Sensing Additional Information
There is no reason, of course, to limit the effort to
improve the automatic stereoperception process to
direct improvement of the image-correlation process.
Techniques which allow other types of information
to be sensed and transmitted to the computer must
also be considered. Pattern-recognition techniques are
of particular interest in this respect, because they
tend to automate the photointerpretation function.
An accompanying papert describes research on a
pattern-recognition approach which employs an auto-
matic statistical analysis of the scanned video signals
from the flying-spot scanners and the video correlator
outputs and is thus directly compatible with the pre-
sent automatic stereoperception equipment.
A pattern-recognition system, integrated with an
automatic stereoplotter, might provide several addi-
tional features. One application is the automatic
recognition of obstacles such as trees and buildings
which produce appreciable elevation errors. A further
step would be automatic measurement of these errors
so that they could be compensated for within the
computer. These techniques would do much to ex-
tend the application of automated instruments to
large-scale mapping.
A pattern-recognition system might also provide
information which would assist the computer in con-
trolling automatic plotting of contours and profiles.
For example, automatic recognition of open fields or
water features might be used to provide more effec-
tive traversing of such areas, while recognition of
image patterns corresponding to sharp changes of
terrain elevation or slope might be used to more care-
fully plot these features.
An additional application for pattern-recognition
techniques is automatic extraction of the planimetric
data from the photographs. These techniques might
be used to replace or assist the operator in tracing out
individual planimetric features. A more fruitful area,
as discussed in an accompanying paper* appears to be
automatic scanning of the entire stereomodel in the
profiling mode, gathering data which can be pro-
cessed to directly generate color-separation manu-
scripts.
Expanding the Role of the Computer
Even though the computer in the present auto-
mated analytical stereoplotter performs major func-
tions in plotting motion generation and overall con-
trol, much of its potential for data storage, data
processing, and decision making remains untapped.
Examples of some of the more significant possibilities
in this area are:
12
(1) Increasing the sophistication of the plotting
control programs to handle more complicated
situations, particularly those encountered in
mountainous terrain.
(2) Providing more extensive processing between
basic model-coordinate data gathered in auto-
matic plotting and the output information
plotted on the coordinatograph.
(3) Preprocessing the stereomodel to gather and
store rough metric and image quality informa-
tion which can provide guidance during the
detailed plotting.
One of the best illustrations of more sophisticated
plotting control is the potential use of the computer
to provide automatic fill-in of the small gaps in the
plotted lines. In a typical situation the operation
would be as follows. Contouring in mountainous ter-
rain, the system plots accurately until a sharp ridge or
valley is reached. Then, as the system negotiates the
turn, a transient elevation error occurs, the stylus is
raised, and a small gap is generated. Generally in these
situations, the correlation system relocates the terrain
surface without computer assistance, but occasionally
the automatic search routine is employed to regain
correlation. In either event, the point where auto-
matic plotting is restarted is, by definition, a good
point; therefore, a special gap fill-in program could be
called on to plot in the reverse direction across the
gap until a point close to the initial point is reached,
and then return to the end of the gap and initiate
plotting in the normal direction. If the'system be-
comes lost during this operation, then it is not pos-
sible to automatically plot across the gap (the system
has attempted to do so from both directions and
failed). However, experience indicates that this strate-
gy would be highly successful in mountainous terrain.
Temporary storage of the model-coordinate plot-
ting data and automatic editing of this data before it
is permanently plotted would provide an efficient
means for improving the manuscript. This approach
would allow questionable information just prior to a
gap to be discarded. Contour data could also be
smoothed to eliminate the effects of system noise or
to provide a small amount of generalization of the
terrain shape, if desired. At the same time, the plot-
ting line could be adjusted to approximately compen-
sate for the effects of finite scan size in curving ter-
rain, and line gaps too small to require editing could
be filled in. This type of process might be performed
within the control computer by delaying the plotting
TR. M. Centner and E. D. Hietanen, “Automatic Pattern Recogni-
tion Techniques for Photographic Analysis."
*ibid.
CHAPELLE, WHITESIDE, AND BYBEE
foi
tiv
ta]
CO: