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 
based on the calibrated reflectance was applied to single, first 
and last echoes. Though the use of the calibrated reflectance 
allowed to strongly reduce the number of points, some 
vegetation elements, like trunks, could not be completely 
removed as they showed reflectance values similar to those of 
ground points. An example of the results of this kind of 
filtering is shown in figure 5, where the red points represent 
single and first targets filtered out by applying a reflectance 
threshold of -6 dB. A lower threshold (-8 dB) was set for the 
group of last echoes because the reflectance values of 
vegetation elements resulted to be more similar to that of 
ground points. As discussed in previous section, this is due to 
the attenuation effect that preceding targets produce on the 
emitted laser pulse, so that reflectance values recorded for the 
last echoes are not range independent and do not actually 
represent the backscattering properties of corresponding 
targets. For each class of echoes, the reflectance thresholds 
were determined empirically through the analysis of the 
reflectance distribution plot provided by RiscanPro software 
and by checking the reflectance value of some points (clearly 
recognized as vegetation and ground) manually picked in the 
3D view of the test area. 
  
Figure 5: Example of vegetation filtering based on the 
calibrated reflectance. 
After the pre-filtering of VZ-400 laser data, the same 
processing steps were applied to the datasets derived by both 
instruments. Two different spatial filters were then applied to 
the corresponding 3D models, in order to eliminate as much 
vegetation as possible: an iterative filter (Axelsson, 2000) 
originally developed for the filtering of ALS data, and a 
custom morphological filter, developed by the authors for the 
mapping and quantification of vegetation in forested areas 
(Pirotti et al., 2011). 
In the iterative the original point cloud S is firstly projected 
on a reference plane p; and then rasterized on a regular grid 
by selecting the point with minimum laser elevation (Z axis 
orthogonal to the plane) Next S is compared with the 
obtained DEM and only the points closer than a distance 
threshold are preserved. This process is repeated iteratively 
by reducing at each step the size of both the grid cell and the 
threshold, until the vegetation is completely removed. 
In the morphological filter three main group of parameters 
are derived from a dataset acquired with an echo-digitizing 
system: spatial coordinates of measured points, amplitude 
and ordinal number of return signal. These two features are 
used in the first two steps of the algorithm in order to extraxt 
candidate ground points from the original point cloud. A 
threshold is applied to the amplitude data recorded with the 
laser measurements. Such threshold is determined by the first 
value of the last quartile of the cumulative distribution 
function of the amplitude values. In a second step, a custom 
515 
morphological filter (Haralick and Shapiro, 1992), composed 
by erosion (E) and dilation (D) operators, is applied to 
maximum and minimum laser elevations falling inside a 
regular grid (eq. 3). By iteratively decreasing the cell size C 
of the grid, a set of DTMs are obtained until the vegetation is 
almost completely removed. The conceptual workflow of this 
algorithm is shown in figure 6. 
D, =max(z,) E,-min(z,) (3) 
Cx ur, OC £x, v. RC 
Space (XYZ) | | Amplitude | | Return number | 
  
  
emma 
i PAN == NoR |- False 
  
True 
{ > threshold "T false 
Ground candidates rue 
M ee I"fiveshold: "distance | 
Closing L__fromplane” ^ ] 
a ee 
U 
  
  
  
  
  
Morph. Filter 
Openin 
Pe 
  
  
  
  
   
  
  
  
  
       
SS À 
Terrain model 
       
  
Surface model Vegetation points 
     
Difference 
  
  
Weighted|Voxel density 
  
  
: Canopy height map : i Vegetation density map : 
d eecocsossonet 
  
Figure 6: Workflow of the morphological filter; PRN is 
denotes the point return number, while NoR the total number 
of returns. 
Numerical results of the application of the iterative and of the 
morphological filters for the vegetation removal are presented 
in table 4, while the final filtered point clouds derived from 
the Riegl LMS-Z620 and VZ-400 laser scanners are shown in 
figures 7a and 7b, respectively. Note that the numerical 
values shown in table 4 were rounded to the most 
significative digit for clarity sake. 
Table 4: Results of the filtering of the Riegl LMS-Z260 and 
Riegl VZ-400 datasets. 
  
LMS-Z620 VZ-400 
# of initial points 2500000 12200000 
# of points after multi- N/A 10750000 
target filtering 
# of points after 
filtering based on N/A 6230000 
calibrated reflectance 
sotpoints loitaier 400000 3020000 
iterative filtering 
# of points left after 415000 3035000 
morphological filter