the results from the A7 and A8 autographs. The larger values for the other two instruments may be
caused by the lack of flatness of the diapositives used, at least for the stereoplanigraph. See preceding
section 1.342. The accuracy to be expected in the final results of the photogrammetric procedure can be
estimated from the root mean square values of the residual y-parallaxes according to the procedure
shown in reference 2.2:1. In section 4.4 the formulas for such tests are summarized.
Reference. 3.2:1. Bachmann, W. K.: Essais sur la précision de la mesure des parallaxes verticales dans
les appareils de restitution du 1er ordre. (Report from Comm. F of OEEPE.)
Photogrammetria XYI 1959—60.
3.3 Practical Tests of the Accuracy of Analytical Relative Orientation
The procedure shown in section 4.4 for analytical photogrammetry has been used for tests of the
analytical relative orientation with the requirement that corresponding pairs of rays shall intersect in
space. The measurement of image coordinates was made in the StK 824 stereocomparator and the com
putations were made in the FACIT electronic computer with a program by J. Talts. In general only two
iterations were necessary. The relative orientation was adjusted from discrepancies in 15 points and the
standard error of unit weight was consequently determined with 10 degrees of freedom. Known regular
errors were corrected. The residual y-parallaxes were simultaneously computed in a great number of
other model points which had been measured but not used in the adjustment. In the table below the
standard errors of unit weight are denoted s 0 is and the root mean square values of the residual y-parall-
axes in additional points are denoted s m . Contact diapositive images from five wide angle cameras
and one superwide angle camera were used. In two experiments with photographs from wide angle
cameras no corrections for regular errors were applied. The measurement and calculations made by
H. Middel, J. Tails and other assistants at the Division of Photogrammetry, Stockholm 70.
Table 3.3:1.
Camera
s ol5
S m
microns
microns
Ag3
4.1
3.1
Wide angle cameras c = 153 mm. Flying
Ag29
4.4
3.3
altitude 1200 m. Known regular errors of
Ag38
5.2
3.1
image coordinates corrected before com-
Ag41
4.9
4.5
putation
Agl96
6.7
4.3
Average
5.1
3.7
Ag3
5.6
5.5
No corrections applied for regular errors
Ag38
9.7
6.9
of image coordinates.
Ag56
6.7
7.9
Superwide angle camera c = 88 mm.
Flying altitude 2600 m. Known regular
errors of image coordinates corrected
before computation.
It is somewhat surprising to find that the root mean square values of the residual y-parallaxes are
generally smaller than the standard errors of unit weight. This may be explained by the weight variations
of the image coordinates as discussed in section 1.62. See fig. 1.62:9. The points in which the residuals are
determined for the computation of the s m are located closer to the principal points than are the
lateral orientation points.
It should also be noted that the standard errors of unit weight are in general smaller than would be
theoretically expected from the standard errors of unit weight of the image coordinates, the root mean
square value of which was found to be about 7 ; «m. See section 1.62. If the image coordinates of the two
photographs from which the model is formed were independent, the standard error of unit weight of the
y-parallaxes theoretically would become y2 times larger as ay-parallax is a difference betweeny-coordi-
nates of two images. But probably there are correlation effects which will reduce the standard errors of
the y-parallaxes as determined from the intersection condition.
In the next section the correlation coefficient will be determined from results of tests of single images
and of parallaxes.