Moire fringes on the structure, but the method may be 
used on a few points only and therefore is not suited for 
this application. Photogrammtry is on the contrary an 
ideal complement to strain gauges, since it provides with 
a dense field of 3D deformation vectors, allowing a 
precise reconstruction of the panel surface at any stage. 
In each loading sequence intermediate configurations 
may be recovered while the load is increased in steps. 
The initial panel shape must also be known, to derive the 
true boundary binding state and stress. The required 
accuracy in the direction normal to the panel surface was 
set in the order of 0.1 mm over the whole panel. 
2. THE DESIGN OF THE SURVEY 
Thanks to the simplicity of the task (no semantic 
involved, just mass point determination on a smooth, 
though very poorly textured, surface) the advantage of 
using an all digital solution (i.e. using digital cameras and 
automatic surface reconstruction) for the panel survey 
was self-evident. In principle, even a near real time 
solution might be achievable (though not required), if a 
sufficient number of cameras and enough computing 
power were available. We opted rather for a mixed 
approach, using an analogue camera for image 
acquisition, followed by scanning and processing of the 
digital images. The main reason for that choice was cost 
containment: since most of the funding allocated to the 
project was used to buy the loading machine, we were 
forced to use the cameras available to our Department, 
all of them still analogue metric or semimetric cameras. 
Given the accuracy goal, a standard digital camera 
would have required a more complex network design, 
while an high resolution camera would have been too 
expensive. For much the same reason, scanning has 
been performed by an inexpensive off-the-shelf DTP 
scanner, with maximum resolution of 800 dpi. 
Even accounting for the additional burden given by the 
scanning process, performing the measurements 
automatically on digital images is clearly much faster 
than tackling the task by a human operator, hardly 
motivated by repeating thousends of times the same 
operation. 
Among the available cameras, we choose the Rolleiflex 
6006 with a 40 mm lens and a 11x11 reseau, for two 
main reasons: 
e fast operation: in each static condition the load is 
maintained by an hydraulic system whose stability 
proved to be not very reliable; the shots relative to 
each loading stage should be therefore completed in 
a very short time, what is hard to achieve using 
metric cameras such as the Wild P31; 
e scanner deformations: since our scanner may 
introduce considerable distortions, we preferred to be 
able to estimate corrections on each image, taking 
advantage of the reseau. 
Simulations were carried out to determine the number of 
stations, the image scale and the degree of convergence 
of the camera axes. The main constraint on the design 
has been the reduced depth of field, which limited the 
minimum incidence angle to the object plane. We were 
also concerned of a possible decrease in the accuracy of 
ls. matching in those areas where the perspective 
deformation heavily shrinks the target. Three basic 
configuration were selected, all of them consisting of 4 
210 
images, taken simmetrically, each covering the whole 
object (see figure 1): 
- a normal case, with horizontal and vertical baselines of 
40 cm and minimum distance to the object of 1.2 m; 
- a slightly convergent case, with horizontal baseline of 
120 cm and vertical baseline of 80 cm and same 
distance as above; 
- a strongly convergent case, with horizontal baseline of 
150 cm and vertical baseline of 70 cm, with minimum 
distance to the object of 70 cm. 
Moreover, a fifth nadir image was taken to simplify the 
target localization procedure (see 5.). 
Assuming an accuracy on the pixel coordinates of 1/20 of 
the pixel size, the photoscale would not have been 
enough to ensure the accuracy on object space at the 
maximum scanning resolution. The negatives where 
therefore enlarged by approximately a factor 2.5 and the 
printed copies digitized. 
The panel surface is poorly textured and therefore 
correspondencies would be hard to find either for 
humans as well as for algorithms. A solution might have 
been using a light projector capable to create some 
pattern on the surface; we preferred instead to signalize 
the panel, making easier, if any, to compare pointwise 
the behaviour of the structure subject to different loading 
conditions and to use template matching rather than 
matching with respect to a reference image. 
Taking into account the expected frequency components 
of the deformation surface, a regular square grid of 
targets (spaced 15 mm) was prepared on adhesive paper 
and fixed to the panel, totalling around 2700 point. The 
targets simply consist of a black circle with a diameter of 
6 mm on a white background and have been prepared on 
a laser printer. Their size was computed, after selecting 
the photoscale and the scanning resolution, to ensure a 
target on average 20 pixel wide. 
3. SCANNER CALIBRATION 
As mentioned above, a UMAX UC840 Max Vision 
scanner has been used; its main the characteristic are as 
follows: 
- scanning format. 216x356 mm; 
module; 
- radiometric resolution: 8 bit; 
- max optical geometric resolution: 400 dpi across scan, 
800 dpi along scan direction; up to 1600 dpi in 
interpolated mode; 
- internal buffer: 2 MB 
The software driving the scanner allows for the standard 
grey value transformations (contrast, brightness, gamma 
correction, histogram equalization, etc.) and for freely 
defineble scanning area. In order to assess possible 
distorsions of the scanning process, radiometric and 
geometric properties of the scanner have been 
investigated. 
no transparency 
3.1 Radiometry 
The g.v. profile through the image of a target has been 
acquired at different times after the power was switched 
on. The maximum g.v. changes are in the order of 15% 
and occurred in the first minutes. The repeatability of the 
g.v. profile, after the warm-up effects vanish, is better 
than 1%. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996 
A radiome 
grey wedc 
sensitivity 
assess the 
tests have 
with and ‘ 
object co 
correction: 
3.2 Geom 
Since our 
module, 4 
reference 
spacing oi 
size of our 
The coord 
Zeiss corr 
digital ima 
coordinate 
which ma 
directions: 
estimated, 
coordinate 
performed 
computed 
later. Non 
of the affit 
be recov 
orientatior 
Figure 2 
Figure 3 
Figures 2 
the X (ac 
can be se 
uniform in 
pixels. Re