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
how far two overlapped echoes separated will influence the
detection results.
2. WAVELET-BASED ECHO DETECTOR
The wavelet transform is a tool to decompose a signal in terms
of elementary contributions over dilated and translated wavelets.
One of the continues wavelet transform(CWT) applications is
resolving overlapped peaks in a signal(Jiao et al., 2008). The
CWT at time uw and scale s can be represented as
wf ws) =(fw..)=| fOw, Od (D
where /{t) 1s the input signal, * denotes the complex conjugate,
y, .(r) 1s the wavelet function controlled by a scale factor s and
a translation factor uw. Wf{u, s) is the so called wavelet
coefficients. Applying CWT to the waveforms can be
considered as measuring the similarity between the waveform
and the wavelets. If the chosen mother wavelet and the
responded echo are similar in shape, then the locations where
WC peaks occur imply the positions of the response echo in the
waveform. Figure 2 shows an example of detecting echoes
using the wavelet-based detector. A signal(waveform) is applied
wavelet transform at two scales. Consequently we can obtain
the WC (Wf (u, s4), Wf (u, s2)) at each scale. Taking the result
at the smaller scale s; firstly, one can see that the locations
where the WC peaks (Wf(u4,s41), Wf (u2, 43), Wf (us, $4) )
occur corresponding to the positions of echoes in the waveform.
However for the case of larger scale s,, only two echoes are
detected. This can be explained that the expanded wavelet
cannot "see" the echoes whose size 1s smaller than the wavelet
itself.
signal
wavelet
(scale 1)
OON REOR M OW NOM
dbi m eomm mos ub oo
xot XE ue $e oim m de de
wow oul mom os xp mom
ww wd an wees
Hmmm
SiO
LB
21
HF s ; Vries, S MG s
Winn) NE rd
f IAN AAAS SAA
® E
E 9
= =
æ æ
æ =
signal 2 %
wavelet 2 NUN
(scale 2) 2 \ x
» m
i n
FETT Wits)
Hb, Sytem i "ers A Tr
Tus, Lr v
Figure 2. the interpretation of detecting echoes by CWT.
Since the wavelet can be scaled by a scale factor, the wavelet-
based detector is able to deal with different system with variant
echo width, for example, the echo width is 5 ns for Leica
ALS60 system and 8 ns for Optech ALTM 3100. Therefore to
530
optimize the detection results, an appropriate mother wavelet
and a scale factor need to be prior determined. Many researches
have considered the responded echo as a Gaussian function. For
this reason the Gaussian wavelet is chosen as the mother
wavelet in our study. Additionally by exploring some waveform
samples, the scale parameter can be determined according to
detection results.
3. EXPERIMENT AND RESULTS
3.1 Waveform simulation
A received waveform is a power function of time and can
be expressed as follows (Wagner et al., 2006; Mallet et al.,
2010):
P(0- Y 5*0 @)
where S(t) is the system waveform of the laser scanner, o;(t) is
the apparent cross-section, N is the number of echoes and k; is a
value varied by range between sensor and target. Eq. (2)
indicates that a return echo is the convolution of system
waveform and the apparent cross-section of a scatter. Wagner et
al.(2006) have reported that the system waveform of Riegl
LMS-Q560 can be well described by a Gaussian model. If the
apparent cross-section of a scatter is assumed to be of Gaussian
function, then the convolution of two Gaussian curves gives
again a Gaussian distribution. The received waveform P.(t)
can be rewritten as:
ny
N =
P (t) = > Pe 255i (3)
i=l
where t; is the round-trip time, s, ; the standard deviation of the
echo pulse, and P; the amplitude of cluster i. For this reason, the
Gaussian function is chosen to simulate the return echo. The
simulated waveform can be represented as follows:
m
w() - 3 g,() n 4)
2
G1)
25°
(5)
|
where w 1s the simulated waveform, m the number of return
echoes, n the noises which have a normal distribution(Unoise»
Onoise), 4 the location of time domain, and s the echo width
which can be represented by full width at half maximum
(FWHM =2+/2In2s). Figure 3 shows an example of a simulated
waveform.
g(t) 2 A-exp -