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
EXPLORING WEAK AND OVERLAPPED RETURNS OF A LIDAR WAVEFORM WITH
A WAVELET-BASED ECHO DETECTOR
C. K. Wang
Dept. of Geomatics, National Cheng Kung University, No. 1, University Road, Tainan, 701, Taiwan -
p6896102@mail.ncku.edu.tw
Commission VII, WG VII/7
KEY WORDS: LIDAR, Detection, Algorithm, Simulation, Accuracy
ABSTRACT:
Full waveform data recording the reflected laser signal from ground objects have been provided by some commercial airborne
LIDAR systems in the last few years. Waveform data enable users to explore more information and characteristics of the earth
surface than conventional LIDAR point cloud. An important application is to extract extra point clouds from waveform data in
addition to the point cloud generated by the online process of echo detection. Some difficult-to-detect points, which may be
important to topographic mapping, can be rediscovered from waveform data. The motivation of this study is to explore weak and
overlapped returns of a waveform. This paper presents a wavelet-based echo detection algorithm, which is compared with the zero-
crossing detection method for evaluation. Some simulated waveforms deteriorated with different noises are made to test the
limitations of the detector. The experimental results show that the wavelet-based detector outperformed the zero-crossing detector in
both difficult-to-detect cases. The detector is also applied to a real waveform dataset. In addition to the total number of echoes
provided by the instrument, the detector found 18% more of echoes. The proposed detector is significant in finding weak and
overlapped returns from waveforms.
1. INTRODUCTION
Studies have demonstrated that an airborne LiDAR system
(ALS) can provide accurate estimation of topographic surfaces.
An important application is for exploring some key
characteristics of a forest such as canopy height, vertical
distribution, above-ground biomass, leaf area indices and terrain
models (Pirotti, 2011). Most of papers, however, have shown
that LIDAR data underestimates the canopy height (Chauve et
al., 2009). The reason is explained that only a small proportion
of laser pulses interacting with the tree apices. Low point
density of tree tops results in underestimation of tree heights.
Another tricky problem is to classify the ground points in a
complex forest area. Usually when a laser emits into a forest
area, only some of the laser energy is scattered back by the tree
tops. The other energy penetrates through and is reflected by
branches, shrubs and the ground. The received echo scattered by
the ground could be too weak to being detected (Figure 1 (a)).
Additionally if scatters are separated around the range
resolution of the instrument, for example, the terrain and shrubs,
an ALS will produce a waveform composed of a superposition
of echoes. The overlapped echoes are difficult to resolve and
usually only an uncertainty point located between the terrain
and shrubs is detected (Mallet and Bretar, 2009) (Figure 1(b)).
The terrain height represented by this point will be a slight
higher than the true terrain. Fortunately, thanks to the
development of LiDAR technology, the laser scanner systems
with full waveform digitizing capabilities have become
available. Compared with the conventional LiDAR system,
waveform LiDAR further encodes the intensity of the reflected
energy along the laser lighting path. Users now can utilize the
waveforms for the interpretation of ground objects and develop
their methods to detect the effective return echoes. This gives us
a great opportunity to improve the canopy height estimation and
terrain model if an approach which can deal with the weak and
overlapped echoes is obtained.
529
UTIO]SA8M
ULIOJO9AE AA
(b)
Figure 1. Interpretation of difficult-to detect echoes, (a) weak
echo due to the loss of laser energy, (b) overlapped echoes due
to the scatters separated around the range resolution of systems.
The purpose of this study is to explore extra return echoes in
waveform data, especially weak and overlapped echoes which
are not detected in the online process of an ALS. However,
weak echoes are difficult to detect due to the low signal-to-noise
(SNR), and overlapped echoes are also hard to resolve because
they are synthesized signal. To deal with this problem, one has
to extract signals from a waveform that is composed of echo
signals and data noises. This paper proposes a wavelet-based
detector for the solution. The detector first applies a wavelet
transform (WT) to decompose a LiDAR waveform in terms of
elementary contributions over dilated and translated wavelets,
and then searches possible echoes in the wavelet coefficients
(WC) at a certain scale. Using the WT of a LiDAR waveform
enables us to detect the echoes at various scales and to suppress
the noises simultaneously(Shao et al, 1998). The detector
detects all possible return echoes from a waveform signal.
Focused on the detection of weak and overlapped echoes, our
analysis aims at how much signal strength relative to noise and