10
illustrated in Fig.8.
In the case of 6 points system, one point is only redundant. And if
this one point happens to have errors which approximately coincides with
the necessary vertical displacement of that point for the appropriate
orientation elements computed by the remaining 5 which is also erroneous,
then the residual vertical parallaxes become very small. We are using
the standard deviation of the residual vertical parallaxes as a macroscopic
measure specifying the quality of the model constructed. But this measure
becomes meaningless in the above erroneous case, and it appears that we
encounter the case quite frequently. It is clear that three redundant
points will scarcely have the errors simultaneously which shall lead to
the result making the standard deviation to be meaningless as in the
case of 6 points system.
We use the standard deviation of vertical residual parallaxes as the
primary measure for the discrimination of the case to be remeasured or not
in practical work. In considerably long period we could not lower this
range less than 20 microns, but at present we are able to lay the limit
almost at 15 microns. This is clearly due to the level up of the skill
for marking of points to be measured and the technique of measuring them.
Here we will show some statistics about the relative orientation. It
is too much laborious to make statistics of the whole data obtained in
these 4 years and we will only make it on the confined data which were
obtained by the analytical aerial triangulation executed in August -
September 1963 an< * in February - March 1964 and are yet available in hand.
They are composed, for the former, of 68 strips and 653 models and for
the latter, of 43 strips and 308 models. In Fig.9 the standard deviations
of residual vertical parallaxes for every model are gathered as frequency
curves for the above two groups. It seems that this curve shows a slight
progress of the latter case in accuracy, and in both cases about 80% falls
in the range less than 15 microns of standard deviation.
Now, about the errors of x-direction. Effects of the error of x-
direction on vertical parallaxes are, as well known, only of a small
amounts, and are not detected by vertical parallaxes. They can be detected
only by discrepancies at minor controls common to adjacent models. They
are expressed in our system by a unit of per mil of the distance between
the projection center of the common photograph and the point concerned and
are printed in table between adjacent two models by computer. The amount
of discrepancies at present seems to be in average about the order of 0.15