DECOMPOSITION OF AIRBORNE LASER SCANNING WAVEFORM DATA BASED ON
EM ALGORITHM
Qi Li*
School of Remote Sensing and Information Engineering, Wuhan University. 129 Luoyu Road, Wuhan 430079
KEY WORDS: LIDAR, Waveform-digitizing, Analysis of waveform, Filter, Smooth, Gaussian decomposition, DSM, DTM, Forest
structure
ABSTRACT:
Data storage capacity and high processing speed available today has made it possible to digitally sample and store the entire reflected
waveform of Small-Footprint Airborne Lidar (light detection and ranging), instead of only extracting the discrete coordinates which
form the so-called point clouds. One of the most important advantages from waveforms data is that it gives the user the chance to
extract three-dimensional coordinates by himself in the post-processing. Decomposition return waveform is a key step during
analyzing waveform data. Conventional algorithm to decompose is maximum and centre of gravity, or simply by using the
thresholding method provided by equipment vendor. Both show lack of high accuracy. In this paper, an improved Expectation
Maximum (EM) algorithm is adopted to extract peak location and pulse width from raw waveform data, proving it is a reliable and
high accurate decomposition algorithm. Moreover, the high-quality point-cloud data could be obtained which provides high-quality
resources for DSM(Digital Surface Model) and DTM (Digital Terrain Model) production. Derived forest biophysical parameters,
such as vegetation height and crown volume are able to describe the horizontal and vertical forest canopy structure.
1. INTRODUCTION
In the last ten years, airborne laser scanning is a rapidly growing
technology which has initially been conceived for topographic
mapping. Airborne laser scanners employ, with few exceptions,
pulsed lasers that repetitively emit short infrared pulses towards
the Earth’s surface. Some of the energy is scattered back to the
sensor where it is measured with an optical receiver. A timer
measures the traveling time of the pulse from the laser scanner
to the Earth’s surface and back, the integration of laser with
GPS and INS for position and orientation determination.
Nowadays, ALS is used routinely for topographic mapping and
is considered to hold a large potential in a range of other
applications such as forestry, 3D city modeling or power line
detection.
Development of airborne laser scanning goes back to the 1970s
and 1980s. The first commercially available airborne laser
scanners recorded the time of one backscattered pulse. The
recording of only one pulse is sufficient if there is only one
target within the laser footprint. State-of-the-art commercial
laser scanners typically measure first and last pulse; some are
able to measure up to five pulses. In order to derive digital
terrain models (DTMs), laser pulses reflected by the ground
surface must be distinguished from non-terrain points. This task
can be achieved using various filtering techniques that classify
the point cloud into terrain and off-terrain points just based on
the spatial relationship of the 3D data. For many applications,
this has been deemed the suitable form of output. However, the
user has no way of knowing how the electronics of his LIDAR
system actually determine the location of the returns they report,
nor of any distortions of the pulse shape that receiver electronics
or surface structures may have imposed upon the pulse echo.
LIDAR system manufacturers are tight-lipped about the pulse
detection methods their systems employ. However, as Wagner
et al. (2004) point out the choice of pulse detection methods has
significant impact on accuracy, and in practice causes a number
of effects that reduce the quality of the measurements, like
amplitude dependant range walk, slope dependency of range,
signal ringing causing outlier measurements below the terrain
level, etc. In addition, with mere range output much of the
informational content about structured surfaces is lost.
The solution is to digitally sample and store the entire echo
waveform of reflected laser pulses. Digitizing and recording the
complete backscattered waveform during the acquisition for
later post-processing has the advantages that algorithms can be
adjusted to tasks, intermediate results are respected, and
neighborhood relations of pulses can be considered. The
technical feasibility has been demonstrated by large-footprint
airborne systems developed by NASA in the 1990s, namely the
Scanning Lidar Imager of Canopies by Echo Recover (SLICER)
and the Laser Vegetation Imaging Sensor (LV1S). Recently,
three commercial airborne systems have become available, such
as Figure 1,namely the RIEGL LMS-Q560> Toposys FalconllL
Leica ALS-II % Optech ALTM 3100E.
Discrete
Returns
.1 si return
} 2nd return
. last return
time
Echo
waveform
Amplitude
outgoing pulse
return signal
1st level
(canopy)
anopy
structures
I level (bushes)
ground
time
Fig. 1 waveform Digitization
An approach based on unsupervised learning is presented where
a mixture of Gaussian distributions are fitted to the waveforms
* Email:qli_rs@yahoo.com.cn