Signal Processing - March 2017 - 67

There are several important and common information sources
for automotive positioning that can be utilized. modern camera-based vehicles can be equipped with traffic sign recognition, etc. If a database of the sign location is available, it fits
perfectly into the map-aided positioning framework by simply
adding a measurement update. Other information sources,
such as maximum allowed speed information for a road, can
be used for making a better proposal in the PF. In the future,
more localization sources will probably be available, e.g.,
information exchange from the infrastructure.

crete time, we have the inertial position p it + 1 = p it + TC ib v bt .
It can be shown that
.
q bi = 1 S (~ bbi) q bi,
2

In this section, the use of inertial sensors such as accelerometers and gyros, both for dead reckoning, sensor fusion, and as
stand-alone velocity estimation is discussed.

where ~ bbi is the angular velocity of the body system relative
to the inertial system described in body coordinates.
Figure 9 shows the ekf estimate based on Imu and GPs
signals from a Google nexus mobile phone using the data logging from [49], together with the GPs position using a discretized model For this application, it is essential that some
velocity estimation algorithm or position sensor is utilized to
mitigate the dead-reckoning problem with unknown biases in
accelerometer and gyro signals. Here the GPs sensor was utilized instead of map-matching techniques.

Inertial sensors

Virtual speedometer

an Imu measures acceleration and angular rotation. From these
measurements it is possible to integrate the underlying system to
achieve an estimate of position, velocity, and direction [14].

For some applications, an accurate velocity estimate is essential. a complementary method to the previously described
estimation is to utilize frequency analysis. It turns out that the
velocity of the vehicle is proportional to vibrations in the
accelerometer signal [50]. This can be utilized in the previously described positioning filter. It is not as accurate as Wss
information, but for a stand-alone application when Wss is
not available, data from a standard smart phone can be utilized. In Figure 10, the lateral accelerometer spectrogram is
depicted together with the angular velocity from the Wss (as
ground truth). The spectrum is formed at every instance
(downsampled to every second) by filtering the periodogram
of the lateral acceleration. There is usually a frequency related to the velocity. note that there are usually some overtones
as well. utilizing this in the frequency domain, it is possible
to construct a simple peak detector to estimate the velocity.
The point estimates utilizing only the maximal peak in the
accelerometer spectrum (batch-wise every second) is depicted

Inertial sensors and velocity estimation

Dead-reckoning IMU
The pure integration or dead-reckoning approach relies on
very accurate and expensive sensors that are not possible to
utilize in commercial passenger vehicles. The main problem
for affordable commercial sensors is that they have both
unknown sensor biases and drifts that are impossible to
remove. For instance, in the longitudinal acceleration direction, both sensor errors and hills will act like an unknown
time-varying bias. Integrating twice to achieve position will
very rapidly yield large position errors. For reliable standalone navigation, without relying on, for instance, GPs, this is
very difficult.

Map-aided positioning using Imu
an Imu-based map-aided position (stand-alone or in combination with Wss) was tested in [46], where the outcome was
that the Wss is a superior velocity sensor due to the aforementioned problems.

GPS Position
Est Position

gps and IMU fusion
North (m)

For modern eCus, the in-vehicle sensor cluster consisting of
rate gyros and accelerometers can be used together with a
GPs sensor to achieve position, velocity, and orientation estimates [14]. However, not all vehicles have these signals, or
they are not readily available. Hence, for many vehicles,
application external sensors might be helpful, particularly utilizing signals available in smart phones.
Consider the following state vector
T

x = ^q p i v b a bh ,
East (m)

where q is the quaternion vector, p i is the inertial position
vector, v b is the velocity in the body system, and a b is the
accelerometer vector in the body system. Let C ib be the conversion matrix from body to inertial systems. Hence, in dis-

Figure 9. The fused position ekf estimate and the gps position depicted
in a street map utilizing sensor data from a google Nexus Android mobile
phone.

IEEE SIgnal ProcESSIng MagazInE

March 2017

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