ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002
algorithm requires no interpolation between the laser points.
This can be achieved by fitting an analytical model of the ditch
height profile to the laser points in both strips (figure 3). In this
way the height gradients that are required for the fitting can be
taken from the analytical model instead of from the laser data.
Figure3: Point cloud of the same ditch as in figure 2 viewed
longitudinally and fitted to an analytical profile.
Because of the relatively low point density, the detection of
such linear features can not be done by a standard edge
detection in a height image. As can be understood from figure 2,
this would result in very fragmented edges. Clustering-like
techniques seem to be more suitable for this task.
Combining the information of several linear structures and
planar faces results into the same information on the offsets
between strips as would otherwise by gathered by the estimation
of corresponding locations. Still, it may be questionable
whether the height data alone can always provide sufficient
information for a three-dimensional strip adjustment.
3. MATCHING REFLECTANCE DATA
Most laser sensors nowadays have the possibility of recording
the intensity of the reflected laser pulse. Several authors
suggested methods to make use of this data for the estimation of
planimetric offsets between strips [Burman 2000, Maas 2001].
Indeed, reflectance data often contains much more detail that
can be used to determine offsets (figure 4). The usage of
reflectance data, however, also has some inherent problems. In
the next paragraph several aspects of the noise characteristics of
reflectance images are discussed. We then present a more
detailed look on how edges are represented in reflectance
images and how this affects the edge location. Finally a
procedure is suggested to partially overcome the noticed
problems.
Figure 4: Height and reflectance data of a road crossing.
3.1 Noise characteristics of reflectance images
Reflectance images are known to be relatively noisy. Several
reasons can be identified for this property:
e The way most laser scanners measure the intensity of the
reflected pulse is by quantising the intensity at some point
of time, instead of integrating the intensity over a small
period around that point.
e Compared to the distance between the laser points the
amount of detail may be very high. Images of urban scenes
generally look noisier than images of rural areas (figure 5).
e If a laser beam hits multiple objects at different heights,
only the energy that is reflected by one of these objects is
used for the determination of the intensity of that pulse.
e Finally, the footprint of a laser scanner is usually much
smaller than the distance between two laser points. Hence,
the intensities only represent the reflectance properties of a
small part of the terrain. The difference between the
footprint size and the distance between two laser points can
be quite large. E.g. a typical scanner has a footprint size of
0.3 m at a flight height of 1000 m. Scanning with an
opening angle of + 20° and a pulse rate of 25 kHz, the
average point distance recorded with this scanner at a flight
height of 1000 m and an aircraft speed of 60 m/s would be
1.3 m. In such a configuration, the footprints only cover
about 4% of the surveyed area. This amplifies the noisy
appearance of reflectance imagery.
Figure 5:
Reflectance images of an urban and an agricultural
scene.
3.2 Edges in reflectance images
In the ideal imaging case the grey value of a pixel represents the
average grey value of the area that is covered by that pixel.
When generating a reflectance image the grey value of a pixel is
based on the reflectance properties of only a small fraction of
the pixel area.
This characteristic has an impact on the location accuracy of
edges in reflectance imagery. In the extreme case of an infinitely
small footprint the measured reflectance intensity will be
representative for the surface properties at only one side of the
edge. As in the case of height jump edges in height imagery, an
edge in such a reflectance image should be considered as an
edge in a binary image with the edge location properties as
described in [Fórstner 1986]. Both the bias and the standard
deviation of edge location depend on the orientation and length
of the edge.
A - 3T]