The rounded corners of the inserts cut the worked metal 
too, therefore it is also necessary to measure these corners. 
The assignment of edge pixels in the upper and the lower 
image can be done with the help of epipolar lines. As the 
accuracy for single points is not sufficient the edge points 
of the corners have to be adjusted by a circle, whose radius 
can be taken out of the construction drawings. 
The positions of the inserts are determined in the coor- 
dinate system defined by the reseau grids. They have to 
be transformed into the tool coordinate system which is 
defined by the rotation axis. As the tools are rotating dur- 
ing the manufacturing process the angular position of the 
edges is not as important as their distance to the rotation 
axis. The determination of the rotation axis of the tools 
has been postponed because this is a standard task and 
it requires a high precision clamping device for the tool 
on the rotation table which was not available. It can be 
easily performed by measuring a single rotating point at 
different heights. The centres of these circles lie on the 
rotation axis. 
5. Reached Accuracy 
In the field of industrial measuring technique it is a com- 
monly accepted way to check the quality of measuring 
devices with end measures which are high precision steal 
blocks. They have two opposite sides with lapped measur- 
ing surfaces. The width of the end measures is manufac- 
tured with an accuracy of less than a tenth of a micron. 
With optical means it is impossible to measure surfaces of 
polished steal without any structure or targets. The edges 
of the end measures are not sharp enough to represent 
the surfaces. Therefore two attachments were taken from 
a measuring microscope and put at the end measure as 
shown in Fig. 8. 
  
  
end measure 
  
measuring 
edges S: 
77 
N 
  
  
  
Fig. 8: test object 
The wedge-shaped attachments have also a lapped measu- 
ring surface which ends in a very sharp edge. The distance 
between the two edges corresponds to the nominal value 
of the end measures. 
Two end measures of 30 mm and of 100 mm length were 
measured repeatedly at different location within the object 
space. As this procedure means measuring a straight line 
in images with known orientations, the test object could 
not be measured in a horizonta! nosition. In this case the 
measuring edges lie vertical, that is parallel to the base 
of the cameras, and the epipolar lines are almost parallel 
to the edges, too. As there are no distinct points there is 
an ambiguity which can only be solved if there is further 
information available, for example the intersection of two 
edges. With triangle shaped inserts this is always the case. 
   
Table 4 shows the results of the repeated measurements. 
  
  
30 mm in | 309mm . [100mm in 
vertical tilted vertical 
position [at 50 gons | position 
accuracy of 4.4 4.2 3.7 
a distance 
accuracy of 31 3.0 2.6 
a single edge 
  
  
  
  
  
  
Tab. 4: results for end measures [um] 
Those values were evaluated out of the differences between 
the measured distances and the nominal value that means 
they give the absolute accuracy. 
This accuracy could only be reached because the sharpness 
of the edges allows a precise edge detection in the images. 
Under these conditions the illumination which is normally a 
critical point for the image processing was easy to handle. 
For round edges the direction of the light rays and the 
position of the cameras might influence the results. This 
point demands further investigations. 
6. Conclusions 
The investigations described above show that it is possi- 
ble to apply photogrammetric methods in the industrial 
measuring technique. Metal edges of cutting tools can be 
measured with an accuracy of 3 um if the edges are well 
defined. 
The automation of the system has still be completed, that 
means the automatic positioning of the cameras and the 
automation of the illumination. It should also be possible 
to measure differently shaped cutting edges. 
As a future step the integration of the system in the ma- 
nufacturing process could also be taken into consideration. 
References 
Jacobsen,K., Hielscher,H., Husen,B., Benter,U., 1990, Pre- 
cise Object Determination By Digital Macro Photo- 
grammetrie, ISPRS Commission V, Zürich 
Pfeifer, T., Fürst, A., Vollard, W., 1982, ProzeBintermittie- 
rende Werkzeug- und Werkstückmessung auf NC- 
Werkzeugmaschinen, Industrie-Anzeiger 104, Nr. 27 
Wang, Y., Jacobsen, K., 1991, Model Based Fast Reco- 
gnition of Industrial Workpieces, 13th DAGM Sym- 
posium, München