ul 2004
——————
ias also
y Stage
fects of
)
1e
| to
e fests
strictions
jour chart
ial to the
side and
1s as the
| at these
ito small
rt patch.
| area of
selected
example,
t data for
with the
nt in an
to a least
dently fit
1. In each
exported
the laser
patch the
to create
| indicate
t each of
ial to the
rt rotated
allowed à
les.
| confirm
the range
chart and
of f-tests
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
were also used to determine whether the spread of points
defining a colour patch were significantly different to those
defining the same colour patch at the different ranges and
angles. Finally, any relationship between the different colours
and the scanner's data quality value from the return signal was
examined. This was achieved by plotting the range error of each
colour patch against the reflectance value measured by the
scanner, as well as the spectral reflectance value of each colour
patch at the scanner's wavelength (532nm). If any systematic
effects could be identified this would offer the possibility of a
correction model capable of further improving measurement
accuracy to similar surfaces.
RESULTS OF THE COLOUR CHART ANALYSIS
Least squares plane fitting
Range parameters derived from least squares plane fits to the
point cloud data demonstrate that different colour patches
exhibited different range offsets. For the greyscale chart patches
the range to the black patch, as determined by the scanner, was
greater than the range to the white patch and the grey patches in
between systematically stepped back from one another. The
standard deviation of each individual colour patch plane fit was
of the order of two to three millimetres. Larger planes were also
fitted to the entire colour chart with all of the borders and edges
omitted in order to obtain a mean range for the complete chart.
Figure 3 shows the plancs fitted to the point cloud at positions
one to three. The plane offsets are consistent at each position,
for example, the planes fitted to the white patches all lie in front
of the large patch fitted to the entire data set.
Figure 3: The large and small planes fitted to the point clouds
of the colour chart at positions one (nearest), two and three
To quantify the observed offset between planes, the distance
between each fitted plane and that of the Neutral 8 grey patch
were calculated. The range offset was recorded from the centre
of the plane perpendicular to the Neutral 8 grcy plane. Figure 4
shows the offset for selected colour patches and the Neutral 8
grey patch across the test range. For data relating to the
remainder of colour patches refer to paper in summer 2004
edition of Survey Review. A t-test (2 tailed 9594 confidence)
Was used to assess the hypothesis that the position of each of the
small planes corresponding to each individual chart colour patch
was significantly different to the position of the large plane
fitted to the entire set of colour patches. The test demonstrated
that the measured range was si gnificantly different.
ligure 4 confirms that the black patch is consistently located a
distance behind the other greyscale patches. The maximum
observed difference in range (12mm) is between the black and
White patches. The range offset between the position of the
black and Neutral 8 grey patch of 11mm is also significant
When compared with quoted scanner performance figures.
i
j 9 ji 4 —$— White
— 4 €
2 731 07 — Black
= S —> Blue
9 *
= Green
©
—t— Red
Range from scanner to neutral 8 grey (m)
Fig.4. Observed offset between selected colour patches and the
neutral 8 grey patch over the test range.
Note the neutral 8 grey patch lies along the x-axis.
These differences in range are greater than the 2 to 3mm
standard deviation of the plane fits and are therefore likely to be
attributable to a difference in reflectivity of each colour patch at
the wavelength of the laser. It is assumed that electronic
processing within the scanner interprets a change in the
intensity of the return signal, processed according to some
internal threshold, as a difference in range. The observed
differences are beyond the single point linear and range
accuracies quoted by the manufacturer (44mm and +6mm
respectively) and may have significant implications for the
precision of scanning projects. In general the magnitude of these
observed differences between the colour patches were
consistent at each position. The most notable exception is the
blue patch where the offset to the Neutral 8 grey patch increases
with range, from five millimetres at position one to twelve
millimetres at position fivc.
Distribution of measurement error per patch
To visualise the distribution of points defining the colour
patches the data surface was triangulated to produce surface
plots against a datum defined by each individual plane fit
(Figure 5). This procedure was carried out for the white, Neutral
8 grey, black, green, blue and red colour patches (only two are
shown). The figures provide visual confirmation that, for all
patches investigated, the distribution of noise appears random
and is therefore unlikely to be biased by a systematic error.
(a) (b)
Fig. 5: (a) Neutral 8 grey and (b) Green plots
Histograms from plane fit residuals — plane orthogonal to
scanner
Least squares plane fitting can be also used to investigate the
distribution of error in the scan data. For this procedure the
residual of each point, expressed normal to the fitted surface,
and the RMS of all the residuals were evaluated. The point
residuals were used to create a histogram for each colour, so
that any local variations in the plane fit could be visualised
graphically. The shape of the histogram and the associated
1033
