The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
1040 
For the manual measurement of the target coordinates, the Riegl 
laser scanner software RiScan Pro was used. As the black 
circles on white background are well distinguishable within the 
intensity images of the laser scanner, these were used for the 
coordinate determination by manual selection of the target 
centre (with integer pixels) and attribution of the associated 
spherical coordinates. The target code can not be used because 
the quality of the laser scanner intensity images does not allow 
a code interpretation. Also an automatic target measurement 
with sub-pixel operators is not supported by the software. 
Therefore the lateral accuracy of the spherical laser scanner 
coordinates is strongly limited by the chosen angular resolution. 
The target design (black on white background) turned out to be 
not really suitable, as the intensity values in the target centre 
were very low which often resulted in suboptimal accuracies of 
the distance determination. In some cases the laser scanner was 
not able to detect a reflected signal from the centre of the target. 
It has to be mentioned here, that white targets on black 
background would be better suitable for the laser scanner 
calibration. However, it has to be considered in principle, that 
the material and colour of the targets influences the distance 
measurement in a way, that distances will be measured 
systematically too short or too long. This fact has to be kept in 
mind for the interpretation of the calibration results, in 
particular regarding the parameters distance offset and scale. 
combination of the two scans as well as the fisheye image 
provides almost the same field of view. 
a) •* b) 
Figure 5. Configurations: scans from the room centre 
For object points on the ceiling, X and Y coordinates can be 
considered as lateral coordinates, Z as the depth direction. 
Table 1 shows that the accuracy of the resulting object point 
coordinates (RMS of estimated standard deviations) can be 
improved using at least one fisheye image additionally, 
particularly in X and Y direction. This result was expected 
since the fisheye image observations were measured with sub 
pixel accuracy operators (while the laser scanner observations 
result from integer pixel measurement within the intensity 
image). The standard deviation of object points resulting from 
the bundle adjustment of calculation b) is better than the used 
angular scan resolution and also better than the distance 
measurement accuracy. 
4.3 Fisheye image data 
In the calibration room mentioned above, five images with a 14- 
Megapixel Kodak DCS 14n Pro camera equipped with an 8 mm 
Nikkor fisheye lens were captured from different positions. 
With a pixel size of 8 pm the object resolution is 4 mm in 4 m 
distance. 
Scans/ 
Images 
Observ./ 
Unknown 
Points 
X 
RMS (mm) 
Y Z 
XYZ 
a) 
2/0 
339/217 
66 
5.00 
3.25 
7.58 
9.64 
b) 
2/1 
457/233 
66 
2.69 
2.00 
4.47 
5.59 
Table 1. Bundle adjustment results (configuration a, b) 
Fisheye images are often characterized by strong effects of 
chromatic aberration ([Luhmann et. al., 2006], [van den Heuvel 
et. al., 2006]). Although the consideration of the chromatic 
aberration in the geometric model is basically possible (e.g. 
[Schwalbe, 2005]), this was not done in the investigations 
described in this paper. Instead, the images were divided into 
their three color channels, and only one channel (green) was 
used for the following analyses. 
For the sub-pixel image point measurement within the 
hemispherical fisheye images and the decoding of point number 
the software Aicon 3D Studio was used. 
4.4 Configurations and results 
In the following, the results from different laser scanner and 
fisheye camera configurations, which have been processed in 
the integrated bundle block adjustment, will be analyzed. In 
order to achieve an optimal utilization of the different 
observation types, a variance component estimation scheme has 
been applied. Each calculation example is processed as free 
network adjustment and with an integrated self-calibration of 
the involved devices fisheye lens camera and terrestrial laser 
scanner. 
4.4.2 Two opposed scans 
The purpose of the following calculations is to analyse, whether 
scans, which enclose optimal intersection angles with the object 
points, result in an improvement of the estimated accuracies. 
Therefore two diametrically opposed scans have been chosen 
(Figure 6, calculation c and d). Calculation e) and f) 
additionally utilize one fisheye image on each laser scanner 
position. In calculation g) and h), two scans and two fisheye 
images are distributed in the room comers to allow for optimal 
intersection geometry. 
4.4.1 Scans from the room centre 
At first, two laser scans from the room centre (Figure 5, 
calculation a) and additionally one fisheye image (Figure 5, 
calculation b) were used to calculate the 3D coordinates of 
object point targets visible in each used scan or image. The 
Figure 6. Configurations: diametrically opposed scans 
Table 2 summarizes the results of the calculations illustrated in 
Figure 6. It has to be noted, that c) and d) are based on a lower 
number of corresponding object points visible in both scans due 
to object occlusions. However, the RMS values of these