Table 4. Some parameters from the twin photograph Hasselblad adjustments
Degrees of | \Variance Factors | RMS Object space coordinate discrepancy, Coordinate | RMS Photo-coordinate residuals
Freedom: 64 standard eviation (mm) (um)
Correction | (1) OQ xp vo, zDl xoi vo |.zo Ixp.] xp Lxo | yo
070 | oso | o74 | 082 | 059 | 070
Raw Plane [1.316 | 1,305 [.970| 975 T 074] 082, [089 ro o ineo] nec [aee
076 .| 063. | 087 | 060. 074} 082
anser fs | oo rp om Ius Teale the
Mean 055 | 06% | 052 | oss | 071 | 049
Bilinear y 105$ | 1081 19851 | 099) | 073) | 086) (007 1 (075 V 9629| 2165 063516216
Local 045 | 053 | 045 | 045 | 053 | 045
puma |] 0997 | 0857 P572). | 99 065) | 032.1.093. 005 ], 551. | 190 [ 05: | 1.90
None of the sets of camera parameters estimated during
each adjustment series were significantly different from
any other set. The similarity between camera models can
again be explained by the relatively weak constraints on
the object space. Figure 6 demonstrates that the
photo-coordinate residuals are virtually independent of
object space datum. This independence does not hold true
for the estimated object space coordinates, as there are
significant systematic errors in the estimated object
coordinates when the survey control does not cover the
complete object volume. A similar effect could also occur
if, for example, an error was present in a surveyed object
coordinate.
There are no statistically significant differences between
the rms object coordinate discrepancies produced by the
raw photo and local bilinear data sets. However there are
significant trends in the estimated object coordinates of
upto 2mm with the raw photo data set, especially where
the control doesn't cover the complete object space. Again
it must be decided if such systematic effects will make a
significant contribution when the problems associated
with the plotting of natural features with line strings in an
analytical plotter are considered.
4. Conclusions.
Given a strong convergent network, image deformation
corrections can be insignificant if object space coordinates
are the prime requirement of the photogrammetric survey.
The pseudo photogrammetric survey was limited as
intended by weak network geometry. The table of rms
object coordinate discrepancies has shown that the camera
system is still capable of a respectable 1 part in 6,000 to
10,000 of the object space if film deformation is taken
explicitly into account. Such results agree well with Fraser
1982. If image refinement is applied, it must be applied
locally at every digitised position on every plotted
linestring. Results have shown that very localised changes
in film surface topology can occur so the degree of image
refinement possible at each plotted position will be a
function of reseau density.
The rms object coordinate discrepancies demonstrate that
the mean bilinear correction can provide better object
coordinate precision than the raw plate or affine methods.
Unfortunately the mean bilinear will not necessarily
remove systematic effects, such as the rotation seen in the
estimated object space coordinates (Figure 8). To remove
this type of trend the complete local bilinear approach is
required. If object space precision requirements are below
1 part in 5,000 it can be concluded that the extra
computational effort involved in carrying out image
refinement is unjustified given the physical properties of
this camera.
The inclusion of constrained inner orientation and
additional parameters in these weak adjustments have
allowed well defined camera calibration parameters to be
included. By this method it is possible to avoid some of
the problems caused by high internal correlation that is
often associated with self calibration and weak network
geometry.
Detection of systematic effects in the data, particularly in
the object space with limited networks is dependent on an
apriori knowledge of the physical processes occurring in
the imaging system. Limitations ofthe UMK camera show
that even a system designed specifically for its geometric
imaging properties can exhibit significant systematic
effects which cannot be removed by use of its reference
mark system. Attention to design of the complete imaging
system for survey requirements is the main factor which
can significantly reduce both the computational and
measurement effort associated with photogrammetric
survey.
5. Acknowledgements.
The author would like to thank Professor M. A.R. Cooper
of City University, Mr S. Brown, Dr. C. Forno, Miss A.
Kearney and Mr S. Oldfield of the Division of Mechanical
and Optical Metrology at the National Physical
Laboratory for their assistance and advice during the
course of this project.
6. References
AGFA (1989). Agfa Holotest Materials, Agfa-Gevaert,
Publication No. NDT1286/2012, NV.B-2510 Mortsel,
Belgium.
COOPER, M.A.R. and ROBSON, S. (1990). Methods for
High Precision Photogrammetric Monitoring of the
Deformation of a Steel Bridge, The Photogrammetric
Record, 13:(76)
