1
1 \ b ]
I A) ♦
. ■ ?
*
iv, 2a
ft.
J- » •
f
,0 s
ul
In this paper we presented a car detecting and a tracking algo
rithm which have been especially chosen and adapted to the given
situation, the flying traffic monitor ARGOS. It was shown how
they work and that they brought satisfying results depending on
the environmental conditions. Furthermore it was shown, where
the approach has problems and continuous work can be done.
Surely the system can be improved in some points and a few of
them should be given here. First of all the street accuracy prob
lem which could be easily solved by using another database. And
it should be mentioned that there was already the attempt to use
the more accurate street database Atkis. On the one hand the co
ordinates were indeed more exact and yielded slightly better de
tection rates, but on the other hand the database divides the street
network into too small segments, which take a lot more time to
process one by one. Additionally the achieved data should be
mapped on Navteq segments, which would not be easy. So the
next step is to integrate the newest version of the Navteq database
being bought at the time.
Furthermore the edge detection could be optimized for example
by running it on the GPU, but it has not been considered so far.
Another idea is to compute the filtering in the frequency space.
The Fourier-transformed images and filters just have to be multi
plied in frequency space and transformed back. The only prob
lem is that the filters change with every street segment, so there
are four filters and four filtered images to be transformed ev
ery time. The approach was already explored, but the Fourier-
transformation implemented in OpenCV needs longer than direct
convolution, because it uses floating point numbers.
Next to this the detected cars could be verified by a more ex
pensive algorithm like a Bayesian Network or a Support Vector
Machine because some of the false positives do not look like a
car at all. So they would be easy to discard.
ACKNOWLEDGEMENTS
The authors would like to thank Dr. Franz Kurz and Dr. Do
minik Rosenbaum (Remote Sensing Technology Institute, Ger
man Aerospace Center, Oberpfaffenhofen) for providing the Geo-
TIFF images and navigation data.
REFERENCES
Figure 11 : Tracked objects filtered by CC-threshold 0.9 ( 100%
correct)
Ernst, I., Hetscher, M., Thiessenhusen, K., Ruhe, M., Bomer, A.
and Zuev, S., 2005. New approaches for real time traffic acqui
sition with airbone systems. Int. Archives of Photogrammetry,
Remote Science and Spatial Information Sciences 36, pp. 68-73.
Grabner, H., Nguyen, T. T., Gruber, B. and Bischof, H., 2008.
On-line boosting-based car detection fron aerial images. ISPRS
Journal of Photogrammetry and Remote Sensing.
Haag, M. and Nagel, H., 1999. Combination of edge element
and optical flow estimates for 3d-model-based vehicle tracking
in traffic image sequences. International Journal of Computer
Vision 35, pp. 295-319.
Hinz, S., 2004. Detection of vehicles and vehicle queues in high
resolution aerial images. Photogrammetrie - Femerkundung -
Geoinformation (PFG) 3/04, pp. 201 - 213.
179
