International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, \ ol XXXV. Part B2. Istanbul 2004 
Using the ground models, sample shapes can be derived for the 
subclasses, e.g. hatchbacks, sedans and caravans in the 
passenger car category and minivans, light trucks and SUVs in 
the MPV category. 
5. CONCLUSIONS 
The vehicle classification and recognition methods described in 
this paper show reasonable performance in categorizing 
vehicles, which proves the capability of LIDAR in supporting 
traffic flow applications. This paper focused on classification 
using ground-based laser scanning datasets that could lead to a 
refined classification in the future. In addition, we developed 
technics to decrease the impact of corrupting points (reflected 
from the side of the vehicle, multipath reflection, other objects 
on the road etc.), since the vehicle points can be filtered either 
by cropping the outliers or filtering the shapes and adjusting the 
resolution along the profile. 
As opposed to airborne measurements, the ground-based laser 
scanning campaign was performed in Hungary, where the 
traffic pattern is significantly different from that in the US. 
Therefore, we have chosen widely used vehicle models as test 
vehicles, which can be representative even in the US. In the 
ground laser dataset the point density was exceptionally high, 
especially compared with the airborne set; in the profile 
determination process the overwhelming number of points 
ensured a very good resolution. The resulted shapes are nicely 
fit within the previously derived buffer zone (with sample 
shape); the only visible difference 1s caused by the higher back 
parts of European cars. 
The shape-based method can be directly used as a classifier, or 
can be used to enhance the previously used PCA based 
classification. Applying ground based laser scanned data, a 
detailed shape library can be established, which can be used to 
distinguish between sub-classes within the categories. 
The potential of using intensity values in the classification 
procedure has also been investigated. Although, the intensity 
maps seem to be applicable for segmentation, the point density 
of our dataset (2.4 points/m^) might not be sufficient for that 
purpose. 
Acknowledgements 
This research was partially supported by the NCRST-F 
program. The authors would like to thank to Woolpert LLC and 
Optech International for providing the airborne LiDAR 
datasets, and to Piline Kft. and Riegl Laser Measurement 
Systems for providing the ground-based laser scanned point 
cloud. 
We also would like to express our gratitude to the Thomas 
Cholnoky foundation, which partially sponsored the initial 
rescarch work in the 2001-2002 period. 
References 
Duda, R. .O. Hart, P. E. Stork, D. G. (2001* Pattern 
Classification, Wiley, New York 
Lovas, T. (2004a): Comparison of Vehicle Recognition 
Methods Using LiDAR Data, Il. PhD Civilexpo Conference, 
Jan 29-30, in press, Budapest 
Lovas T. , Barsi A,, Toth C., (2004b): Detecting Moving 
Targets in Laser Scanning, Proc. ASPRS Annual Conference, 
Proc. ASPRS Annual Conference, Denver, CO, May 23-28, CD 
ROM. 
138 
Rojas, R. (1993): Theorie der neuronalen Netze - Eine 
systematische Einführung, Springer Verlag, Berlin 
Toth C., Grejner-Brzezinska D.. Lovas T. (2003a): Traffic Flow 
Estimates from LIDAR Data, Proc. ASPRS Annual Conference, 
May 5-9, pp. 203-212, CD ROM. 
Toth C., Barsi A., Lovas T. (2003b): Vehicle Recognition from 
LiDAR Data, International Archives of Photogrammetry and 
Remote Sensing, Vol. XXXIV, Part 3/W13, pp. 162-166, CD 
ROM. 
Zhao, Y. (1997): Vehicle Location and Navigation Systems, 
Artech House, Inc., Boston 
Flood, M. (1999): www.airbornelasermapping.com 
ALG( 
ON I 
b Ins 
KEY W 
ABSTR 
This pay 
correctic 
factors r 
brought 
elevatiot 
polynom 
this metl 
(Dali, Y: 
of the fc 
about 2€ 
SAR im 
orthorec 
The geo 
today tl 
(SAR) a 
angle of 
This mo 
SAR in 
applicati 
ortho-im 
Many S; 
orthorec 
DEM 
orthorec 
remote | 
methods 
rectificat 
Range-D 
Polynon 
Since 1 
1975: TF 
polynom 
for recti 
orthogra 
can be u: 
Collinea 
This met 
model, 4 
elements 
nor do ai 
model, t 
same wit 
with the 
conditior 
projectiv