Content uploaded by Van Nguyen
Author content
All content in this area was uploaded by Van Nguyen on Sep 09, 2018
Content may be subject to copyright.
Detection of Motion and Posture Change using an IR-UWB Radar
Van Nguyen, Abdul Q. Javaid, and Mary A. Weitnauer
Abstract— Impulse radio ultra-wide band (IR-UWB) radar
has recently emerged as a promising candidate for non-contact
monitoring of respiration and heart rate. Different studies have
reported various radar based algorithms for estimation of these
physiological parameters. The radar can be placed under a
subject’s mattress as he lays stationary on his back or it can be
attached to the ceiling directly above the subject’s bed. However,
advertent or inadvertent movement on part of the subject and
different postures can affect the radar returned signal and
also the accuracy of the estimated parameters from it. The
detection and analysis of these postural changes can not only
lead to improvement in estimation algorithms but also towards
prevention of bed sores and ulcers in patients who require
periodic posture changes. In this paper, we present an algorithm
that detects and quantifies different types of motion events
using an under-the-mattress IR-UWB radar. The algorithm also
indicates a change in posture after a macro-movement event.
Based on the findings of this paper, we anticipate that IR-UWB
radar can be used for extracting posture related information
in non-clinical enviroments for patients who are bed-ridden.
I. INTRODUCTION
Pressure ulcers or bed-sores are common but preventible
problems in nursing homes especially for physically impaired
elderly patients [1]. These ulcers can be caused in-part by
immobility of the subjects [1]. Laying in one posture for
a long period of time can impede blood flow and decrease
the supply of oxygen and nutrients to the concerned muscles
[1]. The development of pressure ulcers can interfere with
the recovery process and also lead to premature mortality
in some patients. The cost associated with the treatment of
these ulcers is also high and causes significant burden on
annual healthcare expenditure [1].
One solution to prevent these pressure ulcers is to make
sure that patients have changed posture periodically [2].
Many researchers [3]–[6] have studied detection of body
postures with pressure sensor sheets. The pressure applied by
the body as the patient lays on such a sensor sheet creates a
pressure image, which changes with a change in posture. In
these methods, a large number of sensors, hundreds or even
thousands [4], [5], are needed to improve the classification
resolution. Therefore, the amount of data that needs to
be processed in the pressure image analysis is enormous,
leading to high computational cost and excessive delay. Some
studies have proposed methods for posture detection using
wearable accelerometer sensors [7], [8]. Electrocardiogram
sensors have also been used for posture detection as mor-
phology of the QRS-complex in ECG changes with the body
Van Nguyen, Abdul Q. Javaid and Mary A. Weitnauer are with
School of Electrical and Computer Engineering at Georgia Institute of
Technology, Atlanta, Georgia 30332, USA. Email: [vannguyen, aqjavaid,
maweit]@gatech.edu.
posture [9], [10]. However, the drawback associated with
these types of sensors is that they require some contact
with the skin / body in the form of adhesive electrodes or
other attaching mechanisms. These contact sensors may raise
concerns about discomfort and hygienic conditions of the
patient.
Impulse radio ultra-wideband (IR-UWB) radar has
emerged as a promising candidate for non-contact and con-
tinuous monitoring of heart rate (HR) and respiration rate
(RR), the two common forms of vital signs [11]–[14]. An
IR-UWB system transmits a series of extremely low power
and short duration electromagnetic pulses. When these pulses
fall on a moving reflecting surface such as human chest, they
are reflected back to the IR-UWB receiver and contain infor-
mation about the periodic movement of chest wall caused by
breathing and heartbeat. Since chest displacement introduces
time-varying round trip propagation delays in the received
pulses, presence of motion related artifacts caused by sub-
ject movement degrades the signal-to-noise ratio (SNR) of
the received signal and affects the accuracy of estimated
parameters. Thus, the detection and quantification of these
motion corrupted parts is not only important for consistent
and accurate estimation of physiological parameters but this
information can also be leveraged to detect posture changes.
Many researchers have focused on detection of motion and
posture from an IR-UWB radar along with the vital signs
sensing. Khan et al [15] implement a motion detector on
an IR-UWB vital signs monitor based on an autocorrelation
approach. Ota et al [16] compute the changes in the power-
range profile of the IR-UWB received signal over time to
detect various motion types of the subject, such as in-bed
motion (sitting up) and motion that is outside of the bed
but inside the room (wandering and going in and out of
the door). Another approach to detect the type of motion
similar to the latter is presented in [17]. In this approach, a
threshold decision method based on energy detection is used
to detect the human motion that involves moving from one
range cell to another such as walking or running. All these
methods use very high sampling rates for the radar signal
and are also hampered by large storage requirements for the
recorded data.
In this paper we investigate how typical activities of a
patient lying in bed affect the radar received signal of a
IR-UWB vital signs monitor. Specifically we implement a
novel and simple algorithm that detects and quantifies motion
events and also indicates if a macro-movement event has
resulted in a posture change. The presented work serves as a
first step for the development of more sophisticated methods
to classify different types of movements, using IR-UWB
Fig. 1. Experiment set up. The radar was placed under the mattress.
TABLE I
ACTIVITIES
1. Resting 7. Turn onto left side
2. Pick up book* 8. Resting
3. Turn pages 9. Turn onto right side
4. Put down book* 10. Resting
5. Count 11. Turn onto stomach
6. Arm movement & talk 12. Resting
* Some arm movement and talk follow
radar, such as to identify if a period of movement corre-
sponds to limb movement or a turn. The proposed methods
can also be added to heart and respiration rate algorithms to
improve the accuracy of the estimated parameters.
II. MET HODS
A. Protocol
The data for the study were collected from six adult
subjects including 3 males and 3 females (demographics:
age 40.2±13.6 years, BMI 23.1±3.5), under a protocol
approved by Georgia Institute of Technology Institutional
Review Board. Each subject lay still on his or her back for
the first 10-12 minutes and performed a series of various
daily activities including picking up and reading a book,
talking, moving arms, turning to the sides and stomach.
Specifically, the talking was counting from 1 to 20 and the
arms moved round and round perpendicular with the chest.
Table I lists the activities performed by each subject. Note
that the footnote “* Some arm movement and talk follow”
was not part of the protocol but were accidentally added by
one subject.
B. Hardware & Signal Processing
The IR-UWB radar, a 40-inch by 40-inch panel, was
placed under the mattress. A schematic of the system can
be found in [18]. The transmitted pulses were 1 - 4 ns long,
centered at 4.1 GHz. The reflected signal was time-gated and
down-converted to baseband before being hardware-filtered
into the lower respiration band (0.1 - 0.7 Hz) and the higher
heart band (0.6 - 7.2 Hz). Next, the outputs of each filter
band were sampled and quantized with a 12-bit ADC with
sampling rate 128 Hz.
C. Motion and Posture Change Detection
The complete motion and posture change detection al-
gorithm is composed of two parts: (1) detect and quantify
motion events, and (2) detect if a posture change has taken
place following a motion event. These are explained in detail
below:
1) Detection and Quantification of Motion Events: The
radar-based motion detector works on the ADC samples from
each of the heart and respiration bands. To detect motion in
the heart band, for each ADC sample, we set an indicator
function to 1 if the sample amplitude is outside the range
[Lh,Uh], and zero otherwise, where Lhand Uhare operator-
specified lower and upper bounds in the heart signal channel,
respectively. We sum the indicator function values over the
temporal window of length T(this counts the number of
times the ADC value goes out of range) and divide that sum
by the total number of samples in that window. If that ratio,
denoted as ρ, is higher than a threshold ρth, then we set
a heart channel motion flag to one. A similar operation is
done for the respiration band, but we change the bounds to
[Lr,Ur], where Lrand Urare the lower and upper bounds in
the respiration signal channel, respectively, and Lr≤Lhand
Ur≥Uhsince the respiration signal is stronger. The radar-
based motion detector performs the OR operation on the
heart-samples-based motion flag and the respiration-samples-
based motion flag to compute the overall motion flag for the
current temporal window. Then the window slides forward
by ∆≤Tand the whole procedure is repeated to get the next
overall motion flag. The values of all the above mentioned
parameters, used in this work, are shown in Table II.
In order to quantify motion, the out-of-range rates of the
two channels are mapped to a score in a predefined scoring
scale. Assume a motion scoring scale from 0 to S, where a
score of 0 indicates no motion and a score of Sindicates
that motion occurs during most or all of the duration of the
window. One way to compute the motion score in a time
window is through quantization:
score(¯
ρ) = d¯
ρSe,(1)
where ¯
ρis the average of the out-of-range ratios from both
heart and respiration channels in a window.
2) Detection of Postural Change: In order to quantify the
changes in posture, the signal segments before and after a
motion event are analyzed. If there is no motion event in Tw
second period (Tw=60s), then local maxima and minima in
this period are identified. The upper and lower envelopes,
Euand El, are then determined by cubic inteprolation of the
maxima and minima, respectively. The envelope difference
sequence, Ed, is then estimated by subtracting the lower
envelope from the upper envelope, i.e., Ed=Eu−El[19].
The mean envelope difference, Ed, is obtained by averaging
Edover the number of samples in Twsegment. Edis then
chosen as a reference and denoted by αr. Whenever a motion
event is detected, the algorithm checks if there is no motion
TABLE II
MOTI ON DET ECTI ON PARA METE RS AND VAL UES
Parameter Value
Window length (T) 4 s
Window step (∆) 2 s
[Lh,Uh] [1000,3000]
[Lr,Ur] [500,3500]
ρth 0.15
0 50 100 150 200 250
Change metric (.)
0
0.005
0.01
0.015
0.02
0.025
0.03 |H0
|H1
Fig. 2. Histograms of the change metric γconditioned on the null
hypothesis H0and the alternative hypothesis H1. The change metric (γ)
is the difference in Edfor pairs of Twsegments from the data.
event in Twseconds of data that follows the motion event.
Once such a segment is found, Edis calculated for this
new segment and compared with the reference αr. If the
difference between estimated and reference value is greater
than a threshold λ, then posture flag is set to one, indicating
that a posture change has occured. The new value of Ednow
becomes the new reference αr. The value of λis determined
from the data by plotting two histograms shown in Fig. 2.
Specifically, one histogram (H0) is plotted for difference γ
in Edvalues for two non-overlapping Twsegments during
periods without motion of the same posture. The second
histogram (H1) is for the difference γin Edvalues of pairs of
Twsegments without motion, one belongs to one posture and
the other belongs to the next posture. A threshold λcan be
defined as a point of intersection of the two histograms. By
inspection, a value around 45 is appropriate for λ. For our
data, λ=43 works best in terms of maximizing the detection
rate and minimizing the false alarm rate.
III. RES ULTS
A. Motion Detection and Quantification
Fig. 3a-b show the output of the respiration and heart
bands of the system, respectively. The individual motion
detector output for the heart channel and the respiration
channel are indicated also on these plots (the dots at the top
and bottom of each plot). In these plots, the motion flags are
multiplied by 4250 for ease of viewing. The vertical bars
between Fig. 3a and Fig. 3b delimit periods corresponding
to different activities as listed and numbered in Table I.
The overal motion flag is plotted in Fig. 3c. For example,
at t=664 seconds the heart channel flag is set to 1 and
the respiration channel motion flag is set to 0, hence the
overall motion flag is 1, and we say motion is detected in
the time window [t−T,t] or [660 seconds - 664 seconds].
It is worth noting that we made a heuristic selection of the
out-of-range threshold ρth. Let Fsbe the ADC sampling rate,
here Fs=128 Hz, then ρth =0.15 means that a window must
contain at least ρthT Fs=0.6∗Fsor 0.6 seconds of out-of-
range ADC samples to set the motion flag to 1. A typical
body movement typically lasts longer than 0.6 seconds.
This motion detector can be made more or less sensitive
by varying the heart bounds and respiration bounds or by
0
2000
4000
Heart signal
0
2000
4000
Respiration signal
0
1
Motion flag
0
5
10
Motion score
0
1
Posture
(Heart)
0 500 1000 1500 2000 2500
Time (s)
0
1
Posture
(Heart & Resp)
(a)
(b)
(c)
(d)
(e)
(f)
12356910
811 12
7
1.#Resting
2.#Pick#up#book
3.#Turn#pages
4.#Put#book#
down#(in#fact,#
Jiten takes#back#
book)
5.#Count
6.#Arm#move
7.#Left#turn
8.#Resting
9.#Right#turn
10.#Resting
11.#Turn#to#
stomach
12.#Resting
4
Fig. 3. Data of one subject. (a) Heart signal. (b) Respiration signal. (c)
Motion flag. (d) Motion score. (e) Posture-change flag from heart channel
only. (f) Posture-change flag from both heart and respiration channels.
changing the out-of-range threshold. We observe that the
radar-based motion flags agree well with the subject’s time-
stamped series of movements.
The motion score is shown in Fig. 3d. In order to quantify
different types of motion events, we considered two move-
ment events, ‘Pick up a book’ and ‘Put down the book’, and
two macro-movement events resulting in a posture change.
The other activities are located very close to each other in
time and hence are not included in the analysis. The motion
score values were averaged over the duration of the activities.
The mean and standard deviation for all subjects for the
above mentioned activities is shown in Fig. 4. It can be
observed that motion events that correspond to a posture
change have high motion score.
B. Detection of Posture Change from the Radar Signal
The results for detection of posture change in one subject
are shown in Fig. 3e-f. Since the radar records data in
both the respiration and heart band, posture change flag was
calculated with heart channel alone in Fig. 3e. The algorithm
accurately detects the three posture changes for this subject.
Fig. 3f shows value for posture flag if information from
respiration channel is also combined with the heart channel.
Specifically, the posture change flag was calculated from both
channels and an OR operation was performed between them.
All the posture changes are also correctly detected.
The results for posture change for all subjects are summa-
rized in Table III. A Xindicates a correct detection while ×
represents a missed detection. The number of false alarms
is included in the last column. The red symbols or numbers
correspond to when using the heart channel data only. The
black symbols or numbers correspond to when using both
heart and respiration channels. When using the heart channel
data only, the algorithm was able to detect correct posture
change 77.8% of the time for all subjects, with a false
* *
Pick
up
book
Put
down
book
Left
to
right
Right
to
Stomach
Motion score
Fig. 4. Motion scores for all subjects. The blue bars represent motions not
resulting into a posture change while the red bars are for those that cause a
change in posture. (*) indicates that p<0.05 for the posture changing
motion scores as compared to those that were caused by low intensity
movements.
TABLE III
POS TURE DE TECT ION RES ULTS
Subject No. B→L L →R R →S FA
1XXXXXX4, 5
2XX×××X0, 1
3×XXXXX1, 1
4XXXXXX0, 0
5XX××XX1, 1
6XXXXXX0, 0
B: Back, L: Left, R: Right, S: Stomach, FA: False alarm
alarm rate of 20%. When both channels are used in posture
detection, the detection rate is increased to 88.9% but the
false alarm rate is also increased to 26.7%.
IV. DISCUSSION
The results derived in this paper indicate that IR-UWB can
be used to detect motion and postural changes. The motion
detection algorithm operates on small time windows (T=4s)
and the time sensivity of the detector can be changed by
changing the window size. On the other hand, the posture
change is estimated from a window (Tw) of 60-seconds
duration. The larger window size was chosen to decrease
false alarm rate. Moreover, patients in nursing homes, who
require posture change after periodic intervals, normally stay
in a particular posture for a longer time.
As indicated by the results, both channels can accurately
detect motion. Combining posture change information from
both signals improves the detection probability but degrades
the false alarm rate.
A limitation of this work is the extremely small sample
size of 6 subjects. Future work should include data from
patients of all ages and also from sick and bed-ridden patients
in nursing homes. The algorithms presented in this paper
can be augmented with other parameters related to heart
and respiration rate to provide for more accurate detection
of micro- and macro-movements. Future work should also
focus on incorporation of features other than those based on
amplitude of the signal.
V. CONCLUSION
In this paper, we present an algorithm that detects different
types of motion events from an IR-UWB radar signal. The
algorithm assigns a score value to each motion event and
also utilizes signal statistics to determine if a motion event
has caused a posture change. The accuracy of posture change
determined by the algorithm is 88.9%.
ACKNOWLEDGMENT
The authors would like to thank Sensiotec Inc. for provid-
ing the experimental data.
REFERENCES
[1] M. Reddy, S. S. Gill, and P. A. Rochon, “Preventing pressure ulcers:
a systematic review,” Jama, vol. 296, no. 8, pp. 974–984, 2006.
[2] D. Bluestein and A. Javaheri, “Pressure ulcers: prevention, evaluation,
and management.” American family physician, vol. 78, no. 10, 2008.
[3] J. J. Liu, W. Xu, M.-C. Huang, N. Alshurafa, M. Sarrafzadeh, N. Raut,
and B. Yadegar, “Sleep posture analysis using a dense pressure
sensitive bedsheet,” Pervasive and Mobile Computing, vol. 10, pp.
34–50, 2014.
[4] W.-Y. Chang, C.-C. Chen, C.-C. Chang, and C.-L. Yang, “An enhanced
sensing application based on a flexible projected capacitive-sensing
mattress,” Sensors, vol. 14, no. 4, pp. 6922–6937, 2014.
[5] R. Yousefi, S. Ostadabbas, M. Faezipour, M. Farshbaf, M. Nourani,
L. Tamil, and M. Pompeo, “Bed posture classification for pressure
ulcer prevention,” in IEEE Int. Conf. Eng. Med. Bio., 2011.
[6] C. Hsia, K. Liou, A. Aung, V. Foo, W. Huang, and J. Biswas,
“Analysis and comparison of sleeping posture classification methods
using pressure sensitive bed system,” in IEEE Int. Conf. Eng. Med.
Bio., 2009, pp. 6131–6134.
[7] H. Yoon, S. Hwang, D. Jung, S. Choi, K. Joo, J. Choi, Y. Lee, D.-U.
Jeong, and K. Park, “Estimation of sleep posture using a patch-type
accelerometer based device,” in IEEE Int. Conf. Eng. Med. Bio. IEEE,
2015, pp. 4942–4945.
[8] K.-M. Chang and S.-H. Liu, “Wireless portable electrocardiogram
and a tri-axis accelerometer implementation and application on sleep
activity monitoring,” Telemedicine and e-Health, vol. 17, no. 3, pp.
177–184, 2011.
[9] T. Pawar, N. Anantakrishnan, S. Chaudhuri, and S. P. Duttagupta,
“Transition detection in body movement activities for wearable ecg,”
IEEE Trans. Biomed. Eng., vol. 54, no. 6, pp. 1149–1152, 2007.
[10] H. J. Lee, S. H. Hwang, S. M. Lee, Y. G. Lim, and K. S. Park, “Estima-
tion of body postures on bed using unconstrained ecg measurements,”
Biomedical and Health Informatics, IEEE Journal of, vol. 17, no. 6,
pp. 985–993, 2013.
[11] E. M. Staderini, “Uwb radars in medicine,” Aerospace and Electronic
Systems Magazine, IEEE, vol. 17, no. 1, pp. 13–18, 2002.
[12] S. Venkatesh, C. R. Anderson, N. V. Rivera, and R. M. Buehrer,
“Implementation and analysis of respiration-rate estimation using
impulse-based uwb,” in Military Communications Conference, 2005.
MILCOM 2005. IEEE. IEEE, 2005, pp. 3314–3320.
[13] J. C. Y. Lai, Y. Xu, E. Gunawan, E. C.-P. Chua, A. Maskooki, Y. L.
Guan, K.-S. Low, C. B. Soh, and C.-L. Poh, “Wireless sensing of
human respiratory parameters by low-power ultrawideband impulse
radio radar,” Instrumentation and Measurement, IEEE Transactions
on, vol. 60, no. 3, pp. 928–938, 2011.
[14] V. Nguyen, A. Q. Javaid, and M. A. Weitnauer, “Harmonic path (hapa)
algorithm for non-contact vital signs monitoring with ir-uwb radar,”
in IEEE Biomed. Cir. Sys. Conf. (BioCAS), 2013, pp. 146–149.
[15] F. Khan, J. W. Choi, and S. H. Cho, “Vital sign monitoring of a non-
stationary human through IR-UWB radar,” in IEEE Int. Conf. Net.
Infra. Dig. Cont., 2014, 2014, pp. 511–514.
[16] K. Ota, Y. Ota, M. Otsu, and A. Kajiwara, “Elderly-care motion sensor
using uwb-ir,” in Sensors Applications Symposium (SAS), 2011 IEEE.
IEEE, 2011, pp. 159–162.
[17] X. Li, D. Qiao, and Y. Li, “Macro-motion detection using ultra-
wideband impulse radar,” in IEEE Int. Conf. Eng. Med. Bio., 2014.
[18] S. Foo, “Ultra wideband monitoring systems and antennas,” Apr. 23
2013, uS Patent 8,428,696.
[19] A. Q. Javaid, C. M. Noble, R. Rosenberg, and M. A. Weitnauer,
“Towards sleep apnea screening with an under-the-mattress ir-uwb
radar using machine learning,” in IEEE 14th Int. Conf. Machine
Learning and Applications, Dec 2015, pp. 837–842.
