272
va OO
N U C0 Fh
{4 =
ro
A. ® 4 A. Fh (OD tT
[t]
273
only for a wide range of possible transformations of transfered images but
also for their enhancement (Real, 1982). Electronic image transfer also
eliminates the rigid constraints imposed by the optical trains on the overall
design of analytical instruments, and allows for a flexible configuration of
system components (e.g. the photo-stage configuration) and for easier
implementation of techniques for real-time superposition of graphics in the
stereo-model.
The overall organization of software packages and the optimal design of
application programs exploiting the best features of man-machine interaction
are still wide open fields for research and development (Slama, 1982). Some
unique, already developed, on-line techniques for compilation of time-
dependent imagery, for exploitation of multi-model solutions in close-range
photogrammetry and for processing of multi-media photography are good
indicators of the potential of analytical. instruments for opening up new
areas of application.
AUTOMATION OF PROCESSES
Some of the routines and techniques of on-line analytical photogrammetry
which do not require the operator's intervention, such as the routine for
automatic positioning of the measuring mark in predetermined positions or the
editing playback technique, may be considered as automation of processes. On
closer examination, it is evident that these simple automatic processes are
just a consequence of a direct computer drive of positioning devices without
the involvement of any decision making control processes or machine-
intelligence.
Taking in account the present status of the theory and technology in
artificial intelligence a full automation of all the processes needed to,
for example, extract from images all the required information for the
establishment of digital files of a cartographic data base must be deemed
intractable. For instance the pattern interpretation that is an integral
part of all the processes for feature compilation is still confined within
the boundaries of human intelligence. This is the reason that the only
practical way to deal with planimetric features is to generate an orthophoto,
that is, to perform a pictorial-to-pictorial conversion by automatically
transforming the metric information and leaving the semantic information
virtually unchanged.
The level of machine-intelligence needed for the extraction of information
on elevation is somewhat lower since it can be based on relatively simple
automatic image matching techniques. Attempts at automatic generation of
digital terrain models and digitizing of contours have been made by analog
electronic scanner correlators and by optical scanners using digital
correlation techniques. The goal of these devices is the measurement of
parallaxes in an oriented model and the consequent determination of elevation
or generation of a control signal for the positioning of stages. The image
matching in these devices has been based by and large on cross correlation of
two image density functions (Konecny and Pape, 1981). The cross correlation
function has a maximum when the images are shifted the correct amount
relative to one another.
These devices have proven quite successful under favorable conditions. But
they fail in cases of discontinuities (e.g. large scale images of built-up
areas), on steep slopes, and where a correlation is possible simultaneously
at different elevations (e.g. sparsely forested areas). These difficulties
may be overcome with the aid of the human operator, but the processes have