International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
4. DATA ANALYSIS
Algorithms are developed and evaluated with simulated signals
of synthetic objects.
First, a signal preprocessing of the intensity cube with a
matched filter is implemented to increase the precision of the
range measurement and improve the detection rate. These
results are used to analyze the waveform of each pulse for
gaining the pulse properties: range, pulse power and number of
peaks. Then the pulse properties are processed with a region
based segmentation algorithm. By the use of images the region
boundary pixels derived from multiple reflections at the same
spatial position are shared by separate regions. Considering the
received pulse power of the associated spatial neighborhood for
the region boundary delivers the estimation of the edge position
and edge orientation with sub pixel accuracy.
4.1 Signal preprocessing
The data analysis starts with the detection of pulses in the
temporal signal of the intensity cube. Usually this signal is
disturbed by various noise components: background radiation,
amplifier noise, photo detector noise etc. Detecting the received
signal of the pulse in noise and extracting the associated travel
time of the pulse is a well-known problem and is discussed in
detail in radar techniques (Mahafza, 2000; Osche, 2002;
Skolnik, 1980) and system theory (Papoulis, 1984; Turin, 1960;
Unbehauen, 1996). Due to this problem matched filters are
used.
To improve the range accuracy and the signal-to-noise ratio
(SNR) the matched filter for the signal of the backscattered
pulse has to be determined. In practice, it is difficult to
determine the optimal matched filter. In cases where no optimal
matched filter is available, sub-optimum filters may be used,
but at the cost of decreasing the SNR. With the assumption that
the temporal deformation of the. received signal can be
neglected and the waveform is uniformly attenuated (isotropic
attenuation by reflection or transmission of the pulse) the signal
of the emitted pulse is the best choice for the matched filter
determination.
Let us further assume that the noise components of the system
mentioned above are sufficiently described by white noise with
the constant factor Ny. Furthermore the signal energy of the
pulse is known as £, and the maximum SNR occurs if the signal
and the filter match. In this case the associated travel time / of
the delayed pulse is 79 and the SNR is described by
2E
SNR(tg) = — (1)
No
An interesting fact of this result is that the maximum of the
instantaneous SNR depends only on the signal energy of the
emitted pulse and the noise, and is independent of the
waveform.
The matched filter is computed by the cross-correlation Ry
between the signal of the emitted pulse x and the received
signal s. We obtain the output signal v with a local maximum at
the delay time r.
yn =R (1-1) (2)
Then the output signal with improved SNR is analyzed by a
detection filter tor local maxima to determine the travel time of
the pulse. By using the correlation signal for processing the
travel time a higher accuracy is reached than by operating on
the waveform. This is because the specific pulse properties (e.g.
asymmetric shape, spikes) are taken into account and so less
temporal jitter can be expected. For the detection the
preservation of the waveform has no relevance, only
maximizing the SNR is important.
Furthermore, the range resolution of the laser scanning system
generally depends on the temporal width of the emitted pulse.
Short pulses decrease the average signal power and the
associated SNR. To increase the temporal pulse width 7 without
decreasing range resolution we need to consider pulse
compression. The typical time-bandwidth product of a
conventional system is BT 1, but with an increasing bandwidth
B of the receiver and a corresponding waveform the range
resolution AR is improved and we obtain with the speed of
light c
AR=—. (3)
Then the improved range resolution of the laser scanner system
is inversely proportional to the bandwidth.
4.2 Pulse property extraction
Depending on the size of the focused surface geometry in
relation to the laser beam (footprint and wavelength) different
properties can be extracted (Jutzi & Stilla, 2003b). In this paper
we focus on the properties range, pulse power and number of
peaks.
e The range value is processed to determine the distance
from the system to the illuminated surface.
e The pulse power value is computed to get a description
for the reflectance strength of the illuminated area.
e The number of peaks is considered to locate the
boundaries of an object with multiple reflections.
For determining the property values of each pulse the intensity
cube is processed in different ways. First the range value is
found from the measured travel time of pulse by the matched
filter. Then the travel time is used to locate the area of interest
in the intensity cube for temporal signal analysis. We integrate
this area with the received signal s for the pulse power value P
0T.
P=— [toa . (4)
as
Integrating the intensity over a small area instead of quantizing
a single value has the advantage of decreasing the noise
influence (Vosselman, 2002).
Finally the number of peaks is determined by the number of
detected reflections for each emitted pulse. The number of
detected pulses depends on the threshold adjustment for the
local maximum detection of the matched filter output signal.
4.3 Segmentation
The segmentation of the range values provides a higher level
model formation. Segmentation considers the spatial
neighborhood of detected pulses to generate surface primitives
Internat
in the f
power i
multiple
pixels a
foregro
The cor
by the
same p
manner
determi
The re
multipk
single r
region 1
it can b
pixels c
Figure
An te
propert
differer
thresho
elemen!
segmen
44 R
For ph
(edges)
used.
represe
process
edge p:
and anz