reference surfaces using the Leica Axyz ([3]) software (Figure
9).
(1)
Q)
Figure 9: Point offsets from CMM derived surfaces 1.2
4.4.1 Laser dot data
In comparing the laser data to the CMM surface 1, the average
offset from the reference surface is of the order of twice the
estimated target coordinate precision (Table 3). The data also
demonstrates a small negative bias indicating more points are
below the reference surface than above it.
Positive
(mm) Negative
Table 3: Surface 1 to laser dot point cloud comparison results
(42 points)
4.4.2 Pattern projection data
The pattern projection data set provided a more dense surface
than the laser points for surface 1 (Table 4, 5), however the
distribution of surface coverage is limited to a particular region.
In this case though, the data are evenly distributed either side of
the surface.
(mm) Positive
Negative
Table 4: Surface 1 to pattern projection data comparison results
(132 points)
Measurement of surface 2 was hampered by several low
intensity patterns, which in addition to shadow and occlusion
effects cased by the projector and camera geometries, have
resulted in only raised areas of surface being successfully
measured. The data exhibit significant negative bias and a
particularly large discrepancy between precision and accuracy
estimates.
(mm) Positive
Negative
Table 5: Surface 2 to pattern projection data comparison results
(97 points)
4.4.3 White light data
The data set with the projected white light produced point
clouds that covered a greater portion of each surface (Tables 6,
7). This data set exhibits larger maximum discrepancies with
respect to the reference surfaces, which are probably
attributable to regions of lower image texture.
(mm) Positive
Negative
Table 6: Surface 1 to light projection data comparison results
(104 points)
Positive
(mm) Negative
Table 7: Surface 2 to light projection data comparison results
(106 points)
4.44 Accuracy summary
The accuracy assessment for these data sets is relying on the
fidelity of the reference surface, which has been interpolated
from the CMM point data by means of B-spline surfaces. Such
a method is appropriate for smooth surfaces but can have
unpredictable and non-quantifiable effects for complex
surfaces. Also to be considered is the fact that the gearbox
surface in the areas tested has a surface texture resulting in an
increase in CMM probe measurement uncertainty.
All three data sets demonstrate agreement with the reference
surface to the order of 0.3 to 0.4 mm. There is a noticeable
negative bias in all cases, the cause of which cannot be
ascertained from the information available but may be linked to
limitations in the reference surface definition.
5. CONCLUSIONS
This paper presents some results towards the efficiency,
accuracy and reliability of making dense automated
measurements of complex engineering surfaces using multi
station photogrammetry. The developed method has been
shown to provide very dense point clouds of known quality.
The densification process is applicable to both white light and
artificial texture projection techniques. In both cases a dense
point cloud has been automatically provided with very few
outliers following filtering in the bundle adjustment.
19