CMRT09: Object Extraction for 3D City Models, Road Databases and Traffic Monitoring - Concepts, Algorithms, and Evaluation 
qualifier used to quantify the suitability of the whole clustering 
method. 
k k 
K = 
i=1 ;'=1 /=1 
ZZ-v, -Z*- v >.- v j 
'=1 7=1 J /=1 
(13) 
where 
k = number of clusters 
3. EXPERIMENTAL RESULTS 
In this research, we have used two subsets of LiDAR data 
recorded from the city of Stuttgart, Germany. This data is 
recorded in four bands of first and last pulse range and 
intensity. The pixel size of this data is 30 cm. This means the 
average density of the recorded 3D points which is close to 9 
per meter. 
Figure 1. Datasets used in our research, a) first pulse range of the first dataset, b) last pulse range of the first dataset, c) digital aerial 
image of the first dataset, d) first pulse range of the second dataset, e) last pulse range of the second dataset, f) digital aerial image of 
the second dataset 
For better understanding of the objects, digital color (RGB) 
images have been also captured from this area using a medium 
format digital areal camera. In figure 1, color-coded first and 
last pulse images and also the RGB images of the investigated 
areas are illustrated. The trees can be distinguishes by 
comparing first and last pulse images. 
3.1 Results of Feature Extraction Algorithms 
The level of the discrepancy between first and last return 
heights before and after applying the gradient filter is shown in 
figures 2, 3 for our two datasets. The discrepancy was larger 
than zero in the tree regions as expected. 
a b 
Figure 2. The normalized difference of the first and last pulse 
range images for our first dataset, a) before gradient filtering, 
b) after gradient filtering 
a b 
Figure 3. The normalized difference of the first and last pulse 
range images for our second dataset, a) before gradient filtering, 
b) after gradient filtering 
The feature image of applying the morphological operator on 
last pulse range image with 5*5 structuring element is 
illustrated in figure 4. Here, the size of structuring element is 
selected by experiments on these two datasets. 
62