ı and 
luring 
‘ation 
stems 
992). 
ment. 
to be 
ment 
- side 
sured 
ed to 
liper. 
great 
since 
s are 
g the 
  
  
307 
mirror was developed. Systematic deformations due to unflatness of the mirror are neglected yet, since they are assumed to be very 
small. They have to be investigated in the future. 
At the moment, no illumination is used but it is already in development. A ring of red LEDs will be fixed around the lenses to 
illuminate the interior of the mouth. No structured light or other artificial texture projected on the teeth will be used. The surface of 
the teeth will be reconstructed only using their natural texture by geometric constrained least squares template matching (Gruen 
and Stallmann 1991). Problems will arise for the deformation analysis since neither corresponding points on the teeth are known 
nor can they be marked. Thus every single tooth will be described by features that are extracted from the DSM. It's not necessary 
that a complete DSM is generated by geometric constrained least squares template matching. A DSM of the structured surface of - 
the teeth is sufficient. DSM points on the vertical parts of the tooth or points on the flesh are not necessary and very difficult to 
measure. On principle, only some reliable points on the sturctured surface of the teeth are needed to extract features describing the 
three dimensional orientation and position of ervery single tooth. By comparing the features, position and orientation changes can 
be derived. 
3. SYSTEM CALIBRATION 
The imaging system is calibrated periodically with a small white testfield of 150 x 150 mm size. It shows 49 black non 
retroreflective ring coded targets. The ring code corresponds to the point number. At the moment a maximum of 980 targets can be 
coded, but only a few are needed for DigiDent. Ring coded targets are detected automatically using a special decoder algorithm. As 
a result approximation coordinates for least squares template matching are derived. Final coordinates are determined using least 
squares template matching. In the final step only the center of each target is matched, using the same template for all points. 
The targets on the testfield are produced by CAD and printed with a laser printer on one sheet of paper. This sheet is fixed on a 
metal plate. The geometric performance of a laser printer and the flatness of the metal plate seem to be sufficient, that the testfield 
can be used for calibration purposes. Reference values are derived from the CAD-drawing. This means that the targets are not 
measured independently. Their accuracy has not been investigated yet. 
Eight images are taken from 4 stations with a roll of 0 and 90 degrees (Beyer 1992). Both cameras are calibrated independently 
using a set of 10 additional parameters. Non determinable parameters are excluded (Gruen 1986) This is very important since the 
testfield is flat. 
  
+ + ++ 
E ae "EE 
  
  
  
Fig. 3: Photogrammetric net design for calibration Fig. 4: Image showing testfield with 49 coded targets 
DigiDent shall measure position changes with an accuracy better than 100 mircons. This corresponds to % pixel in image space. 
Taking into consideration errors introduced by DSM generation and feature extraction used for deformation analysis the cameras 
have to be calibrated better than !/,, of a pixel. 
4. IMAGE DATA ACQUISITION 
Taking images from living objects for measuring purposes is very difficult. Especially taking images in a very close range like the 
human mouth is not simple and a lot of constraints have to be taken into consideration. The number of images has to be reduced to 
a minimum always taking into account the conditioning of the photogrammetric network. In this case only four images are needed, 
two of the upper and two of the lower jaw. The cameras are placed outside the mouth in a distance of approximately 150 mm. The 
cameras themselves are placed next to each other in approximately 100 mm distance. Inside the mouth between upper and lower 
jaw a mirror of approximately 50 mm x 70 mm size is placed to enable image acquisition. 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995