International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012 
XXII ISPRS Congress, 25 August - 01 September 2012, Melbourne, Australia 
  
local point density variations to include the required number of 
points for reliable plane definition. 
3.4 Boundary Detection 
The major drawback of parameter-domain segmentation 
techniques is that the spatial connectivity of points belonging to 
each segment is not considered. Therefore, the points belonging 
to coplanar but spatially disconnected planes will be segmented 
into the same group. To resolve such ambiguity, a 
neighborhood analysis is conducted through boundary detection 
of the clustered points. The process of searching for each 
boundary point is carried out in the local neighbourhood of the 
previous boundary point. In order to define adaptive 
neighbourhoods for sequentially finding the boundary points, 
the estimated local point density indices should be taken into 
consideration. 
3.5 Terrain/Off-terrain Classification 
In order to classify the clusters of LiDAR points into those 
belonging to terrain or off-terrain objects, the discontinuity 
measures between adjacent clusters should be considered. The 
adjacency relationship between these clusters is defined by 
analyzing the neighbourhoods of each cluster's boundary 
points. For all the points in the boundary of each cluster, there 
exist neighbourhoods which include the points belonging to its 
adjacent clusters. In order to define adaptive neighbourhoods 
which include the points belonging to adjacent segments, the 
local point density index at each point's location should be 
considered. 
4. EXPERIMENTAL RESULTS 
In this section, the performance of the newly developed 
methods for the estimation of the local point density indices and 
the impact of considering them on the quality of LiDAR data 
segmentation results will be investigated by conducting 
experiments using airborne and terrestrial LiDAR datasets. 
4.1 Airborne LiDAR Data 
The utilized airborne LiDAR dataset for this experiment 
(Figure 7.3) has been collected over an urban area in 
Switzerland with the Scan2Map mapping system. This dataset 
exhibits significant local point density variations (estimated by 
the approximate method) as shown in Figure 7.b. The results of 
the planarity check for the individual points using the eigen- 
value analysis and adaptive cylinder methods are presented in 
figures 7.c and 7.e, respectively. Once the planarity of 
individual points was checked and local point density indices 
were calculated, the point density maps for the points belonging 
to planar surfaces are generated using estimated local point 
density indices (figures 7.d and 7.1). 
    
] 28.89 Pnts/m" 
0.12 Pnts/m? 
    
W 28.05 Pnts/m? 
   
   
EE Non-Planar \ s 
0.03 Pnts/m? 
l| 26.24 Pnts/m* 
38 Non-Planar \ ii 
(e) (f) 
Figure 7. Airborne LiDAR dataset: (a) original LiDAR data, 
(b) generated point density map using the approximate method, 
(c) planarity check result using the eigen-value analysis relative 
to the point in question, (d) generated point density map based 
on eigen-value analysis relative to the point in question, 
(e) planarity check result using the adaptive cylinder method, 
and (f) generated point density map based on adaptive cylinder 
method 
  
0.004 Pnts/m° 
To verify the importance of the processing of LiDAR data while 
considering the estimated local point density indices, the 
provided airborne datasets is processed using an adaptive 
segmentation approach (Lari et al., 2011). The segmentation 
process is carried out with and without considering local point 
density variations. Figure 8.a shows the result of the airborne 
LiDAR data segmentation without considering local point 
density variations while Figure 8.b shows the result of the 
segmentation of the same data considering the local point 
density variations. Qualitative evaluation of the derived 
segmentation results through visual inspection of Figures 8.a 
and 8.b shows that considering the local point density indices 
avoids some problems in the segmentation results, the most 
visible one is the over-segmentation problem — as highlighted 
within the red rectangles. 
  
Figure 8. Airborne LiDAR dataset segmentation results: 
(a) without considering local point density variations and 
(b) considering local point density variations 
4.2 Terrestrial LiDAR Data 
The terrestrial LiDAR dataset (Figure 9.a) has been obtained 
from a building façade in the University of Calgary campus 
Int 
using 
map 
show 
indiv 
cylin 
respe 
plane 
densi