International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
Sector Cover N° of samples
Si Forest cover 10
S2 Test plot B 15
S3 Test plot C 15
s4 Test plot A 15
S5 Test plot D 15
Table 2. Characteristics of the four test sectors.
The samples in the ROI where chosen with specific care of
leaving out any surface which does not have a vegetation cover,
to limit variability of the geometry and radiometry of the return
echo, since, as it is described in literature (Jutsi and Stilla, 2003)
there is significant variability even between different elements
inside an urban environment.
2.5 Waveform extraction and analysis
The process for extracting the waveform data was done by
implementing custom routines developed in c++ language which
have been integrated in a tool with graphical user interface. This
software is in very early development stages and is work in
progress; anyone interested is welcome to contact the author for
testing purposes. The objective of the software is to bundle
methods for processing waveforms directly and for exporting
waveforms to more common formats such as ASPRS' LAS 1.3
or future open standards.
The routine requires a shapefile with points representing the
center of each sample plot, the radius of the plots and the folder
path were all of the files are stored. It then proceeds in reading
the plot coordinates, and filling a container for each sample with
the pulses falling inside the sample area. Each pulse is then
linked to the corresponding waveform data using the GPS
timestamp to search the waveform file (see section 2.3).
5 Sector Areas in ROI
| 10/15 Samples / Sectors |
| ~5000 laser Pulses / Sample
| 1-8 waveform Segments / Pulse
| ^ 10-256 samples (1 ns) / Segment
Figure 3. Plot of eight waveforms of the outgoing pulse.
Figure 3 shows a scalar relationship from the ROI down to the
single 1 ns sample in the waveform segment.
The position in 3D space of each return sample at time # in the
waveform segment can be calculated with the following: the
sensor position, the sensor orientation, the pulse scan angle and
the time interval (t;) between the beginning of emitted pulse (TO)
and the beginning of the n™ return echo waveform segment
(Tn), using the following formula:
525
Xp =X, +R, sin0 cosy —R,, singcosy
Van ^ Y, + Ra) Sin 0° siny +R, sin cosy
Zu Zu 7 Ra) CoOSO cos ó
where, at time t, $, 0, y are respectively the pitch, roll and yaw
angles of sensor, o is the scan angle of pulse, X,Y, Z are the
coordinates of the sensor center, and R, is the range of the first
waveform sample.
The range Ry, is calculated using the speed of the laser pulse and
the time interval between the maximum value of the TO pulse
(see figure 2) and the first sample over the baseline of the return
waveform segment Tn:
10°
where P is atmospheric pressure in mbar ant T is air temperature
in degrees Celsius.
Pressure and temperature were considered for the sake of
comprehensiveness - a difference of 30° C and 200 mbar
pressure brings a difference of about 3 mm which is not
significant considering the other sources of error in a airborne
laser scanner survey.
The method reported above was used to fill a voxel grid with 0.3
m resolution with information on the waveform data. The range
was calculated for each waveform data which was significantly
above the baseline of the return signal; an empirical value of 5
digitizer counts was used as threshold. This means that a surface
which causes a reflection will imprint a high value in the voxel it
falls.
The voxel grid is then used to extract statistical information on
vegetation height and density over the whole sample, and
successively over the whole sectors. The Kolmogorov-Smirnov
two-sample test (K-S) is a non-parametric test that is sensitive to
both the shape and location of peaks within the distribution and
was used to compare between samples in the same sector and
between samples over different sectors to check if inter- and
intra- variances are significantly different thus leading to the
possibility of using the method to discriminate areas with
different vegetation structure.
2.99792458-10°
ee P .
2731547
Ro = T
2E T 78.7
3. RESULTS AND DISCUSSION
This method sample the waveform information in 3D space, and
will be the basis of further work for vegetation analysis. It is
different from more common methods which use fitting of
waveforms to discriminate between returns and use peak values
to have a dense point cloud with width and amplitude
information. In this case we base our method on statistics on the
voxel grid, which overrides any signal analysis step necessary
for determining the point in space which causes the reflection. In
this case the retum waveform from a reflecting surface will not
supply information to a single point in space or to a single voxel,
but will be supplying information to many voxels along its path.
The resolution of 0.3 m was chosen empirically as it is a space
which can include —2 successive samples. À bigger voxel size
would be less discriminate, whereas a smaller voxel size would
not improve results, but increase threefold the size of the files.
As can be seen in figure 4, results can visually report vegetation
structure. Table 3 reports the results of the K-S test over the
distribution of metrics extracted from the voxel grid, showing
Q)
