Wheel Extraction based on Micro Doppler
Distribution using High-Resolution Radar
Dominik Kellner
∗
, Michael Barjenbruch
∗
, Jens Klappstein
†
, J
¨
urgen Dickmann
†
and Klaus Dietmayer
∗
∗
driveU / Institute of Measurement, Control and Microtechnology Ulm, Germany
Email: firstname.lastname@uni-ulm.de
†
Daimler AG, Ulm, Germany
Email: firstname.lastname@daimler.com
Abstract—With the advent of advanced driver assistant systems
(ADAS) in urban scenarios, a fast and reliable classification and
motion estimation of wheel-based vehicles such as cars, trucks or
motorcycles is crucial. The fact that the wheels’ velocities differ
from the vehicle’s chassis velocity is exploited. For the first time,
a fully automated approach based on the Doppler distribution
extracts the exact positions of the wheels. The Normalized Doppler
Moment is calculated, describing the Doppler signature of each
reflection based on the Doppler distributions of wheels. Locations
with high values reveal the positions of the wheels. Besides the
classification, the vehicle’s orientation and therefore the driving
direction can be estimated. Furthermore the position of the rear
axle is estimated, which is essential for a reliable prediction
of rotational movements and yaw rate estimation. Experimental
results with a 77 GHz automotive radar sensor demonstrate the
feasibility of the approach.
I. INTRODUCTION
When using high resolution DBF (digital beam forming)
radars, more than one reflection is received from an extended
object. A universal radar response model which describes the
number and position of received reflections for an unidentified
object with unknown aspect angle is not available. To detect
and react to critical situations, it is important to precisely
capture other traffic participants. Their motion state is resolved
model-free by analyzing their velocity profile (Doppler veloc-
ity over azimuth angle) [1]. The position of detected wheels
can be used for classification and as a representation of the
spatial position, extension and orientation.
An orientation estimation is possible if two wheels are
captured on either a common side or common axle. Identifying
the rear axle through a detected wheel and assuming no lateral
drift, the center of rotation of the vehicle is determined. The
center of rotation is the point on the vehicle where the velocity
vector is parallel to the object orientation. This is essential in
object tracking to correctly predict the object’s pose during
turn maneuvers [2]. Inaccurate assumptions of the vehicle
length and ratio length/rear axle are normally used for this
task.
The approach is based on Micro Doppler effect, which arise
when vibrations or relative motion of parts of the illuminated
objects are present. Additional patterns are induced in the
Doppler frequency spectrum, which appear as side-bands
around the Doppler frequency of the bulk motion [3].
II. RELATED WORK
The analysis of the Doppler signature is mainly used for
classification tasks in Inverse Synthetic Aperture Radar (ISAR)
[4]–[7] or Synthetic Aperture Radar (SAR) [8].
Using 2D-ISAR, an analysis of the Micro Doppler signature
of rotating wheels in combination with high-range resolution
over time is performed [4]. No automated extraction is pro-
posed, but the Doppler velocity distribution is analyzed only
manually on a single sequence.
Different rotating objects (e.g. wheels) are simulated and
their Doppler signatures are analyzed with 1D-ISAR in the
joint time-frequency domain [5]. Pedestrians, wheeled vehicles
and tracked tanks are classified using the correlation of the Mi-
cro Doppler signature with a previously extracted pattern [6].
The approach analyses the Doppler signature using dynamic
time warping in relation to the aspect angle, but does not run
in real-time. An Empirical Mode Decomposition is applied to
classify tracked vehicles and wheeled vehicles mainly by the
larger Micro Doppler spread of tracked vehicles, whereas it is
assumed that a wheeled vehicle has a component only from
the bulk motion [7].
To the best knowledge of the authors this is the first
work which not only detects, but also performs an automated
extraction of 2D wheel positions of a vehicle. For application
in the driving assistant domain, the approach has to deal with
a moving sensor and an unknown movement of the target
vehicle, therefore the application of SAR and ISAR is not
possible. An automotive DBF radar is used and the extraction
is performed in a single measurement cycle in both range and
cross-range direction. Multiple wheels are detected in real time
and a precise estimation of their position is achieved.
III. THEORY
A Doppler radar is only able to measure the radial velocity
(v
D
) over the azimuth angle (θ). Derivation shows that inde-
pendent of the motion state of the vehicle with 3 degrees of
freedom (DOF) the Doppler velocities always form a cosine
over θ with 2 DOF (amplitude and phase shift) - called the
velocity profile [1].
Among the variation of the Doppler velocity caused by the
object motion and different azimuth angles, Micro Doppler
arises due to the deviating velocity of the wheels. The ap-
proach has to handle different wheel types (e.g. aluminum or
2015 IEEE MTT-S International Conference on Microwaves for Intelligent Mobility
978-1-4799-7215-9/15/$31.00 ©2015 IEEE