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

Data acquisition was performed by using Leica's ALS60 sensor

with WDM6S, mounted onboard a Cessna aircraft. WDM65

records the complete waveform of the reflection from the

surface intercepting the laser footprint. The full waveform data

was collected as 256 samples (à) 1ns. Field of View (FOV) of

the scanner was set at 28 degrees, scan rate was set at 33.7 Hz

and laser pulse rate used was 54000 Hz. The data acquisition

was conducted by flying the aircraft at an altitude of 2491 m

from the mean sea level. The terrain height of the area varied

from 439 m to 790 m. The average point density of the

transmitted pulse was 1point/sq.m.

A computer program was developed to generate georeferenced

point data from the full waveform return pulses. The point

cloud generation algorithm implemented in our approach is

explained below with an illustration of a typical case.

(1) Discrete return pulse (black square dot) and the

corresponding waveform data for the laser pulse are depicted in

Figure 2 (top). The blue dots in the top image in Figure 2 are

full waveform data plotted against time. As seen in the figure,

the full waveform data are generally embedded with noise

generated from the variety of sources. Hence, a 12 order Finite-

duration Impulse Response (FIR) zero-phase digital filter

technique was applied to smooth the waveform and to reduce

the noise for the further analysis. For a 12 order FIR fiiter

(Figure 3), the output is a weighted sum of the current and a

finite number of previous values of the input as described in

equation 1.

y(n) 2 x(n)x h(0) - x( n -1)x h(1) ---- x(n-12)x h(12) (1)

The red curve in Figure 2 (top) is the smooth waveform data

obtained after the processing. The smooth waveform obtained

from this technique preserves peak position, pulse width and

skew of the laser reflection (Wong and Antoniou, 1994).

(2) After the waveform smoothening, the next step is to

distinguish the noise from the actual return intensity. For that

purpose, we selected a threshold and assumed all the reflection

below the given threshold as noise and excluded those values

for further processing. This is a very straight forward approach

and computationally very efficient. Two threshold values of

intensity 16 and 17 were tested. When the threshold was set at

16, in some cases the points below ground level under the tree

canopies were also generated. However with threshold 17, the

last return peak of the waveform corresponds very well with the

ground position level, under the tree canopies. Hence for the

surveyed region, all the waveform return intensity below 17

was assumed noise and considered their intensity to be 0.

(3) In the third step, the number of peaks and the location of the

peaks in the smooth waveform data were, at first, estimated by

finding the local maxima in the smooth waveform. Then an

algorithm based on the Expectation Maximization (EM)

(Persson et.al., 2005) was applied to correct the positions and

number of the peaks by Gaussian decomposition. The crosshair

symbols at the peak of the Gaussian bumps in Figure 2 (top) are

the corrected peak positions estimated by the EM algorithm.

(4) Finally, the correctly estimated peak positions were

georeferenced to create a point cloud data.

506

The bottom image in Figure 2 shows the discrete LIDAR pulses

(green dots), obtained during September, overlaid on a

Triangulated Irregular Network (TIN) of a bare ground

constructed from the discrete LiDAR pulses obtained during

December. The red dots in the image are the position data

obtained by georeferencing the Gaussian peaks of the smooth

waveform data, depicted by the cross hair symbols in the Figure

2 (top image). From the figure it is clear that, for the particular

laser pulse, the discrete return could record only one reflection

from the top canopy layer however the proposed algorithm

could detect 3 reflections of which the first one corresponds to

the same discrete return and the last one corresponds to the

ground level.

Intensity

145163" 185 205 225

Time (ns)

E Smooth Curve : Waveform data X GaussianPeak ® Discrete |

Figure 2. Top: Discrete return pulse (black square dot), raw

full waveform data (blue dots), smoothened waveform (red

curve) and Gaussian peaks (crosshair symbols). Bottom:

Discrete return pulses (green dots), full waveform return

pulses (red dots) constructed from the Gaussian peaks

depicted in the top image