inbul 2004 
reviously 
rolled by 
lels. 
nalize for 
one factor 
eters); the 
0 speed. 
/e similar 
pe curve) 
the key to 
es have to 
w vehicle 
ular buffer 
n red, the 
as similar 
ices high 
he sensor 
methods, 
urements. 
nted with 
es and the 
e attached 
  
mage) 
ind laser 
| mapped 
there 1s 
ed point 
LiDAR 
rly to the 
etry. 
from the 
for the 
ofile, we 
a packed 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
parking lot. The goal was to get information about as many 
vehicle categories, as possible. 
Categorizing Issues 
The ground based laser scanning measurement was performed 
in Hungary, where the traffic patterns are basically 
representative for Europe, while in our research we used 
airborne data sets acquired in the USA and Canada. where 
characteristics of the traffic are noticeably different. From the 
nineties, there are much more MPVs (especially light trucks) 
sold in the US as passenger cars. In Europe, there are much less 
MPVs running on the roads, and most of them are minivans, not 
SUVs and light trucks, while in the passenger car category, the 
proportion of hatchback cars is definitely higher, whereas the 
proportion of the sedans is much higher in the US. Therefore, 
we focused on vehicles that can be representative in both 
regions. 
   
Figure 5. The point clouds of the test vehicles 
For our model-based investigations, we used the data set 
acquired about a Ford Mondeo with a conventional sedan 
profile, and about a VW Golf, which is a hatchback: two very 
popular cars with very typical shapes (Figure 5). 
PCA Test 
First, these two typical vehicles were classified using PCA: the 
cars have been categorized correctly. In the PCA, where 6 
parameters were involved (length, width, heights), these cars 
are on the border of two groups (passenger cars traveling along 
and against flight direction): the cars from the airborne mapping 
are either elongated or shortened, that is why the moving 
directions can also be distinguished in the 6 parameter case. 
  
  
«1 pire 1 - i SORA: : Leo i À 
0 5 10 15 20 2 30 35 40 45 50 
oh 
Figure 6. The new cars in the PCA-based clustering 
In our ground based laser measurements the cars parked in a 
parking lot, therefore, their length values in the dataset fall 
between the shortened and elongated length values of the 
airborne dataset (Figure 6). 
Profile Determination 
In order to derive the shape of vehicles. the sides have been 
cropped, since the points reflected from the side of the vehicle 
are not included in the profile determination. Just for the sake of 
comparison, in the airborne campaign, typically 20-30 points 
are reflected from an elongated car, in the ground based dataset 
these test vehicles have about 200-300 000 points, despite the 
side-looking sensor position (the back of the car is shadowed). 
18 
Em 
  
Paci T > 
PLN 
| Jo S v 
| / V ^ + 
Î ^ 
i MÀ y REN 
1 | € 4 S M C 
| = 
| > 
| PRE 7 K 
i e a ^ 
| pa J^ \ 
| PA 
i Pd 
08 i 
| 
| 
i 
€ i 
o 5 10 15 
Figure 8. Shapes of the airborne campaign 
  
Figure 9. The combined shape curves and envelopes 
As it can be seen in the Figures 7 to 9, the shapes of the cars 
measured with the ground sensor are fit in the previously 
defined buffer zone. The only difference is the back part of the 
vehicles, where the hatchbacks are remarkable higher than the 
sedans. The classification cannot be exclusively based on the 
length, since it depends on the speed of the vehicle; therefore, 
the shapes in the figures are normalized in length.