background image texture. The network consisted of 28 images 4.2 Densification results 4.3 
acquired from two cameras sweeping two curves of image 
stations at fixed positions (Figure 6). The data sets with the pattern and white light projection were Mea: 
M c processed using the densification algorithm described in $3. using 
The laser dot projection data set covered most of the area comi 
viewed by the image network (Figure 5). This dataset was point 
processed in VMS to provide a comparison against the point to us 
cloud measurements derived from the other two datasets. as. e 
targe 
vuv prodi 
PETTITTE sumr 
sigm 
gives 
estim 
of th 
poste 
. p — vas 131 
Figure 2: Laser dot projection data set M pee n 
The second data set involved imaging the object using The r 
approximately the same image network camera locations as Figure 5: 3D view of image network and laser dot targets expec 
previously. In this set a pattern was projected on to the object by th 
surface using a digital projector (Figure 3). The illumination simul 
conditions were appropriately modified to allow acquisition of por datasets two and three the densification process was target 
the pattern information as well as the retro reflective targets. initialised from twelve retro reflective targets and a single is at 
The pattern Was projected to enhance image texture iteration was sufficient to produce a dense point cloud (Figure meas 
information, which is a common technique used in a variety of 6). The densification process was initiated in two areas of the also | 
photogrammetric measurement systems (Siebert & Marshall, gearbox, for which CMM measurements were also available for 
2000; D'Apuzzo, 2002). accuracy comparison purposes. 
Targ 
All p 
Table 
datas 
4.3.2 
The s 
target 
2). TI 
overle 
illumi 
Figure 3: Pattern projection data set 
The third data set was acquired using approximately the same Targ 
camera network configuration. The object surface was All p 
illuminated by projecting white light from the digital projector, Table 
thus producing a similar effect to applying strobe light datase 
illumination (Figure 4). This set-up provided images with 
natural image texture content. 433 
The t 
concei 
The re 
(2) set wi 
Figure 6: Densification of initial target triangulation for target 
projected pattern data set (1) and white light data set (2) combi 
a-prioi 
In both data sets the densification process has provided points 
in parts of the areas of interest. Lack of points in some parts is — 
attributed to the shadow and occlusion problems arising from Tar 
the complexity of the underlying surfaces as well as the All pr 
limitations of the image network. Table 
  
Figure 4: White light projection data set 
70