In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Vol. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009
FAST VEHICLE DETECTION AND TRACKING IN AERIAL IMAGE BURSTS
Karsten Kozempel and Ralf Reulke
German Aerospace Center (DLR e.V.), Institute for Transportation Systems
RutherfordstraBe 2
12489 Berlin
karsten.kozempel@dlr.de, ralf.reulke@dlr.de
KEY WORDS: aerial, image, detection, tracking, matching
ABSTRACT:
Caused by the rising interest in traffic surveillance for simulations and decision management many publications concentrate on auto
matic vehicle detection or tracking. Quantities and velocities of different car classes form the data basis for almost every traffic model.
Especially during mass events or disasters a wide-area traffic monitoring on demand is needed which can only be provided by airborne
systems. This means a massive amount of image information to be handled. In this paper we present a combination of vehicle detection
and tracking which is adapted to the special restrictions given on image size and flow but nevertheless yields reliable information about
the traffic situation.
Combining a set of modified edge filters it is possible to detect cars of different sizes and orientations with minimum computing effort,
if some a priori information about the street network is used. The found vehicles are tracked between two consecutive images by
an algorithm using Singular Value Decomposition. Concerning their distance and correlation the features are assigned pairwise with
respect to their global positioning among each other. Choosing only the best correlating assignments it is possible to compute reliable
values for the average velocities.
1 INTRODUCTION
1.1 Motivation
The gathering of traffic information is a base for all kinds of traf
fic modeling, simulation and prediction for tasks like emission
reduction, efficient use of infrastructure or extension planing of
the road network as well as the intervention and resource planing.
Next to the use of inductive loops, Video Image Detection Sys
tems (VIDS) have become a common alternative due to their low
price as well as their simplicity and effort of installation. Further
more inductive loops can’t cover the whole road network and a
lot of data has to be estimated. Especially during mass events or
disasters with huge congestions or road blocks, they can't yield
reliable information.
For this special purpose the German Aerospace Center (DLR
e.V.) developed the ANTAR system for airborne traffic monitor
ing on demand. During the soccer world cup 2006 it was success
fully applied to gather traffic data and predict traffic situation in
three German cities (Ruhe et al., 2007). Based on this the DLR is
developing the ARGOS system for wide-area traffic monitoring
(fig.l). It contains next to a radar system the 3K-Cam, a device
of three digital cameras with 16 mega pixels each. Together they
cover an area of 2,5 km x 0,7 km with a resolution of 20 cm at
an altitude of 1000 m over ground. Additionally a GPS/IMU-unit
is used to record positioning and orientation data for every image
taken. Thereby the achieved image data gets orthorectified and
georeferenced on-board which means that the images arriving the
traffic detecting software can be used as map images with given
orientation and scale. A fact that makes measuring distances and
computing velocities less complex.
In the first chapter the conditions related to the observation sys
tem are explained as well as the published work on this area. The
second chapter describes the used algorithms, a modified edge
filter for fast vehicle detection and an extended singular value de
composition concerning distances and correlations for tracking in
very short sequences. After this the results with a few examples
are presented. Finally a conclusion with considering possible fur
ther research will close the paper.
48 Mpix camera system
radar system
^ gigabit
ethemet
GPS/IMU
data interpretation
data recording data processing & storage
Figure 1 : Traffic monitoring system ARGOS
1.2 Special conditions
There are two special points to consider while developing de
tection and tracking. It should be respected that the preprocessed
images depending on their altitude over ground can be very large,
in the shown case 25-30 mega pixels. That’s why the detecting
algorithm should be rather fast than exact. Already the previous
system ANTAR demonstrated that for an overview of the traffic
situation a completeness of two thirds is acceptable.
Due to the mentioned size of the images (original size is 16 mega
pixels) they cannot be transmitted continuously. After a burst of
a few images (2-4) the stream is cut to save them. Therefore it is
not necessary to implement a complex tracking filter which needs
a long period to adapt to the scene.
1.3 Related work
A grand variety of approaches in vehicle detection as well as in
object tracking has been released in the last years.
Detection methods can be divided into two groups, depending
on the kind of model being used. The use of explicit models
