2004
with
t was
strate
n the
also
points
ease in
ibution
is most
at the
ges in
t. For
9 white
les the
ypeared
h patch
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
was scanned at increasingly greater angles. However, it should
be noted that although the histograms for the chart scanned at
60° exhibit the narrowest distribution, implying that a greater
majority of points lie close or on the plane, the density of points
captured is noticeably reduced. These results therefore suggest
that for matt surfaces it could be preferable to scan an object at
an angle of 20° to 40° to produce the optimal point cloud with a
reduced proportion of noise compared to an object scanned
normal to the scanner. These findings need to be validated
further with real objects with different surface properties to
ascertain if scanning an object at an angle significantly
increases the precision of the dataset for practical purposes such
as 3D modelling.
Correlation with laser scanner data quality value
The data was analysed to assess any relationship between colour
patch reflectance at the laser wavelength, the data quality
measurement recorded by the scanner and the range error. Any
such relationship could be important in generating a correction
method for the observed range discrepancies.
Figure 8 demonstrates the difference in range measurement for
each patch against the spectral reflectance value of each colour
patch. Range offsets are recorded with respect to the Neutral 8
grey patch. Colour patch spectral reflectance was derived
according to the known spectral reflectance of each chart patch
at the wavelength of the scanner's green laser [5]. A systematic,
but non-linear, relationship between the two variables is
apparent. Figure 9 shows a similar relationship between range
offset and average scanner measurement quality value for each
colour patch indicating that the quality measure is closely
related to patch reflectance. In both cases the actual quality
value recorded by the scanner was used rather than the
normalised value which is used to colour code the resultant
point cloud.
a
r
do
o
o
e
o o o
o
t3 Ln
Offset (mm)
M lI
0.2 0.4 0.6 0.8 o 1
4 t
=
22
GretagM acbeth Colour Chart Value at 532nm
Fig. 8: GretagMacbeth spectral reflectance value against patch
offset (Neutral 8 reference) at position five (7.75m)
© 12
© © 10
© Q8
: + s ©
s 20
S Le
-600 -400 -200 20 00
+4
Average Measurement Quality Value
Fig 9: Average scanner measurement quality value against patch
offset (Neutral 8 reference) at position five (7.75m)
The above figures show that there is a clear relationship
between the error in range and the scanner’s measurement
quality value. In general, as the offset from the Neutral 8 grey
patch increases, the spectral reflectance and the quality values
decrease. This information can be used to apply a correction to
the range error, which could potentially increase range
measurement accuracy.
Development of a simple correction factor
Figure 10 shows the quality measure recorded by the scanner
against the patch offset from Neutral 8 for chart positions one to
six. A cubic regression curve was fitted to each set of data. The
curves fitted to the data fall into three groups, those from
positions one to four (4.08m to 5.5m), those from positions five
and six (7.75m and 9.5m) and those from positions seven to
eight (14m to 24m). The reason behind this grouping is not fully
understood without knowledge of manufacturer's calibration
models. It is possible that the scanner applies discrete correction
factors for different groups of ranges, which could account for
the offset between the three sets of data. To test this hypothesis
further it would be necessary to carry out the tests over both a
greater range of distances and a denser sampling interval.
Offset (mm)
T T T
-400 -300 -200 -100 0 100
r | Scanner Measurement Quality Value
Position |
Position 2 |
Position 3 |
Position 4
Position 5
Position 6
|
|
Position 7
9+nDmn4<400
Postion &
=
eeression lines
Fig 10: Nonlinear regression analysis for the intensity at 532nm
and offset (from neutral 8 grey) for positions one to eight. Note
the three distinctive groups of curves.
In order to compute a correction factor that could be used for
close range work, the average regression cubic curve was
plotted for positions one to four corresponding to the first set of
curves.
As ‘a test, corrections derived from the cubic regression curve
were applied to the black patch point data at position one
orthogonal to the scanner. A plane was fitted to the corrected
points and analysed by creating a histogram of the residual
values for each point. The histogram was the same shape and
had very similar descriptive statistics to the original black patch
data. However, the overall patch position had moved closer to
the Neutral 8 grey patch. Therefore the corrected values for the
black patch data had increased the accuracy, but had not
increased the precision of the data.
Similar corrections were applied to all of the colour chart
patches including the pastel shades on the top two rows. Figure
I'l allows visualisation of the results. After the correction has
been applied the entire point set lie closer to the average plane
fitted through all points. For example, the plane fitted to the
black patch lies 2mm behind the average patch after the
correction has been applied. In comparison the uncorrected
black patch is positioned Smm behind the average plane. The
points comprising the corrected colour chart (Figure 11) can be
seen to be positioned nearer the plane than the uncorrected chart
(Figure 3). The greyscale patches in the uncorrected point cloud
