Applications to Industrial Quality Inspection - 3D Computer Vision: Efficient Methods and Applications - page 268

Fig. 6.26 Comparison of a cross section through the three-dimensional surface profile obtained by
specular stereo, relying on intensity and polarisation angle data, and the ground truth determined
with a laser profilometer (connection rod example)
Fig. 6.27 Comparison of height differences on the surface obtained by specular stereo and the
corresponding ground truth values determined based on tactile measurement (star pattern example)
For the star pattern example it was not possible to obtain ground truth values
with the laser profilometer, as the depth differences across the profile are too large
and the surface slopes are too steep. Hence, we determined height differences at
two representative locations on the surface by tactile measurement (cf. Fig. 6.27 ).
At both locations, the correspondence between the specular stereo reconstruction
and the ground truth is better than 0 . 1 mm. For the ring-shaped flange example, a
similar technique was used to examine the surface reconstruction accuracy.
For the shallow depression marked in the lower right point cloud in Fig. 6.28 ,
a depth of 0 . 23 mm was obtained by specular stereo, which is in good agreement
with the ground truth value of 0 . 25 mm obtained by tactile measurement. The same
depth value was inferred from the dense surface profile shown in Fig. 6.24 ob-
tained based on the stereo point cloud and the available photometric and polarimetric
data.
6.3.4.3 Self-consistency Measures for Three-Dimensional Reconstruction
Accuracy
The specular stereo algorithm of Wöhler and d'Angelo ( 2009 ) yields several self-
consistency measures which provide an impression of how accurately the surface