then transformed to the search 
window via the affine parameters, defining the corresponding 
point there. These two points are mapped onto the next lower 
pyramid level and the LSM repeats. At the end of point track- 
ing, the tie point list is updated with image coordinate meas- 
urements in the original image resolution. Since the number of 
tie points in an image can be unnecessarily large for the block 
adjustment, the 2-fold tie points are tracked only selectively. 
Detailed description of the matching approach can be found in 
(Tang, 1996). 
3.2 Checking the fully automatic measurement 
If, during fully automatic measurement, areas such as large 
water bodies are found in which no or too few tie points could 
be measured, these are recorded. After the completion of fully 
automatic measurement, the operator is guided to these areas 
and can, if required and possible, measure further points either 
manually or semiautomatically. 
For spot checking the fully automatic measurements, the block 
display affords any image of the block to be displayed in a 
separate measurement window. There are no limits to the 
number of measurement windows. Since it is easy to get lost 
with only one display unit, PHODIS AT allows the measure- 
ment windows to be distributed over several display units. 
The measurement windows are autonomous entities. They have 
their own zoom, measuring and display functions. This allows 
the independent use of each measurement window. In addition, 
common functions such as a common zoom for all windows are 
also available. 
3.3 Semiautomatic control and new point 
measurement 
Since an automatic identification of ground control points 
(GCPs) in images is still hardly possible (Giilch, 1995), a 
semiautomatic GCP measurement is supported in PHODIS AT. 
The human operator identifies and prepositions a GCP in 
images. An intensity-based least squares matching algorithm 
(ISM) (IfP, 1996) takes then the responsibility for an accurate 
measurement. If new points are required, they also have to be 
measured semiautomatically. These measurements can be 
made at any time, i.e. also during fully automatic tie point 
determination. 
The point measurement proceeds as follows: 
e Select the manual, semiautomatic or automatic mode of 
measurement, 
e Select the zoom factor desired in the measurement win- 
dows, 
e Select the name of the GCP to be measured in the GCP 
list, or activate the new point measurement for the initial 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996 
Since 
ticule 
offer: 
e N 
t CF peel Mr - "- o 
This 
and c 
6 23 6 O. 2 
PHO 
just 
adjus 
oper: 
Inter 
these 
on th 
direc 
block 
ers. 
Adju 
block 
PHO 
new 
these 
cont: