Signal Processing - March 2017 - 68

Accelerometer Spectrogram

250

Acceleration

Angular Frequency (rad/s)

WSS
200

Figure 12. The general idea for stand-alone map-aided positioning when the
Wss is not available is to use the acceleration vibration speed estimation as
input to the map-aided positioning algorithm. With proper outlier rejection to
the velocity estimates, the algorithm can function without Wss data.

150
100
50
0

200

400

600
Time (s)

800

1,000

Figure 10. An accelerometer spectrogram (filtered periodogram of the lateral
acceleration as a function of time) compared to the velocity from the Wss,
where the harmonics correspond to the wheel rotation frequency.

Est Speed
True Speed

Velocity (m/s)

50
40
30
20
10
0
0

200

400

600
Time (s)

800

1,000

1,200

Figure 11. Velocity from the Wss compared to estimates derived from
peaks in the accelerometer spectrum. sometimes there are outliers or the
wrong harmonic is selected, however, this can later be corrected by the
overall velocity filter.

in Figure 11. usually this correlates to the true velocity, but
there are some outliers. It is possible to improve the detector
by incorporating the knowledge of overtones, but this is not
done in this article.
For map-aided positioning, Wss information is always available. But for stand-alone applications, the vibration-based velocity estimate could be used as a complementary measurements
for map-aided positioning. Figure 12 illustrates how the speed
estimation can be used as input signal (instead of the Wss data)
to the map-aided positioning. For low velocities, this is not accurate enough, but pure accelerometer integration can potentially
be used for short periods of time to support the algorithm.

Discussion and conclusions
We have discussed the needs in future automotive localization
algorithms and pointed out that both accuracy and integrity
68

Velocity
VibrationMap-Aided Position
Based Speed
.
Positioning
ψ
Estimation

have to be improved compared to the navigation systems
today that rely on Gnss. For this purpose, we outlined a path
to future automotive localization algorithms based on a statistical signal processing approach, where information from various sensors and information sources are fused based on given
sensor models and an odometric motion model. The possible
sensors include in-vehicle sensors such as Wss, accelerometers, gyros, and external ones such as GPs. However, localization concerns the relative position of the own vehicle
compared to the surrounding, so the position relative the road
network is more important that the absolute longitude and latitude. a road map is the key information source for this purpose, and we have discussed the concepts of map matching
(basically projection of a position to the road network) and
map-aided positioning (where the road map is treated as a sensor). Furthermore, landmarks such as road signs detected by a
camera and the inclusion of car to infrastructure information,
and wireless sources (Bluetooth, Wi-Fi, and mobile positioning) will be crucial in the future, and this information is also
easily incorporated in our framework.
We have, in particular, highlighted the crucial concept of
map-aided positioning. utilizing measurements from a yaw rate
and wheel speed signals, we have shown that it is possible in
urban areas to position a vehicle with almost GPs accuracy without using any external Gnss positioning sensor when utilizing
the vehicle model and accurate road map information in a PF.
The incorporation of inertial sensor measurements for velocity
estimation utilizing accelerometer vibrations was demonstrated,
as well as the basic principle when using it for positioning.

Acknowledgment
This work was partially supported by the Wallenberg autonomous systems Program.

Authors
Rickard Karlsson (rickard@isy.liu.se) received his m.sc.
degree in applied physics and electrical engineering in 1996
and his Ph.d degree in automatic control in 2005, both from
Linköping university, sweden, where he is currently an associate professor. He has worked with automotive signal processing applications at nIRa dynamics since 2007 and target
tracking applications at saab dynamics in Linköping between
1997 and 2002. His research interests include positioning and
tracking applications mainly using particle filters.
Fredrik Gustafsson (fredrik@isy.liu.se) received his m.sc.
degree in electrical engineering in 1988 and his Ph.d. degree in

IEEE SIgnal ProcESSIng MagazInE

March 2017

http://www.m.sc http://www.m.sc

Table of Contents for the Digital Edition of Signal Processing - March 2017