|A
t Monti
)s
Result, About 85 % of the pairs of coordinates had no partial correlation, The corresponding result fromthe Experiment
Reichenbach was no partial correlation.
The hypothesis no significant difference between inner and margin comparison points was checked with the F-test.
Table 6.24.
Result. No significant difference at the 0.1 % and 0.01 % levels between inner and margin points for the model 2 for
all three negative scales and for model 1 for the negative scale 1:12 000. For model 1 in 1:3 500 there occured
significant differences. For model 1 in 1:6 000 there was a significant difference for the point error but not for the
z-error and the error in space. In the Experiment Reichenbach 1962-1964 there were small differences in standard
error between inner and margin points, of little im portance in practice.
7. ERROR OF DISTANCE
The error of a distance between two comparison points was determined as the difference between lengths calculated
from geodetic coordinates and the corresponding photogrammetric coordinates. The lengths were sorted in classes with
50 m interval together with the corresponding errors of distance. All combinations of two points within the group of
all comparison points were calculated. Two series were chosen for study, one with the distances No 5, 10, 15 ...
and the other with the distances No 1, 6, 11, 16 .... The results are presented in table 7.1 and also graphically in
figures 7. 1, 2,3.
We normalized these figures in the following way. Distances were expressed in units of base lengths of the stereo-
models. The standard deviations of distances were expressed in units of the mean square standard deviation of all
distances for each negative scale. We plotted the normalized values and also the arithmetic mean of them. Figure
7.4. It is obvious that the standard deviation of distances increases linearly with distance, for distances shorter than
the base length, is constant for distances between one and two base lengths and decreases for distances longer than
two base lengths.
8. THREE DIMENSIONAL ERROR DISTRIBUTION
8.0 General Discussion.
Stereophotogrammetry deals with points in space. For that reason its error theory should be concerned with the simul-
taneous three -dimensional error distribution. In the Reichenbach Experiment the x y z errors of the same points were
found to be independent of each other, and not in all cases normally distributed and also not always having homogeneous
variance, The three -dimensional photogrammetric error -distribution is probably truncated, and as stated in section 6.2
hardly normal. Consequently it is hardly possible to study the problem theoretically under the assumption of independent
stochastic variables with symmetric normal distribution and homogeneous variance.
One possible solution is to study the Empirical Photogrammetric Error Distribution. About 50,000 errors of coordinates
are available in the Experiment, derived from 3 negative scales, each scale with 2 models, each model with about
50 points and each point with 3 coordinates. That will give about 9 000 pieces of data per model, 150 per point and