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 
Discrete time system 
  
Processed signal 
y(n) 
Observed signal 
x(n) ———4» hn) 
  
  
  
ha) = ast 110,9,16,2124,25242116,9,0,-11] 
Figure 3. Block diagram of FIR filter 
3. ACCURACY EVALUATION OF LIDAR DERIVED 
TERRAIN DATA 
We performed a test to evaluate and to compare the vertical 
accuracy of LiDAR derived terrain data constructed from the 
discrete LiDAR pulses and the proposed full waveform 
exploitation technique obtained during autumn and winter. For 
the accuracy evaluation, a typical hilly area with varying 
topography covered by Japanese cedar (Sugi) trees, Japanese 
cypress (Hinoki) trees and other deciduous trees was selected. 
Ground truth data were measured by performing total station 
survey. 
Figure 4 shows the terrain model of the site and total station 
surveyed check points. The check points were collected from 
the stripe of about 400m x 10 m area at the density of about 1pt/ 
4.5 sqm. The area was covered with both evergreen and 
deciduous tree varieties and the topography of the area varied 
from flat plane to rolling hill. The maximum slope of the test 
site was about 30 %. 
To evaluate the accuracy of LiDAR derived ground elevation, 
DTM of the ground terrain with 1 m grid size was constructed 
from the discrete LIDAR pulses obtained during September, full 
waveform LiDAR pulses obtained during September, discrete 
LiDAR pulses obtained during December and full waveform 
LiDAR pulses obtained during December respectively. The 
elevation of the ground on each DTM that corresponds to the 
horizontal location of the total station surveyed check points 
were then calculated directly by using ArcGis’s Surface Spot 
tool. 
The difference between measured point elevation data and 
LiDAR derived DTM elevations for the four cases are presented 
in Table 1. The RMSE between measured data and DTM data 
obtained by using discrete LiDAR pulses was 0.73 m during 
September, when deciduous trees were full of leaves. By using 
the full waveform exploitation technique, the DTM thus 
constructed showed an increase in the terrain data accuracy. 
The RMSE for the terrain data obtained by using full waveform 
LiDAR pulses was 0.59 m. 
During December, when the leaves fall off from the deciduous 
trees, RMSE between measured data and DTM constructed by 
using discrete LiDAR pulses was 0.22 m. For the same winter 
data set, the RMSE of the terrain data obtained by using full 
waveform LiDAR pulses was 0.21 m. 
507 
  
Figure 4. TIN of terrain model and ground truth surveyed 
points (blue dots) 
  
  
  
  
  
  
A NSSDA Accuracy; (m) 
DTM model UR (NSSDA Accuracy — 
* 
(RMSE, m) RN) 
September 0.73 143 
discrete 
September 
«m 0.59 TIS 
waveform 
December 0.22 0.43 
discrete 
December 
EIE 0.21 0.41 
waveform 
  
  
  
  
Table 1. Accuracy evaluation of LiDAR derived terrain data 
4. RESULTS AND DISCUSSION 
Point cloud generated from the discrete return pulses and full 
waveform return pulses along, a part of, the longitudinal section 
of the concerned area are compared and demonstrated in Figure 
5 for autumn and winter. The white dots in Figure 5 represent 
discrete return pulses and yellow dots represent the full 
waveform return pulses. For both seasons, we can see that the 
point cloud generated from the full waveform return pulses 
increased the vegetation/canopy detection and moreover 
increased the ground penetration significantly. 
To quantify the increment in the ground penetration, the point 
cloud generated from the full waveform return pulses and 
discrete return pulses were filtered automatically by using 
commercial software, TerraScan, to separate the ground points 
and vegetation points. The ground points obtained from the full 
waveform point cloud and the discrete return pulses are 
compared for autumn and winter respectively and presented 
below. 
During autumn, the ground penetration by discrete LIDAR was 
relatively poor with large gaps or missing ground points in 
some of the topographically complex terrains. However, by 
using the full waveform exploitation technique, ground point 
detection was considerably increased throughout the region.