The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B5. Beijing 2008 
947 
mean differences from the background image. This test has 
been carried out for object distances ranging from 60cm to 
260cm and integration times from 4><2.2ms to 4x20.2ms. 
However, none of them resulted in statistically significant 
deviations from zero, why the effect of scattering is assumed to 
be negligible in the following. By the way, the same has been 
found to hold for the amplitude observations. 
Frame-Wise Arithmetic Mean of Amplitudes 
1000 1500 2000 400 450 500 550 600 
Frame-Wise Arithmetic Mean of Distances 
0 500 1000 1500 2000 
Frame Number 
Figure 1. Frame-wise arithmetic mean of amplitudes (top) and 
distances (bottom). Enlarged details pointing out the periodicity 
of the signals (right). 
Figure 2. Arithmetic mean of the background subtracted range 
observations, depending on the distance between the 
observing pixels and the instantaneous position of the 
imaged target (red, solid), surrounded by its standard 
deviation (red, dashed), and corresponding observation 
counts (black). The results are shown for object 
distances and integration times of 0.6m/4x2.2ms (left), 
and 1.35m/4x20.2ms (right). 
3. ROBUST TARGET TRACKING 
For the purpose of gathering large amounts of reference data, 
the coordinates of the target centres pictured in the amplitude 
images must be determined automatically. As the calibration 
data shall cover both a wide range of integration times and 
object distances, low-amplitude data is inevitable. In addition to 
the resulting high noise levels, the employed calculus must cope 
with repetitive image structures (cf. section 4.1). Because of 
both the vignetting of the objective and the (roughly) radially 
decreasing power of illumination, the amplitude generally 
decreases towards the image borders. As no image 
preprocessing shall be applied, the algorithm must manage 
hugely imaged targets featuring inhomogeneous amplitude 
values. 
Among other, more common target centroiding methods, the 
stepwise adjustment of the target model described in (Otepka, 
2004) demonstrates best performance concerning the 
shortcomings of the amplitude data in use. Furthermore, this 
procedure yields additional results that can be used for 
plausibility checks and weighting in subsequent adjustments. 
The tracking of targets as implemented for this work is based on 
the detection of potential target areas by morphological image 
processing, adjustments of the aforementioned target model, the 
matching of the resulting target centres with their projection 
from the test field (based on the recent exterior orientation, 
nearest neighbours and considering the planar shift from the 
recent image found by correlation), and the subsequent 
adjustment of the camera’s current exterior orientation using 
LI- and L2-norms. Furthermore, some plausibility checks are 
performed. 
4. CALIBRATION 
Sensors of digital amateur cameras are reported to be rather 
loosely coupled to their casing, resulting in an unstable interior 
orientation possibly sensitive to agitation or somehow following 
gravity. In order to avoid displacements of the sensor with 
respect to the projection centre during data capture, it is thought 
to be advantageous to keep the optical axis directing either up- 
or downwards and to avoid rotations about it. Lacking a convex 
3-D surface, the calibration of the ranging system needs to be 
realized on a 2-D test field in order to avoid multipath effects in 
object space. The camera calibration described in this work 
bases on known control point positions, but treats both the 
camera’s interior and exterior orientations as unknowns. Using 
a planar field of control points, unknown interior and exterior 
orientations are however prone to be considerably correlated to 
each other, especially if images with different rotations about 
the optical axis are absent. For the purpose of a more thorough 
investigation of the interior orientation, thus a calibration on a 
3-D test field is conducted, which exclusively uses the 
amplitude images (cf. subsection 4.1). The calibration 
integrating the range measurements however uses a 2-D field 
(cf. subsection 4.2). 
All bundle block adjustments are performed using the 
photogrammetric program system ORIENT/ORPHEUS (Kager 
et al., 2002). The parameters presented in the following thus 
refer to the respective definitions. 
4.1 Separate Calibration of the Interior Orientation 
The calibration of solely the interior orientation makes use of 
the amplitude data only, and is carried out using a 3-D test field. 
In order to maximize the number of discernible targets and 
hence permit precise spatial resections, flat, circular target 
markers are arranged in regular grids on three pairwise 
orthogonal planes. During data capture, the camera is guided in 
a way that the target areas fill the whole image i.e. the distance 
to the point of intersection of the supporting planes stays 
approximately the same. 
In order to test the short-term reproducibility of the interior 
orientation and the distortion parameters, 3 image sequences 
comprising 1000 frames each are captured within an hour. For 
the purpose of testing the influence of gravity, the optical axis 
constantly points upwards throughout the first two sequences, 
while it points downwards in the third one, avoiding rotations