32
3. THE RELATIVE ORIENTATION
This operation is founded upon the projective requirement that corresponding rays in overlapping
bundles shall intersect in space. The accuracy of the relative orientation affects the geometrical shape
of the model, whether expressed numerically or in some other manner. It is also noteworthy that the
accuracy of the relative orientation is a good, but not complete, indication of the accuracy of the pre
ceding operations. There are different methods in photogrammetric practice for the correction ofy-parall-
axes in stereoscopic plotting instruments. It is of practical interest to compare the precision of the most
common methods. K. Torlegârd, MSE, has conducted comprehensive tests and some of his results are
presented here for the first time.
Detailed investigations of the geometrical quality of the relative orientation have been made. See
reference 1.342:1. It seems particularly important to have such tests included in the international
controlled experiments as conducted by commission IV I.S.P., as this is an excellent opportunity to
test the error propagation from the relative orientation to the final results of the procedure. Results of
relative orientation tests from actual controlled experiments are shown below, as well as results from
analytical relative orientation.
3.1 The Precision of Various Methods for y-parallax Correction
These investigations were performed by K. Torlegard, MSE, who will publish a separate, detailed
report. A brief summary is given here. Nine operators took part in this experiment and three different
methods for correction of y-parallaxes in stereoscopic projection instruments were used viz.:
1. Stereoscopic correction. The Dove-prisms are used to rotate the images optically and tliey-parall-
axes are transformed into ^-parallaxes before correction using the stereoscopic setting of the measuring p
mark.
2. Monocular correction. The Dove-prisms are not rotated. With the left eye a detail of the left image
is brought to coincide with the left measuring mark. Then with the right eye, the same coincidence is
required between the right measuring mark and the image.
3. Binocular correction. The Dove-prisms are not rotated. Measuring marks and images are observed
with both eyes at the same time and the floating mark is made as distinct as possible by the y-parallax
correction.
The nine operators determined the precision of the three methods of y-parallax correction from
replicated observations in large scale terrain models and grid models in several instruments. Twenty-
five replications were made and the precision was expressed as the standard deviation of one correction.
In the diagrams following, the operators are numbered on the horizontal axis in order of skill as determin
ed by the root mean square value of all standard deviation for each operator.
The figures 3.1:1 through 6 in the Appendix show the precision as standard deviations of oney-parallax
correction in terrain models. Three model points of differing clarity were used. For each standard devia- ?
tion the confidence interval is also indicated. Fig. 3.1:7 shows the root mean squares values of the standard
deviations taken from the preceding diagrams. In fig. 3.1:8 the results taken from the grid model tests
are shown.
The following is concluded from this investigation. The stereoscopic and monocular correction me
thods appear to give better results than the binocular method for most operators. The more skillful
operators have better results from the stereoscopic method, and the less skillful operators have better
results from the monocular method. Training and familiarity with the instrument is of course important.
The root mean square value of the standard deviation of one stereoscopic y-parallax correction in a t
model point of high quality is about 2 am, monocular correction about 3 ( «m and binocular correction
about 5 y/m, see fig. 3.1:1. Of course the standard deviation of the average of replicated corrections
decreases linearly with the square root of the replications.
