(ts 
es 
ler 
lle 
be 
ve 
he 
50 
of 
1s 
T- 
he 
m 
er, 
us 
he 
us 
ce 
correspondences cannot be established reliably. 
Examples of applications: 
To show the potential of the method some appli- 
cations on different, more or less complex objects will 
be given in the following. The applications shall basi- 
cally show the versatility of the method independently 
from the fact that are competing methods for certain 
applications. 
A relatively simple application is a deformation meas- 
urement of a carbon wing panel (Figure 2). This 500 x 
300 mm? panel is put under load and deformations are 
to be measured on a regular grid and compared with 
finite elements computations at multiple load steps. 
The panel surface was marked by ~5000 projected 
dots, the coordinates of which were determined by a 
three-camera system and interpolated to the grid. 
Repeated measurements in unloaded estate yielded a 
standard deviation of about 25 pm for deformations in 
depth coordinate direction. 
  
Figure 2: Carbon wing panel put under load 
(10 times enhanced) 
As the projected pattern is moving relative to the 
surface during deformations it does well mark the 
surface, but it allows only for the determination of 
deformations in depth coordinate direction but not for 
the computation of local strains and shears. For this 
purpose a dot raster has to be directly applied to the 
surface with the advantages that also strains and 
shears can be computed and that the projection 
density and dot size can be adapted locally; however, 
the method can no longer be called a non-contact 
measurement method then. 
A significantly more complex object is given by the 
model car shown in Figure 3. The surface modula- 
tions of the car are significantly larger than those of 
the carbon wing panel, the surface shows discontinui- 
ties and the metallic paint leads to strong local reflec- 
tions. Moreover, occlusions or steep modulations do 
often cause dots being detected only in one image but 
missed in other views. 
In this application the method failed completely when 
Figure 3: Model car with reflecting surface 
only two cameras were used. With three cameras and 
~1000 projected dots the failure rate in the establish- 
ment of correspondences was about 1%, with four 
cameras it was decreased to less than 0.1%. In general 
wrong matches cause gross errors and can easily be 
detected in the surface coordinates dataset and filtered 
out as they appear as large peaks on the surface. To 
increase the spatial resolution and to be able to 
comprehend edges in a better way several exposures 
with the pattern phase-shifted were taken from each 
camera. 
  
Figure 4: Model car - 0.5 mm isoline plot 
The result of the surface measurement with 6800 
projected dots in total is shown in Figure 4. The 
model is incomplete in some regions because only 
one projector position was chosen. The discontinuities 
on the roof visible in the isoline plot show the ventila- 
tion slits and wing door splits. 
To be able to measure the surface of a complete object 
from all sides in one common coordinate system the 
method has to be combined with photogrammetric 
bundle triangulation methods. This has been done in a 
diploma thesis with the task to generate a surface 
model of a bust of Ludwig van Beethoven (Zanini, 
1991). 
The surface properties were well-behaved for the 
pattern projection (dull white gypsum material), the 
shape, however, can be seen as relatively complex. In 
total 12 projector positions with 4 camera positions 
each and some additional exposures for the connec- 
tion in the photogrammetric bundle triangulation were 
necessary. To be able to perform the method in a 
strictly non-contact manner connection points were 
signalized on a frame posed around the object.