> EE
precision applications, signalized points have to be
measured, the pixel size has to be derived from the
requirement that the minimum target diameter in
image space should be four to five pixels. Five pixels at
10 micron would result in an image diameter of 50
micron. Therefore, what can be considered an optimal
size for analytical triangulation constitutes the bare
minimum for digital triangulation. This has to be
carefully considered when object target size, image
scale and scan pixel size are selected. Depending on
the image scale and scan pixel size digital triangulation
may require very large and inconvenient signals on the
ground.
(c) A third issue to be addressed is the fact that com-
mercial triangulation systems by and large are still
following even the analogue photogrammetry concept.
Blocks are built up through measurements in strip
direction, collecting image data from one strip after the
other. Bundle adjustment is performed in batch mode
only after all images have been measured. This
procedure ignores the fact that blunder detection can
be performed much better on-line if the images are
measured according to the avalanche principle out of a
particular block corner, building up highly reliable units
very quickly (Gruen, 1983). The possibility of
displaying a set of six or even more conjugate image
regions greatly facilitates this measurement concept
(compared to the corresponding procedure on analy-
tical plotters). Such on-line control of measured data
should be supported by the appropriate computational
strategy. For that, suitable sequential estimation
algorithms are available for on-line bundle triangulation
for quite some time (e.g. Gruen, 1985b, Kersten et al.,
1992). To be able to detect even small blunders not
very far above the noise level, additional parameters
for systematic error compensation should be included
in the sequential estimation.
(d) On the positive side we note that semi-automatic
triangulation, even with current commercial technolo-
gy, allows to measure many more tie points than
conventionally, in much shorter time. Practitioners
report about 35-40 aerial images being measured in
half a day. However, this does not include the time for
scanning, preprocessing (minification, compression)
and data handling. These procedures (1,5 hours per
photograph have been reported) slow down the gross
time substantially. Also, if the disk cannot accom-
modate the image data for a full block, data transfer
time and thus the overall triangulation time will inc-
rease tremendously.
3.6 Orthoimages, orthomaps
It is generally acknowledged that orthoimages are the first
(and so far only (?)) practically requested products from
Digital Stations. Once a DTM is available, the process of
orthophoto production is straightforward and could be
implemented on a desktop computer. Also, the supporting
software like image enhancement, mosaicking, etc. is fairly
standard today, as are map annotation programs. There-
fore, under these conditions, it is somehow amazing that
users invest a lot of money into highend Digital Stations with
all their redundant functionality, just for the purpose of
BEE AA A HEN
orthoimage generation.
Orthoimages are correct representations, if the object is
sufficiently smooth. Since buildings and other man-made
objects are usually not well modelled in DTMs/DSMs the
locations of roofs, etc. are false. Roof correction software
has so far been offered by only one vendor (Leica/Helava),
although the correction algorithm and its implementation
does not pose any serious problems (Dan, 1996).
3.7 Monoplotting
As DTMs and DSMs are becoming increasingly available
the issue of monoplotting deserves much more attention.
Monoplotting is fast, easy to perform and does neither
require bulky and expensive stereo display nor highly
skilled stereo operators. However, monoplotting is only as
accurate as the underlying DSM. This will restrict it to
certain classes of application, of which there could be plenty
around.
Image matching supported monoplotting is another capable
mode for feature and object extraction. While the operator
positions the cursor in just one image on the feature of
interest the matching is performed, preferably in a multi-
image mode, in the background and on-line. A first
rudimentary solution of this kind can be observed on Vision
International's Softplotter.
3.8 Automation in general
The current level of automation on digital stations is fairly
low. The production of an orthoimage, given a digital image,
its orientation and a DTM, cannot figure under ,automation®.
We see first reluctant steps towards automation in the
following areas and on some few stations only:
(a) Interior orientation. Should work in general. Works in
some cases.
(b) Relative orientation. Is offered on some systems. Has
not shown to work safely under general conditions.
(c) Absolute orientation. Control points, no matter if
natural or signalized points, have still to be measured
manually.
(d) Triangulation. Tie point measurement is reported to
work largely automatically, but even in a major system
an error rate of 30% is reported. Control points and
new points have to be measured manually.
(e) DTM generation. Fully automated solutions are
offered, but results need much editing.
(f) Feature extraction, mapping. One of the most impor-
tant functions. Automation does virtually not exist.
In summary, what is unsually called , automation” has to be
put into the right perspective. There is no reliable black box
in photogrammetry. Many algorithms which are called
„automatic“ might work without operator interference, but
they usually require some, often many, parameters to be
set in advance. The results will then heavily depend on this
parameter selection. This is the point where either
photogrammetric expertise or longstanding project
experience has to enter the picture. Correctly, Colomer,
Colomina, 1994 remark: Furthermore, (semi-) automatic
software uses parameters for tuning results to different
types of projects. Although manufacturers provide default
values for these parameters and some hints on how to use
132
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
the
be
ap
fou
the
aim
ma
str
WF
rea
cor
an
imp
cor
the
wit|
tec
fun
site
of
for
Wi
par
anc
reti
anc
var
196
cor
res
ha
for
up
Sys
Dig
sci
sive
lea
