Inertial Sensors and Their Applications - Signal Processing Systems - page 88

4 Human Motion Classification with Accelerometers

In addition to providing data for navigation purposes, inertial sensors can be
used for motion classification. As shown in Fig. 10 , the waveform of the norm
of acceleration ( 6 ) has different characteristics depending on motion mode of the
person carrying the unit. When walking, foot impacts clearly increase the variability
of the signal. When driving a car, engine vibrations, vehicle accelerations, and
road imperfections cause variations which are smaller than those occurring during
walking, yet distinguishable from the case of a stationary device where the only
source of variation is measurement noise.
In order to have a means for identifying these motion modes, features such
as sample variance or peak frequency need to be extracted from the acceleration
data. Figure 11 shows two such features: the sample standard deviation and the

peak frequency from non-overlapping five-second windows. In this example, the
classification is relatively easy, as the characteristics are clearly distinguishable. In
practice, there are more motion modes to classify and proper statistical tools are
needed to obtain useful classification results. In this section a brief introduction to
such tools is given.

σ

a b c

σ =

.

Fig. 10 Norm of accelerometer output in different motion modes: ( a ) walking, 0 24 g, ( b )

driving, σ = 0 . 071 g, and ( c ) stationary, σ = 0 . 0084 g

σ ( g )

0.3

Walk
Drive
Static

0.2

0.1

Fig. 11 Standard deviation
and peak frequency as
features

0

0 5 10 15 20

Peak frequency (Hz)