ding noise, 
je 
sis of warm 
) x 150 mm 
round were 
mages was 
ates were 
inate set of 
systematic 
ues for the 
0.018 pixel 
Is to 1/55th 
ntially every 
ate set was 
e excluded. 
differences 
0.024 pixel 
the frame 
imately 300 
‚on hour is 
  
| know how 
jon system 
stigation is 
analysis of 
i the frame 
e the test. 
were taken 
. reference, 
NS, that the 
3MS values 
pixel when 
  
ra 
3.3 Warm Up Effect of Frame Grabber 
Like every electronic device a frame grabber shows a 
warm up effect. Eight images were acquired within 2% 
hours. The camera was turned on several hours earlier. 
The last image is used as a reference and systematic 
shifts are excluded. Figure 3 shows that the warm up of 
the frame grabber is not finished after 1% hours. The 
RMS values are a factor 2 larger compared to the 
warmed up system. 
  
  
    
  
  
  
0,1 
0 
pond 
$01 
e 
i 
—@-— trend x [pixel] 
-0,2 —— trend y [pixel] 
—A— RMS x [pixel] 
—3— RMS y [pixel] 
-0,3 
0 1000 2000 3000 4000 5000 
time [seconds] 
Figure 3: Warm up effect of frame grabber 
3.4 Noise Analysis 
To analyse the noise of the imaging system a set of five 
images was acquired in video rate. The averaged image 
was used as reference. The RMS of greyvalue 
differences is less than one greyvalue. This result is 
suprisingly good. 
  
Figure 4: Noise of the imaging system 
3.5 System Calibration 
The system was calibrated using a small testfield of 
150 x 150 mm size that shows 49 black targets on white 
background. Image coordinates were measured using 
LSTM. The average image scale was 1:20. A total of 8 
Images was taken from 4 stations with a roll of 0 and 90 
degrees. Every target was imaged in average in 6 
Images. To compensate systematic errors a set of 10 
additional parameters was used. Non significant and non 
determinable parameters were excluded from the 
estimation process (Gruen 1986). A RMS of image 
coordinate residuals of 1/10th of a pixel was reached 
009 to 20 microns in object space. (1 part in 
249 
4. IMAGE ACQUISITION 
Due to many restrictions and constraints originating in 
the nature of the project image acquisition is very 
difficult. The problem can be divided into two parts: firstly 
image acquisition itself. This means how to place the 
cameras to be able to image teeth. Secondly the problem 
of illumination. The teeth have to be illuminated in a way 
that they can be measured by using their natural 
structure. 
In accordance to a method that has already been used by 
orthodontists to acquire images for documentation 
purposes a special mirror has been developed. This 
mirror is placed between upper and lower jaw. The 
cameras are placed in front of the mouth and acquire a 
mirrored view of a the upper jaw or the lower jaw 
respectively. They are approximately 250 mm in front of 
the mirror an placed beside in a distance of 60 mm 
corresponding to the base. At the mirrors backside 11 
retroreflective control points are engraved in the mirroring 
layer, so that they can be seen from the front. In addition 
they are placed in a special arrangement that shall 
minimize the overlay of control points and teeth in the 
images. Since the control points are retroreflective they 
are imaged as bright targets, the same way like strong 
reflections. By eliminating the influence of strong 
reflections on the measuring process the influence of 
overlaying control points can be eliminated too. 
The illumination system consists of two parts: red LEDs 
that are fixed around the cameras lens and an additional 
diffuse illumination of white light. The red LEDs are used 
to illuminate the retrotargets fixed on the mirrors 
backside. The teeth’s surface is illuminated by the diffuse 
light. 
     
     
    
camera 
  
      
mirror 
camera 
   
mirror with teeth e 8 e e 
control 
points @ b 
@ 
Y 260mm 
o e 
cameras 
e @ 
  
  
60mm 
Figure 6,7: Camera arrangement, Control points 
5. DSM PROCESSING 
For every tooth of a jaw a DSM (digital surface model) is 
generated. The process can be divided into three steps. 
Firstly all image points that belong to one single tooth 
have to be defined, secondly the generation of the DSM 
and thirdly postprocessing including detection of gross 
errors and smoothing of the DSM. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996