Karsten Jacobsen
5 MODEL SETUP
The absolute accuracy is only one result. For the model setup, the relative accuracy, represented by the y-parallax, is
more important because a large y-parallax is preventing the stereoscopic view. The relative accuracy of the ground
coordinates computed by common intersection - the accuracy of a point in relation to another point in the
neighborhood - for distances up to 500m is RSX=0.19m, RSY=0.10m and RSZ=0.36m. This corresponds to a
mean square influence of 42pm which is exceeding the limit of the tolerance for the model set up, but the relative
accuracy is only indicating and does not say directly something about the y-parallax.
Also the neighbored orientation values are correlated, for phi the correlation of neighbored orientations is r=0.81,
for omega r=0.57. This leads to a relative accuracy of RSphi=0.011 grads, RSomega=0.010 grads and
RSkappa=0.005 grads. For the relative orientation especially omega is important. Just the value for omega is
reaching 0.010 grads e focal length 305mm = 53pm, exceeding the tolerance limit of 30pm for the model setup.
In the check areas the y-parallaxes have been checked by intersections based on the direct determined sensor
orientations and the available image coordinates. The mean square value of the y-parallaxes for all intersections is
56pm, that means it is too large for the model setup. 64% of the models are exceeding a mean square y-parallax of
30pm, which can be used as a tolerance limit, 36% are exceeding 50pm and 6% are exceeding 100pm. This shows
the problems connected with this data set. If it is compared with other published data sets, it has to be respected that a
normal angle camera has been used. The influence of the attitude data to a wide angle camera is only half of this.
Figure 9 and 10 are showing on the left hand side the y-parallaxes of the intersection of 2 models based on the direct
determined sensor orientation. The dominating constant shift is obvious - this is typical for all models. By this reason
Figure 9: y-parallaxes model 188/189
| hoi à E or (manual measurement)
| l left hand: intersection based on direct
D [ determined sensor orientation
| i mean square y-parallax: 141 um
| : right hand: y-parallax after shift of one
l , I image in y-direction
mean square difference: 4 um
Figure 10: y-parallaxes model
11050/11051
(automatic aero triangulation)
left hand: intersection based on direct
determined sensor orientation
mean square y-parallax: 132 um
right hand: y-parallax after shift of one
image in y-direction
mean square difference: 15 pm
: !
200um m 4, Ma 20um
PL
the mean shift has been removed, leading to much better results. The mean square value of the y-parallaxes for all
intersections is reduced by this from 56pm to 14pm. No mean square shifted y-parallax for a model is exceeding
30pm and only 33% are exceeding 20pm. That means, if the models shall be set up, a simple method for a y-shift of
one image against the other can solve the problem. The direct determined sensor orientation have to be transferred to
the different photogrammetric workstations, which is possible by the different transfer programs included in program
system BLUH. Based on measured y-parallaxes only the projection centers of the transfer parameters have to be
changed. Of course it is better to include this as a function in the workstation software, but this usually has to be
made by the different photogrammetric software companies.
434 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.