The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
calculations are clearly below those of calculation a) and b).
This fact confirms that convergent laser scanner positions
processed in a bundle adjustment help to increase the accuracy
of object points. Moreover, the consideration of one fisheye
image on each laser scanner position (calculation e and f)
results in higher accuracies. The result of calculation g) and h)
is slightly better, due to an even better overall intersection
geometry.
Scans/
Images
Observ./
Unknown
Points
X
RMS (mm)
Y Z
XYZ
c)
2/0
318/178
53
1.20
1.16
1.43
2.20
d)
2/0
306/ 172
51
0.97
1.00
1.22
1.85
e)
2/2
605 / 239
66
0.66
0.55
0.88
1.23
0
2/2
585 /239
66
0.61
0.60
0.67
1.09
g)
2/2
580/239
66
0.62
0.57
0.79
1.15
h)
2/2
582/239
66
0.54
0.49
0.68
1.00
Table 2. Bundle adjustment results (configuration c - h)
The results allow for the statement, that it is reasonable to use
additional images in a combined bundle adjustment, in order to
achieve a higher accuracy in terms of registration and
instrument calibration. This applies particularly to fisheye
images since they often cover the same field of view as the laser
scanner. If the camera is actually mounted on the laser scanner,
it is simple to use their images in a combined bundle adjustment,
since an approximate orientation with respect to the laser
scanner is already known.
4.4.3 Multiple scans from the room corners
In order to demonstrate the potential of the combined
processing, observations of further scans and fisheye images
were additionally introduced into the bundle adjustment (Figure
7). For this purpose, at first i) 4 and j) 6 laser scans were
adjusted separately as well as k) 4 and 1) 5 fisheye images as
comparison. Calculations m) and n) combine the laser scans and
fisheye images in an integrated adjustment.
O 0
< >
r V V
k) -
£
m) ^
% 0
A >
£
g
e •
K»
m
j) ' f *
‘ x* y
1) -
„) : * * v
Figure 7. Configurations: multiple scans from the room
comers
Applying configurations i) and j), which only use laser scanner
observations, a slightly better accuracy (RMS of estimated
standard deviations of object point coordinates) was achieved in
comparison to calculation examples k) and 1), which only use
fisheye images for the 3D object point determination (table 3).
But from table 3 it becomes obvious, that the combination of
both leads to a significant improvement of the accuracy of
object point coordinates. Calculations m) and n) show the
potential of a combined processing of laser scanner and fisheye
image observations.
Scans/
Images
Observ./
Unknown
Points
X
RMS (mm)
Y Z
XYZ
i)
4/0
696 / 223
64
0.53
0.54
0.63
0.98
j)
6/0
1014/245
66
0.47
0.51
0.64
0.96
k)
0/4
472 / 232
66
0.75
0.72
0.59
1.20
1)
0/5
568/238
66
0.64
0.57
0.63
1.06
m)
4/4
1157/263
66
0.29
0.30
0.40
0.58
n)
6/5
1534/285
66
0.28
0.27
0.36
0.53
Table 3. Bundle adjustment results (configuration i - n)
4.5 Calibration results
An advantage of the processing of observations in a bundle
adjustment is the possibility of self-calibration. This means that
the used measurement device (laser scanner and/or camera) can
be calibrated simultaneously, since the calibration parameters
can be handled as unknowns in the same procedure. This was
applied successfully in the calculation examples presented
above.
Table 5 shows those laser scanner and camera calibration
parameters, which could be determined significantly from the
calculation examples j), 1) and n). While j) and 1) consider laser
scanner and fisheye lens camera separately, calculation n)
integrates both in one calculation. In addition to the parameter
values their estimated standard deviation and significance level
(in brackets) is presented.
Concerning the laser scanner only a few parameters could be
determined significantly: The sine coefficient of vertical circle
eccentricity (c/) is most significant in calculation j). The
horizontal and vertical collimation axis eccentricity (b 5 , c 3 )
could be determined on a 99% significance level; their
significance was increased in calculation n). While the trunnion
axis error (b 2 ) and the vertical circle index error (c 0 ) could not
be estimated significantly at all, the collimation axis error (6/)
was determined on a very low significance level. Additional
parameters for the compensation of distance errors (offset a 0
and scale ai) were included, too. Due to a high correlation
between the two parameters, their significance is low. It might
have been reasonable to omit one of both.
Also fisheye lens camera calibration parameters (interior
orientation and additional parameters) were estimated, although
not all of them are presented in table 5 (lens distortion
parameters are rather uncritical herein in terms of correlations).
The standard deviation of the principal distance c and the
principal point coordinate y 0 ’ could be improved in calculation
n). The same applies to most of the additional parameters
(distortion, affinity and shear).
It can be summarized, that the integrated processing of laser
scanner data and fisheye image data (calculation n) results in
calibration values, which have a higher accuracy and
significance in comparison to calculation j) with scanner data
and 1) image data separately. That means that scanner and
camera aid one another successfully in the self-calibration
process.
1041