EXPERIMENTAL RESEARCH ON SEVERAL TYPES, ETC.
9
dently on the measure errors. In fact, in computation 6-3, the use of three heights
for the scale transfer brings about mean residual parallaxes almost three times as
large as the preceding ones, while, going from computation 18-1 to 18-6, such an
increase is negligeable. The value we can obtain for residual parallaxes is evidently
the 9 cm (7,5 g m) corresponding to computation 18-1. If, instead of 18 points
for orientation, we had used twice as many, we would have probably come to
the same result.
The x deviations are, in any case, very small and in the four types of bridgings
no meaningful increase can be noticed. If we analyse also the 8 Y and 8 Z, we can
remark that between 6-1 and 18-1 the increase of accuracy in the model formation,
due to the better relative orientation, is very small : less than 10 %, whereas, in the
theoretical research, we had found that the accuracy of the orientation elements of
the cameras reached an increase of 42 %, when we used 18 instead of 6 orientation
points. This difference can be explained bearing in mind that the experimental
results refer to the model coordinates and not to the cameras orientation, and that
in the theoretical study, the mean measure error of the plate coordinates had been
thought to be independent on the position of the points : in practice this fact is
certainly no longer true. Furthermore, the experimental observations depend not
only on the purely accidental measure errors, but also on the systematic errors due
to the imperfect correction of distortion, to the lack of flattness of the plates surface
and so on. This question will be further examined in the last paragraph.
The deviations are definitely reduced in those bridgings in which the scale
transfer was made by using several points placed in the overlapping zone between
models. In fact, from 6-1 and 6-3, there is a decrease both in 8 Y and 8 Z, that
almost reaches 40 %. It is obvious that this result is due to the method we employed,
that determines the orientation elements also in function of the connection bet
ween subsequent models. By the way, the consequence of eventual errors are far
different if, for the scale transfer, we use more points instead of one. We could
suppose that this procedure will bring about a remarkable deformation of the
model; on the contrary, we can see that the continuity between models is obtained
through a negligeable increase of parallaxes.
The opportunity of using 6 points for scale transfer is quite questionable :
from 6-3 and 18-6 we have an improvement of about 15 %, partially due to a better
formation of the model, partially due to the three more points used for the scale
transfer. However, we must obvioulsy inquire whether a 15 % improvement is
worth the larger time required by measures and computations. In fact, the measu
res of 27 points on each plate (and, of course, of all the known points and of all
the pass points between strips) has required about two hours and a half for each
photogram, whereas, had we measured only six, as it is usually done, half an hour
would have been enough. As far as the computation is concerned, we can assume
that bridgings 18-1 and 18-6 require, on the average, computations three times
as long as bridgings 6-3, 6-1.
In figures 2 and 3, we have graphically show all the values of the mean paral
laxes and deviations relative to strip 1 B (see table 1) and 3 A. This latter strip is
interrupted toward the end, namely, there is a zone in which the overlapping bet-