Quantitative Gait Measurement with Wave Doppler-radar for
Elderly Walking Speed Recognition
Maha Reda
1
, Aly Chkeir
1
, Racha Soubra
1
and Mohamad Nassereddine
2
1
University of Technology of Troyes, Troyes, France
2
Lebanese University, Faculty of Sciences, Hadath, Lebanon
Keywords: Frailty, Radar-doppler, Velocity, Irregular Sampling, Vicon.
Abstract: This paper studies the use of a device based on a Doppler sensor in estimating the gait velocity in a non-
controlled environment. It provides signals of the instantaneous velocity with an irregular time sampling. A
high accuracy motion capture system, Vicon, was employed to provide the reference data for device
evaluation. The gait parameters have been validated with a Vicon motion capture system in our lab. A proper
algorithm based on the Lomb-Scargle periodogram was proposed to extract features from the radar signals
such as the dominant frequency and the number of steps performed. These features were then used to calculate
gait parameters such as the gait velocity and the step duration on a 5m walking sequence at a normal pace.
The results showed the reliability of the Doppler device in estimating the gait velocity (mean error = 5.1%).
1 INTRODUCTION
The elderly population in most European countries is
growing significantly. This high increase emphasizes
the need for a strategy to maintain the wellbeing of
the elderly and enable them to live in good conditions
(Chłoń-Domińczak et al. 2014). Even when in their
normal environments, old people are subject to highly
stressful events due to the decreasing physical ability
and activity and to the high risk of pathologies that in
most cases lead to frailty. Frail people are vulnerable
to the risk of falling that causes tremendous amounts
of mortality and morbidity. Researches have shown
that the study of factors contributing to frailty can
help in predicting falls and eventually preventing
them.
Linda Fried’s model presents a description of
physical frailty based on five indicators (Fried, 2001).
Among these indicators, walking speed was proven to
be the one that has the greatest correlation with the
frailty index (Theou et al. 2011). Walking speed is
easy to calculate and this is usually done in clinical
tests by measuring the time taken to travel a certain
distance (usually 10 m) at a normal pace. However,
in order to assess the risk of falling, a continuous
study of the walking speed is needed, especially in the
home environments of the elderly where they are
walking at their normal pace and going about their
daily life activities. For that reason, new technologies
are needed that are able to calculate the walking speed
on a daily basis.
In the work presented herein, we demonstrate the
use of a device based on a commercially available
Doppler sensor that calculates the instantaneous value
of the walking speed. The device emits an
electromagnetic wave and returns a square signal with
a frequency equal to the frequency shifts.
The velocity is then calculated using the following
equation:
∆=
∆×
(1)
where ∆is the calculated velocity,∆ is the Doppler
frequency, is the speed of light and
the
fundamental frequency of the radar.
In a previous study (
Jaber et al. 2014). The
performance of the device was evaluated with
reference to clinical tests over a 3m distance. With
preliminary signal processing, it was shown to be a
reliable device in calculating both the instantaneous
and the mean walking speed.
In this study, the range of the Doppler sensor was
adjusted so that motion can be captured up to 10m
away. We employed a high-accuracy motion capture
camera system, Vicon, for ground truth data, and
more advanced signal processing techniques in order
to improve the capacity of the device to calculate the
410
Reda, M., Chkeir, A., Soubra, R. and Nassereddine, M.
Quantitative Gait Measurement with Wave Doppler-radar for Elderly Walking Speed Recognition.
DOI: 10.5220/0007521604100414
In Proceedings of the 12th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2019), pages 410-414
ISBN: 978-989-758-353-7
Copyright
c
2019 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
mean velocity. We considered that the highest
contribution to the velocity signal of the radar comes
from the pelvis since it has the largest surface area
(Yardibi et al. 2011). The mean velocity calculated
using the Doppler radar signals is then compared to
the mean velocity of the center of mass of the pelvis
obtained using Vicon. The main challenge in this
study was the irregularity of the signals obtained from
the Doppler device, which was resolved using the
Lomb-Scargle Periodogram (LS) (Scargle et al.
1982). A proper algorithm was proposed to extract
features from the radar signals that would help in
calculating the gait velocity.
2 EXPERIMENTAL SETUP
2.1 Doppler
The radar used in our study is a commercially
available Doppler sensor (X-Band Doppler Motion
Detector MDU 1130, Microwave Solutions LTD.,
Marlow United Kingdom) with a carrier frequency of
9.9 GHz. The device was placed on a 1m high table
and was put in a friendly, decorative box in order to
be more acceptable for future experiments in an
elderly environment. The radar was connected to an
application installed on a Tablet via Bluetooth, in
such a way that the recorded values of the velocity are
stored in a file and sent directly to the Tablet once
acquisition is over.
2.2 Vicon
The commercial 3-D motion analysis system, Vicon
system, used in this study consists of eight infrared
cameras and motion capture software installed in a
computer. During the acquisition, the cameras emit
infrared light that is reflected by the retro reflective
markers put on the moving subject. The reflected light
is then picked up by the cameras and eventually the
spatial position of each marker in an x, y, and z
coordinate system is obtained. In our study, we used
16 reflective markers that were put on the toes, heels,
wrists, fingers, shoulders and pelvis. Out of these
markers, we were able to extract the exact
instantaneous location of the center of mass of the
feet, hands, pelvis and shoulders. We considered that
the mean velocity of the center of mass of the pelvis
yields approximately the mean gait velocity. This
velocity was calculated simply by dividing the
distance travelled by the center of mass of the pelvis
by the time (over 10 ms intervals).
A synchronization system was built in order to
make sure that the radar and Vicon are acquiring the
same data for the exact walking sequence of each
subject. The system consisted of two infrared sensors
barriers that detect the start and the end of the walking
sequence. These two sensors are connected
simultaneously to the Tablet and the Vicon system.
2.3 Protocol and Data Processing
The aim of this study was to compare the velocities
obtained by the Doppler device to those obtained by
Vicon. Four persons, all of whom were members of
the laboratory, had given their informed consent to
participate in the experiments. Each subject
conducted five 5m walks towards the device,
resulting in 20 walking sequences. The signals
obtained from the radar contain the instantaneous
velocity of each movement detected in an irregular
time sampling. To filter the signals and remove high
and low frequencies, we used a Butterworth bandpass
filter between 5 and 100 Hz, applied to the signal in
the time domain. The frequencies contained in the
filtered signal were then obtained using the LS
periodogram (Eq. 2) which is known for detecting and
characterizing periodic signals in an unevenly
sampled data (
VanderPlas et al. 2018). A time delay
(Eq. 3) was added in order to overcome the problem
of the irregularity of the signals.
(
)
=
∑
∑
+
∑
∑
(2)
tan(2)=
∑
(
)
∑
(
)
(3)
The frequency with the highest peak corresponds to
the frequency of the steps made during the walk (Fig.
3), i.e. the number of steps performed in one second.
By inversing this frequency, we obtained the time
taken to complete each step for each of the subjects
who participated in the experiments. Furthermore, an
FFT was also conducted on the radar signals, with a
sampling frequency equal to the length of the signal.
The frequency peaks obtained from the FFT
corresponded to the number of steps performed by the
subject during the whole walking sequence. The
velocity was then calculated using this equation:
=
()
×
(4)
As for Vicon, an FFT was also conducted to obtain
the frequency of the velocity signals of each center of
gravity and then it was compared to the frequency
obtained from the LS periodogram applied to the
corresponding radar signals. We considered that the
Quantitative Gait Measurement with Wave Doppler-radar for Elderly Walking Speed Recognition
411
gait velocity corresponds to the mean velocity of the
center of the pelvis. This velocity is calculated by
taking the average of the mean velocities of the four
markers put on the pelvis. The mean error was then
calculated using this equation:
=
|
|
(5)
2.4 Algorithm
The algorithm proposed is described in the following
steps:
• Filter the radar velocity signal using a Butterworth
bandpass filter between 5 and 100 Hz
• Apply the LS periodogram and extract the
frequency with the highest energy peak that
corresponds to the number of steps performed per
second
• Apply an FFT to extract the number of steps
performed during the walk
• Calculate the velocity using Eq. 4.
The LS periodogram used was the one integrated
in Matlab, where the algorithm was implemented.
Note also that in the FFT performed in the third step,
the sampling frequency was equal to the Doppler
signal length in order to obtain the number of steps.
The algorithm was applied to all 20 walking
sequences
3 RESULTS AND DISCUSSION
An example of the signal obtained by the radar is
given in Fig. 1.
Figure 1: Radar signal.
This signal contains the velocity of each
movement detected by the sensor. When applying the
filter, we remove the high velocity peaks that
correspond to the movement of the limbs and are
considered as noise. When comparing it to the Vicon
data, we can see that the filtered signal corresponds
approximately to the velocity signal of the center of
the pelvis (Fig. 2). The frequency content of the
Vicon signal is presented in (Fig. 4). The maximum
frequency was calculated for all the signals produced
by the radar and compared to the corresponding
signal obtained from Vicon (Fig.3). We can see that,
after filtering, the frequencies of the movement of the
limbs are totally removed, and the remaining
frequency was that of the pelvis. The number of steps
performed during the whole walking sequence is
presented in (Fig. 5). By using Eq. 4 we were able to
calculate the gait velocity from the radar signals and
compare them to the Vicon velocities. Results of
Figure 2: Graphical representation of the velocity signals of
the centers of the right, the left feet, the pelvis and the
filtered signal obtained from the radar (blue).
Figure 3: The LS periodogram showing the frequency peaks
of a filtered radar signal.
HEALTHINF 2019 - 12th International Conference on Health Informatics
412
Figure 4: FFT of the pelvis signal obtained from Vicon.
18 walking sequences were obtained (Fig. 6). Two
sequences were found to be corrupted and were not
used in the study.
However, the results shown here correspond to a
small group of young people (between 23 and 28
years old) and does not represent the results on the
elderlies. Another limitation of the study was the
necessity to know the distance travelled beforehand
in order to perform the algorithm.
The mean error obtained is 5.1%. Compared to
previous studies (
Cuddihy et al. 2012), where the radar
signals had a regular time sampling, in this study we
were able to obtain an approximate value of the gait
velocity with a lower mean error, even when we were
dealing with irregular velocity signals.
4 CONCLUSION
In this study, we used a device based on Doppler
sensor in order to estimate the gait velocity. The
device was designed in a friendly, non-intrusive way
in order to be more acceptable to elderly
environments. The performance of the radar was
evaluated by comparing its values to the values
obtained from the Vicon motion capture system. An
algorithm based on the LS periodogram was proposed
to estimate gait velocity and step duration.
The periodogram has shown its reliability in
extracting the frequency content of the Doppler
irregular signals without the need to specify an exact
sampling frequency The results were compared to the
ground truth data of Vicon and demonstrated the
reliability of this algorithm in estimating the gait
velocity. In future work, more results need to be
validated on a larger group consisting of elderlies and
a new method should be developed that could
calculate relevant gait parameters without previous
knowledge of the distance travelled.
Figure 5: FFT of a radar signal showing the number of
steps (9.316 steps).
Figure 6: Mean gait velocity estimates obtained from Vicon
(red) and the radar (blue).
ACKNOWLEDGEMENTS
This work was supported by the regional council and
the European Regional Development Fund (FEDER).
REFERENCES
Chłoń-Domińczak, et al, 2014. Population ageing in
Europe: facts, implications and policies, 2014.
Fried, et al. 2001. "Frailty in Older Adults: Evidence for a
Phenotype" J Ger A Bio Sci Med Sci, vol.56 pp. 146-
157.
Theou, et al 2011. "A comparison of the relationship of 14
performance-based measures with frailty in older
women," Appl. Phy. Nutr. Metab., vol. 36, pp. 928-938.
Jaber et al., 2014. "A New Device to Assess Gait Velocity
at Home," in IFMBE proceedings.
Yardibi, et al. 2011. "Gait Characterization via Pulse-
Doppler Radar," in IEEE International Conference on
Quantitative Gait Measurement with Wave Doppler-radar for Elderly Walking Speed Recognition
413
Pervasive Computing and Communications Workshops
(PERCOM Workshops), Seattle, WA, USA.
Scargle, et al. 1982. "Studies in astronomical time series
analysis. II - Statistical aspects of spectral analysis of
unevenly spaced data". ApJ 1:263 pp. 835-853.
VanderPlas, 2018. "Understanding the Lomb–Scargle
Periodogram," The Astrophysical Journal Supplement
Series, vol. 236, no. 116, p. 28.
Cuddihy, et al., 2012. "Radar Walking Speed
Measurements of seniors in their Apartments:
Technology for Fall Prevention," in 34th An Int.
Conference of the IEEE EMBS, California USA.
HEALTHINF 2019 - 12th International Conference on Health Informatics
414
