The Influence factor
2
max
M -A Ey]
(Sy = Eu) (17)
measures the maximum increase x — Xy of the uncer-
tainty 315; of the estimated parameters y, if the observati-
ons group x; is omitted with respect to X55. It reduces to
pu? = (1— rj)/ri for single observations [cf. BAARDA W. 87].
The empirical sensitivity
6; = T; - wi (18)
measures the maximum influence of the observations group
+; onto the estimated parameters. If this group is omitted
an arbitary function f = a” - y of the estimated parameters
y with variance 0 = a"E,ya does not change more than
(19)
A great value indicates the observation group z; to be neces-
sary, namely the control point model i to improve the result.
In case the value is small the result is determined reliably
as the control point is checked by the others and the control
point model only slightly influences the result.
Vif «€ "4e;
The theoretical sensitivity
6o = 60 . Hi (20)
(external reliability according to BAARDA) gives the maxi-
mum influence of undetected errors in observation group i
onto the estimated parameters. The influence of an undetec-
ted error in the observation group i is bounded by
Voi f < 6o 07 (21)
with 60 depending on the significancd level and the required
power of the test, we use 69 = 4.13.
Small values indicate that an error not detectable by the ro-
bust estimation has an influence onto the result that can be
neglegted. Large values indicate a inacceptable control point
arrangement.
9 Conclutions
An automatic outer orientation procedure of aerial images
based on 3-D wireframe models of natural control points was
presented. The matching procedures for finding correspon-
dencies between image and model features, namely straight
line segments, have shown to be robust, with respect to
wrong or missing correspondencies. The examples demon-
strates the feasibility of object location whith this approach
even in case of weak image information, being the normal
case in natural scenes especially aerial images.
The sensitivity analysis applied to groups of observations
here used as a means for selfdiagnosis has shown as a po-
werfull tool for the automated system to be used in practice.
The program system AMOR has been tested on 32 aerial
images. In 5 cases the sensitivity analysis correctly supposed
a weakly determined configuration, though the orientation
parameters were correct. In 11 cases the clustering has made
à wrong correspondence which has correctly been detected
by the robust estimation, therefore the orientation parame-
ters has been correct. In the other cases all the control points
596
has been correctly located. The program system AMOR will
be implemented this year into the automatic orthophoto pro-
duction system at the Survey Department Bonn.
Though the actual procedure is optimized with respect to
the task of automatic outer orientation of aerial images, the
concept and most of the modules, especially the matching
procedures, may be transferred to other applications. In sec-
tion 6 an example is presented using the presented matching
procedures for semiautomatic mapping.
6 Examples
Example 1 :
Detecting bad geometric control point configurations via
Selfdiagnosis.
^y
ft
OB
[2^ 7
H o8 Es"
O 4
5
Fig. 4 Configuration of 8 control points in an aerial image
^ y!
"x
Fig. 5 Configuration of 5 control points in an aerial image