0.001, this error passes undetected. The bundle block adjustment is ex-
ecuted as before, without self-calibration. The estimated 6, is
1 and ba are 8.97 and
-17.88, respectively, which means that these parameters are sig-
nificantly different from zero. In a practical situation, this would be
all the information available to the photogrammetrist. He would decide
to use self-calibration because 6, is significantly different from the
^
g s 6.93 and the test values for the parameters b
expected go 4 ym.
Following this line of thought, a second run of the block adjustment was
done with self-calibration with parameter b. because it had the largest
test value. This resulted in 9 - 5.59 and again the same test values
for bi and b,. The other parameters of the Ebner set had no test values
significant at an a - 0.001 level. The computed parameter b, gives an
image deformation as shown in figure 4. The value of S is still large
compared with the expected e it is therefore reasonable to use
parameter b, too because it has a significant test value. This was done
in a third run of the adjustment. The result was then 6 - 5.20. The
contributions of the computed parameters bi and bo give an image defor-
mation as shown in figure 5. After this third run, the estimate 8
improved from 6.93 to 5.20 um; thus self-calibration did give better
adjustment results. There is no reason to use other parameters because
there are no more significant test values. The fact that 6 is still
somewhat larger than the expected 9,7 4 um might be explained by the
fact that the ground control was considered as not stochastic in the
block adjustment.
4. A SECOND LOOK AT THE RESULTS
These conclusions sound reasonable for practical situations where no
further information is available. Because we are dealing with a simu-
lated network and block, however, we can have a second look at the
results and analyse them a bit more. We therefore used the check values
as we did for the adjustment with the undistorted network. Residuals
were found by comparing coordinates computed from the block adjustment
with the distorted network with their check values. Figure 6 shows a
rather systematic pattern for these residuals without self-calibration.
The root mean square errors computed from these residuals are in terrain
scale in cm.
H H H E
X y z Xy
perimeter 37.15 11.23 18.52 27.44
inside 35.97 6.86 12.57 25.89
These are large indeed. Computing the same variates after self-
calibration with only b, shows no improvement at all. A graph of the
- 114 -