The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
930
334 mm at the edge of the image format. Another large
correction of note is the 400 mm rangefinder offset. Little or no
improvement is gained by adding the two empirical parameters
in case 4 except in range, for which there is slight improvement.
RMS
% improvement
Case
X
(pm)
y
(pm)
P
(mm)
X
y
P
1
11
8
171
-
-
-
2
10
7
160
9
12
6
3
6
5
50
45
38
71
4
7
5
46
36
38
73
Table 2. RMS of self-calibration residuals and %
improvements
Focusing on case 3, the overall precision as measured by the
RMS of residuals is quite good in both x and y at 6 pm and 5
pm, respectively, which represent 0.15 and 0.125 of the pixel
size. This indicates that 1) the lens-system correction model
was effective and 2) the image point measurement method of
intersecting best-fit edge lines was very precise. At 50 mm, the
range results are somewhat less impressive, though. This is
most likely the due to the target design in which the observed
distance was interpolated at the boundary between the white
and black target components. Range biases were found between
these two differently-coloured materials, which cased severe
inflation of the range residuals. They can be seen in Figure 4 as
the recessed rectangles in the range image. The magnitude of
the bias can also be seen to vary with distance to the target.
Note, however, that the gradual curving of the planar surfaces is
expected in a range image.
Figure 4. Range image showing biases due to the different
target components as recessed rectangles.
This effect has also likely caused the estimated parameters, in
particular the rangefinder offset, to be biased and, due to the
high dispersion of observational errors, to be estimated with
low precision (±2 mm for do in case 3). The resulting high
dispersion of the range residuals for case 3 can be seen in
Figure 5 as a function of range. It is planned to use a circular
target design for future calibrations.
4.3 Rangefinder Error Examples
Figure 6 shows the range residuals from case 3 excluding the
cyclic error terms d 6 and d 7 to show the effect of this systematic
error. Clearly a large-magnitude, periodic error exists. The
nominal wavelength is 1.875 m, one-quarter of the unit
wavelength. Lindner and Kolb (2006) report a systematic
deviation from the sinusoidal pattern of the cyclic errors at
close range. No such effect was found in these data, though this
may be due to the slightly-larger minimum range mentioned
earlier.
Figure 7 shows the effect of excluding the e 2 clock-skew error
term from the case 3 self-calibration solution. The result is a
very strong linear trend in the range residuals as a function of
the y image co-ordinate.
p(m)
Figure 5. Range residuals vs. range for case 3.
250
onn
-200
-250 1 ‘ 1 1 '
1 2 3 4 5 6 7
p(m)
Figure 6. Range residuals vs. range for case 3 excluding the
cyclic error parameters d^, and d 7 .
Figure 7. Range residuals vs. y for case 3 excluding the clock-
skew parameter e 2 .