Based on experiences with numerous test cases similar to those described
here, a design feature of the current program is a strategy definition
option, in which the user may define the sequence of steps, the
acceptence/rejection criteria, and the window size/shape parameters which
will be used for the estimation of each data point. This strategy
definition can be modified to suit the image texture characteristics, and
the terrain characteristics for a particular project, or set of
photographs.
With regard to item (2) there are several possibilities which are being
studied. One could possibly pre-sample the photographs before scanning for
correlation, to determine which areas have sufficient information content
for matching. In order to better interpret data which has been collected
one could record a "figure of merit" to indicate the ccnfidence that we
have in the elevation value. This could be used for weighting in a post-
processing interpolation step to strengthen weak points or to fill in
where the quality of correlation was very poor. A threshold level of this
figure of merit could be used to trigger a manual re-observation of the
point in question. This would fit well into a strategy where an entire
model is scanned without operator intervention, and then all of the manual
assists are performed at the conclusion of the automatic scanning. Another
possibility which is being considered is to manually observe a sparse DEM
grid over the model prior to scanning for correlation. The scanning
program could then obtain, by interpolation from the sparse data points, a
reasonably gocd approximation of the terrain height at each of the final
grid points.
CONCLUSIONS
The tools are at hand for applying the techniques of digital image
processing to numerous measurement tasks currently performed manually by
stereo operators. The design of the DSR11 has permitted easy integration
of video acquisition hardware. Such hardware is extremely flexible, and
new applications usually call for only new software, without modifying the
instrument itself.
The ACOR system has been shown to produce good terrain elevation data when
the requirements of image definition and terrain slope are met. Many
enhancements and alternate image matching techniques are being actively
studied both within the Kern company and by other independent researchers
working with this same equipment. The system, which is already usable for
much photographic source material, should become even more accurate and
reliable.
REFERENCES
1. Cogan, L. and Hunter, D., "DIM Collection and the Kern Correlator",
ISPRS, June 1984
2. Bethel, J., "The DSR11 Image Correlator", ASP Proceedings, March 1986
568