The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
1042
j)
n)
a 0 [10' 3 mm]
a, [IO 3 ]
b, [10‘ 3 rad ]
b 5 [10' 3 mm]
Cj [10' 3 mm]
C; [ 10" 3 rad ]
4.00 ±2.23 (90%)
1.60 ±0.64 (98%)
2.91 ± 1.79 (80%)
1.50 ±0.51 (99.0%)
2.86 ± 1.02 (99.0%)
0.85 ±0.09 (99.9%)
5.09 ±2.18 (98%)
1.37 ±0.63 (95 %)
1.94 ± 1.54 (-)
1.58 ±0.46 (99.9%)
3.97 ±0.88 (99.9%)
0.97 ±0.08 (99.9%)
8.007 ± 0.002
-153.7 ±1.3 (99.9%)
-75.2 ±1.8 (99.9%)
1)
c [mm ]
x 0 ‘ [ 10' 3 mm]
y 0 ‘ [ 10' 3 mm]
8.011 ±0.006
-153.9 ±1.2 (99.9%)
-72.7 ± 3.7 (99.9 %)
Table 5. Estimated additional calibration parameters of laser
scanner and fisheye lens camera
Own independent investigations have shown that the laser
scanner calibration values vary due to different measurement
conditions (distance range, scan resolution, target design, etc.).
Therefore it is reasonable to implement the self-calibration
strategy into the laser scanner processing in order to obtain
values, which are particularly effective under the measurement
conditions at hand. Thus, the accuracy of the laser scanner data
can be improved in general.
4.6 Variance component estimation results
Table 6 shows the estimated a-priori standard deviations of the
observation groups as a result from the bundle adjustment with
variance component estimation (Schneider & Maas, 2007) of
example j), 1) and n). The observations are separated in distance
D, horizontal and vertical scan angle a, f and image coordinates
x y ’. These values provide information on the accuracy of the
observations, which depends on the accuracy and stability of
the used instrument, on the measurement conditions as well as
on the correctness of the geometric model used for the
calculation.
j)
1)
n)
D
8.75 mm
8.68 mm
a
15.0 mgon
-
14.9 mgon
ß
15.3 mgon
-
15.1 mgon
x ’’y’
-
0.228 pixel
0.176 pixel
Table 6. Estimated variance components of observations
It can be seen that the estimated standard deviations of each
observation group were slightly improved in the integrated
processing. The reason of this reduction is the higher reliability
of the results due to different types of observations which are
able to control each other within the bundle adjustment, i.e.
outliers can be detected easier. Therefore a few more observa
tions have been identified as outliers in calculation (n) in com
parison to (j) and (1). This fact causes an improvement of the
standard deviation of the observations as well as a slight impro
vement of the standard deviations of the unknown parameters.
5. CONCLUSIONS
Terrestrial laser scanner and fisheye lens camera complement
one another quite well in an integrated processing scheme.
Application-wise, a terrestrial laser scanner is mainly used for
3D modelling by an object representation based on stochastic
distributed points, while a camera image is used for coordinate
determination of discrete points as well as colorization of laser
scanner point clouds or texturization of 3D models.
The simultaneous bundle adjustment of laser scanner and
fisheye image observations as presented in this paper provides
numerous advantages. One advantage of this approach is, that
the camera can be orientated and calibrated on-site, which
promises an optimal registration between both data sets.
Furthermore, the camera can not only be used for providing
colour information, but it is also able to participate in the
determination of object geometries in terms of coordinates of
object points in a multi-station configuration. Depending on the
image resolution and camera stability, the camera even has the
potential to improve the accuracy of 3D object points in
comparison to the pure laser scanner measurement and to
support the self-calibration of the laser scanner and thus to
increase the accuracy of the laser scanner point cloud in general.
Due to different types of observations used in one calculation
process, the reliability of the parameter and coordinate
determination can be enhanced. The observations control each
other, resulting in improved outlier identification.
Strictly spoken, the results and the drawn conclusions presented
in this paper only apply to the actual recording and analysis
parameters (scan resolution, sub-pixel image measurement of
signalised points, etc.). Nevertheless, the potential of the
presented approach (in terms of instrument calibration, sensor
registration, enhancement of accuracy and reliability) has been
shown.
In practical applications it is recommended to choose the laser
scanner positions according to optimal visibility of the object
details without occlusions and to capture a few fisheye images
additionally, either from the same position as the laser scanner
(if the camera is mounted on the laser scanner) or from different
positions allowing for an optimal intersection geometry.
Finally it has to be noted that these conclusions also apply for
conventional central perspective images, but with the limitation
of a smaller field of view in comparison to fisheye images.
REFERENCES
Abraham, S., Förstner, W., 2005. Fish-eye-stereo calibration
and epipolar rectification. ISPRS Journal of Photogrammetry &
Remote Sensing, Vol. 59 (2005), 278-288.
Bakstein, H., Pajdla, T., 2002. Panoramic Mosaicing with a
180° Field of View Lens. In Proceedings of the IEEE Workshop
on Omnidirectional Workshop, pp. 60-67, IEEE press.
Beers, B.J., 1997: 3-D landsurveying using the FRANK method:
CycloMedia Mapper. In: Gruen/Kahmen: Optical 3-D
Measurement Techniques IV. Wichmann Verlag, pp. 283-290.
Böhler, W., Marbs, A., 2004: Vergleichende Untersuchung zur
Genauigkeit und Auflösung verschiedener Scanner. Luhmann,
Müller (Hrsg.): Photogrammetrie, Laserscanning, Optische 3D-
Messtechnik - Oldenburger 3D-Tage 2004, Wichmann Verlag.
Brown, D., 1971. Close-Range Camera Calibration. Photogram-
metric Engineering, Vol. 37, No. 8.