glass plate flat in the image plane and are probably coupled
with humidity effects. In a twin camera survey without
apriori knowledge of all locations in the object space it is
difficult to see how systematic effects such as those in
figure 3, could be detected and modelled reliably within
the adjustment process especially given only four fiducial
coordinates. A possible approach would be to stabilise the
physical properties of the imaging system and to derive a
physical model which could be applied either before
adjustment or by the method of prior constraints. An
indicator of the gross departures from the model could be
based on a knowledge of fiducial mark discrepancies
occurring under calibration conditions.
An interesting question can be posed at this point; are the
object coordinate discrepancies significant in a real
situation where we are interested in plotting three
dimensional line strings? Results have been based on high
contrast target images produced using high resolution
emulsion, not natural features which tend to be less
distinct. Also measurements were made to below lium
rms as compared to a conventional analytical plotter with
a typical rms of 3 jum. Such considerations mean that any
accuracies predicted by network design must take into
account the pointing ability to real features. It must then
be decided if systematic effects due to departures of the
emulsion surface from the assumed image plane are likely
to be significant.
3.2. Adjustments featuring the modified Hasselblad
SWC camera.
Three sets of adjustments were computed from the
Hasselblad photo-coordinate data. The first set included
all 8 images with a datum based on the first simulated
survey control set, The other two adjustments used just the
top left and right photographs ("3" and "4" in figure 1) in
conjunction with the first and second survey control sets
respectively. Each adjustment was further divided to
include the four methods of film deformation correction
described previously. In total there were 12 adjustment
permutations.
3.2.1. Results from the 8 photograph Hasselblad
camera adjustments.
Within this set of adjustments, the only variable was the
method of film deformation correction applied. In such
cases the variance factor can be used as a crude global
descriptor of film deformation correction effectiveness.
Table 3. Some parameters from the 8 photo
Hasselblad Adjustment, Set (1)
Dess RMS Object Coordinate RMS
; : i Photo-coordinate
Freedom [Variance discrepancies and
758 Factor Standard Deviations (mm) residuals (p.m)
Correction X Y Z X y
0.21 0.27 0.26
; 2.69 2.03
Raw Plate | 1.642 (0.81) | (0.88) | (0.76)
0.19 0.33 0.23
Affine 2.547 (1127) | (1.44) | (1.16) 2.95 3.44
Mean 0.28 0.26 0.21
. : .96
Bilinear 1713 (1.06) | (1.20) | (0.97) 249 2
Local 0.25 0.26 0.19
s .206 2.18 2.68
Bilinear H2
Hasselblad 8-Photo
Affine
3 4
AN i5, \ t
=» A ’r AJ +
7 4 > da d ye +}
l \ + / Saas LA
bobine CS
LM 7, $4 Y
x^ pot. 7 /
A as
10 \um JO um
Local BllInear
3 4
m} = + vol
X x tt
NS y > v / cvs +
et à Ard
| A 1 +, ANY , à
X X +
/ Ve X
a
+ T, \, >
ye spn TR /
Lory :
10 um 10 um
Figure 4. À selection of photo-coordinate residuals from
the 8-photo Hasselblad Adjustment.
The data set which utilised the affine transformation has
produced the largest variance factor, and arguably the
worst solution. À probable reason for this is that larger
image deformations occurring at the frame edges have
made a significant contribution to all photo-coordinate
corrections. The mean bilinear method has also suffered
because atypicalities between frames have lead to spurious
photo-coordinate corrections.
Ob Ject Space DIscrepancles
Front Vi
Affine C Cu Local BllInear
= 5 = uos T d \ = L — M €
cn Ni" VAR sens
x \ I 1 NS — 2S —
-— -- > i emm = ~ e
dr X - Ta = 7
i x eM OA 7 th m =
x Ed Fo # | 2X A
eT \
—
2 mm
Aff Ine Top View Local Blllnear
AREE: K ker
^A) | v V
Figure 5. Object Space discrepancies from the 8-photo
Hasselblad Adjustment.