A Falls Risk Screening Tool Based on Millimetre-Wave Radar

Elif Dogu

1 a

, Jose A. Paredes

1,2 b

, Akram Alomainy

1 c

, Janelle M. Jones

3 d

and

Khalid Z. Rajab

1 e

School of Electronic Engineering and Computer Science, Queen Mary University of London, London E1 4NS, U.K.

School of Arts, Humanities and Social Sciences, University of Roehampton, London SW15 5PU, U.K.

School of Biological and Behavioural Sciences, Queen Mary University of London, London E1 4NS, U.K.

Keywords:

FMCW, mm-Wave Radar, Health Informatics, Medical Decision Support, Fall Risk Assessment,

Timed Up and Go, Gait, Mobility.

Abstract:

Falls among the older adults pose a global health concern, necessitating innovative approaches for timely and

effective falls risk screening. Aiming to develop a real-time falls risk screening tool, this study explores the

integration of millimeter-wave (mmWave) radar technology with the Timed Up and Go (TUG) test, which

is a widely used screening tool that combines parameters measuring a person’s dynamic balance and func-

tional mobility. Radar technology has emerged as a promising tool for non-intrusive, continuous monitoring

of movements – including gait patterns and mobility – in real-life scenarios. By leveraging Frequency Mod-

ulated Continuous Wave (FMCW) radar, the study assesses its performance against video recordings in TUG

completion time measurement. The completion time, conventionally measured manually with a timer in clini-

cal settings, was derived from radar measurements using two different methods based on distance, and micro-

Doppler (i.e. velocity). Results indicate radar’s superior accuracy in distance-based measures with 3.48%

error and a correlation of 0.9996, surpassing manual timing (4.26% error, 0.9960 correlation) and demonstrat-

ing viability for falls risk screening protocols. The velocity-based determination performed slightly poorer

(6.49% error, 0.9936 correlation), which is attributable to the very high sensitivity of the radar in detecting

small motions, such as shufﬂing in a chair, that are not a part of the TUG sequence. This study contributes to

healthcare technology innovation, emphasising radar’s transformative role beyond falls risk assessment. The

precision of radar-based measurements opens avenues for enhanced diagnostics, monitoring, and personalised

care.

1 INTRODUCTION

A fall is deﬁned as an event which results in a per-

son coming to rest inadvertently on the ground or

ﬂoor or other lower level (WHO, 2021). Every year,

millions of older adults, and their families face the

life-altering consequences of falls. The World Health

Organization (WHO) reports that falls represent the

second leading cause of unintentional injury deaths

worldwide (WHO, 2021). These incidents not only

have immediate critical impacts such as injury, pain,

disability, and increased mortality (NICE, 2013), but

they also cast a long shadow of psychological distress

https://orcid.org/0000-0003-4883-3450

https://orcid.org/0000-0002-0412-0179

https://orcid.org/0000-0002-5802-9194

https://orcid.org/0000-0002-0910-2404

https://orcid.org/0000-0003-1337-2965

leading to fear of falling (Laved

an et al., 2018), re-

duced daily activity and reduced self-conﬁdence in

mobility.

The American Geriatrics Society and the British

Geriatrics Society recommend that all adults over the

age of 65 should be screened for falls risk at least once

a year (Moncada, 2011). However, effective fall risk

screening is still underutilised and not routinely inte-

grated into clinical practice (Sun and Sosnoff, 2018).

There might be several reasons such as constraints of

clinical time and environment.

From a broader perspective, another limitation of

current falls risk screening tools is that they only pro-

vide a snapshot of the patient’s condition at their best,

under clinical conditions. The tests are carried out

during the working hours of clinicians, when patients

are less likely to be affected by factors like hunger,

thirst, or fatigue. Meaningful insights into how the

Dogu, E., Paredes, J., Alomainy, A., Jones, J. and Rajab, K.

A Falls Risk Screening Tool Based on Millimetre-Wave Radar.

DOI: 10.5220/0012618900003699

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 10th International Conference on Information and Communication Technologies for Ageing Well and e-Health (ICT4AWE 2024), pages 161-168

ISBN: 978-989-758-700-9; ISSN: 2184-4984

161

patient’s gait pattern may vary at different times of

the day, such as when they wake up in the morn-

ing or when they get up at night, remain understud-

ied.In light of the weak association between natural

gait speed and in-laboratory gait speed (Takayanagi

et al., 2019), integrating technology with established

practices can offer a promising avenue for real-time

and continuous monitoring beyond clinical conﬁnes,

in the diverse and dynamic scenarios of everyday life.

In the present research, a millimetre-wave

(mmWave) radar was chosen to integrate with the

TUG test, to develop a falls risk screening tool, as a

digital health technology tool (DHTT) (Taylor et al.,

2020).

Radar technology is a promising solution for am-

bient sensing, as it is minimally intrusive: it does not

rely on wearables, nor does it capture camera images

or video. It can provide real-time remote monitoring

of individuals in a home, allowing for the continu-

ous evaluation of gait (Gambi et al., 2020; Alanazi

et al., 2022), mobility patterns, and overall wellbeing

(Cardillo et al., 2022). FMCW (Frequency Modulated

Continuous Wave) radar captures subtle movements

with high precision, making it an effective tool for

classifying daily-life activities such as walking, sit-

ting, standing and sleeping (Rajab et al., 2021; Yu

et al., 2022). Radars also have the capability to op-

erate through various materials and work in challeng-

ing environmental conditions such as adverse weather

and low visibility. In homes, the radars provide effec-

tive continuous monitoring which enables early de-

tection of deviations from normal mobility patterns.

In this work, we observe the feasibility of integrat-

ing radar technology with the TUG test and assess

its performance compared to video camera record-

ings. For this purpose, the paper is organised as fol-

lows: in section 2, previous studies implementing new

technologies to the TUG test are summarised. Sec-

tion 3 provides information on the procedures fol-

lowed while designing the experiments. Section 4

presents the results of the experiments and their analy-

ses. Discussions of the outcomes are provided in Sec-

tion 5 and the paper is concluded in Section 6.

2 RELATED WORK

The TUG test is widely used in current clinical prac-

tice to detect gait and balance impairment. To conduct

a TUG test, patients wear their regular shoes, and may

use a walking aid if necessary. The setup involves a

standard stable chair and a three-meter line in front

of it. The patient is instructed as: “When I say ‘Go’,

I want you to stand up from the chair, walk to the

line on the ﬂoor at a normal pace, turn around, walk

back to the chair at a normal pace and sit down again.”

The timing starts on the word ‘Go’ and ends when the

patient sits back down. The TUG completion time

is measured by the clinician manually, using a timer.

If a patient’s TUG test takes more than 15s, they are

identiﬁed as being at risk of falling. Therefore, by

deﬁnition, the TUG test involves 4 parameters: sit-to-

stand time, 3m walking time (where the gait speed can

be inferred), a 180

◦

turn time, and stand-to-sit time.

The TUG test is usually preferred for its simplicity. It

has shown to be predictive in lower functioning adults

(Beck Jepsen et al., 2022).

The TUG test recently attracted signiﬁcant atten-

tion from researchers. New technologies have been

developed to instrument the TUG test in several ways,

including wearable sensors, cameras, and ambient

sensors.

Greene, Doheny, O’Halloran and Kenny used the

TUG test with shank-mounted inertial sensors and

developed a regression-based method for the assess-

ment of frailty. The experiments were conducted with

399 community-dwelling older adults (Greene et al.,

2014). Using inertial sensor data obtained during the

TUG test, the participants were classiﬁed as frail or

non-frail with mean accuracy of 75.20% (stratiﬁed

by gender). Spina et al. investigated the parame-

ter of 180-degree turns derived from an instrumented

TUG test in stroke patients (Spina et al., 2022). They

placed a single inertial measurement unit (IMU) sen-

sor on the lower back of patients. 48 chronic stroke

patients and 23 healthy controls were included in the

experiment. They reported turning speeds as accu-

rate measures of mobility, capable of discriminat-

ing stroke patients with normal and impaired mo-

bility. Fudickar, Hellmers, Lau, Diekmann, Bauer,

and Hein introduced an unsupervised screening sys-

tem for older adults and evaluated its validity for the

TUG and Five Times Sit-to-Stand test (Fudickar et al.,

2020). The system involved both wearable and ambi-

ent movement sensors. The sensor data sets of 91 par-

ticipants aged 73 to 89 years was compared to conven-

tional measurement with stopwatches. With ambient

sensor data, signiﬁcant correlations of 0.89 and 0.73

were detected for TUG and sit-to-stand, respectively.

With wearable sensors, they were reported as 0.78 and

0.87.

Choi, Parker, Knarr, Gwon and Youn proposed

a model that predicts the TUG test using three-

dimensional acceleration data collected from wear-

able sensors during normal walking (Choi et al.,

2021). They recruited 37 older adults with an average

age of 69.6 and used seven IMU-based wearable sen-

sors for the experiments. They obtained better predic-

ICT4AWE 2024 - 10th International Conference on Information and Communication Technologies for Ageing Well and e-Health

162

tion accuracy with two foot sensors. However, they

recommended the use of a single IMU sensor at the

pelvis for greater comfort. Diao et al. developed an

environment-adaptive TUG test with wearable iner-

tial sensors attached to the two shanks (Diao et al.,

2021). They conducted the experiment with 103 older

adults, reporting an average accuracy 9.52% higher

than the conventional TUG in classiﬁcation of high-

risk and low-risk groups for falls. Dierick, Stoffel,

Sch

utz, and Buisseret proposed placing a single wear-

able IMU sensor on the lower back during the TUG

test to enhance the predictive accuracy (Dierick et al.,

2022). They recruited 73 nursing home residents for

the experiments. They reported 74% accuracy, with a

speciﬁcity of 95.9% and a sensitivity of 29.2% using

the sensor, while the conventional TUG was 64% ac-

curate. Kataoka et al. observed lower-limb kinemat-

ics of patients with Locomotive Syndrome (LS), using

a TUG test with wearable gait sensors (Kataoka et al.,

2023). They recruited 140 patients with an average

age of 72.6 in Japan for the experiments. Their sen-

sors consisted of tri-axial acceleration and gyro sen-

sors that were placed on seven lower-limb body seg-

ments. Their results showed that the subjects with LS

have longer TUG completion time than the subjects

with non-LS.

Wearable devices in assessing gait and mobility

with the TUG test have yielded promising results. The

main concern in these studies was the comfort associ-

ated with wearing the sensors. Conﬁgurations involv-

ing wires or tapes on the body, while suitable for clin-

ical or research environments, may pose challenges

in everyday life due to their intrusive nature. More-

over, their practicality is constrained by the factor of

limited battery life.

Savoie, Cameron, Kaye, and Scheme proposed

the automation of the TUG test using a single con-

ventional video camera (Savoie et al., 2020). They

recorded 30 healthy participants with both Microsoft

Kinect V2 and a standard video camera while per-

forming two versions of the TUG test with 3-meter

and 1.5-meter walking distances. They developed a

video-based vTUG system leveraging advanced com-

puter vision techniques. vTUG system yielded the

same error as the standard Kinect-based system for

all six key transitions points, and average errors of

less than 0.15 seconds from a multi-observer hand-

labelled ground truth. Li et al. developed a video-

based activity classiﬁcation method to infer the TUG

parameters of Parkinson’s disease patients (Li et al.,

2018). 24 patients were recruited, and their TUG

test videos were recorded in semi-controlled environ-

ments having different backgrounds. They reported

more than 90% on average for the classiﬁcation of

frames.

Cameras, coupled with computer vision technol-

ogy, have shown good performance with, however

their use for continuous monitoring can be limited due

to privacy concerns.

Frenken, Brell, G

overcin, Wegel, and Hein used

a light barrier, four force sensors, and a laser range

scanner built into a single apparatus and proposed the

ambient TUG (aTUG) test for gait and balance anal-

ysis (Frenken et al., 2013). The experiment was con-

ducted with ﬁve older patients in a residential care

facility, demonstrating that aTUG could reliably and

precisely measure total duration of TUG and dura-

tions of the single components with a mean error of

only 0.05 seconds and mean standard deviation of

0.59 s. Ayena, Chioukh, Otis, and Deslandes com-

bined the performance of ultra-wideband radar and in-

strumented insole for an unobtrusive TUG test (Ayena

et al., 2021). They conducted 14 tests with a sin-

gle participant and the results reported the feasibility

of the TUG test using a sensor combination. Soubra

Mourad-Chehade, and Chkeir proposed the automa-

tion of TUG test using Doppler radar system (Soubra

et al., 2023). The radar was set into the backrest of the

chair used for TUG test. They recruited 26 healthy

participants, aged between 22 and 60. The partici-

pants performed 3 slow, 3 normal, 3 fast TUG tests as

experiments. An infrared camera system was used as

the ground truth. The Doppler radar system achieved

4.8% error on the TUG completion time.

Ambient sensors have the potential to deal with

the limitations of wearable sensors and cameras.

However, the generalisability of previous ﬁndings

may be constrained by small sample sizes. There is

a need for the radar to be validated for TUG test, and

other falls screening tools to disseminate their use in

various settings.

When all the risk factors are detected, the third

step of designing the targeted intervention and con-

trol measures begins. This step may include exercise

and education programs, modiﬁcations in home envi-

ronment, revision of medications and other actions to

eliminate the risk factors one by one from patients’

lives.

3 SYSTEM OVERVIEW AND

SETUP

Our standard TUG test setup includes a chair with a

3m line as the walking lane, as depicted in Fig. 1.

During the experiments, the radar device is placed

at two different locations to observe the variations. In

both positions, the height of the radar was 0.5m. In

A Falls Risk Screening Tool Based on Millimetre-Wave Radar

163

Figure 1: Two positions of the radar for the TUG test.

position 1, the radar was placed 1m behind the chair

and in position 2, 4m in front of the chair.

3.1 Radar Data Acquisition

A millimeter-wave (mmWave) radar system emits

high-frequency chirp-like signals within the GHz

range. Certain conﬁgurations incorporate multiple

receive (RX) and transmit (TX) antennas, constitut-

ing Multiple-Input Multiple-Output (MIMO) radars.

These systems present the capability to extract com-

prehensive information about range, angle (azimuth,

elevation) and radial velocity.

The Doppler frequency, which is the difference

between emitted and received signals, is measured by

the radar as f

– a.k.a. intermediate frequency –, and

the distance r to the object can be expressed as:

r =

, (1)

where c represents the speed of light, T signiﬁes the

emission period, and B denotes the signal bandwidth.

For a more comprehensive exposition, please consult

(Richards, 2014).

The emission of several signals in a row, what is

known as a frame, enables the precise measurement

of radial velocities. By emitting a sequence of signals

equally spaced in time by T

, a resultant phase shift

∆ϕ appears. Then, radial velocity can be computed

by:

v =

4πT

∆ϕ, (2)

where λ is the wavelength of the original emitted sig-

nal.

In this work, a Fast Fourier Transform (FFT) is

applied over each emitting-receiving antenna pair, in

order to obtain the distance r from equation (1). And

a Capon beamforming algorithm is used to determine

the velocity from equation (2).

The radar utilised in this work was manufac-

tured by NodeNs Medical Ltd (NodeNs Medical

Ltd., 2023) and is based on the Texas Instruments

IWR6843 chipset. It operates within the unlicensed

60 GHz band. For detailed speciﬁcations, its conﬁgu-

ration is outlined in Table 1.

Table 1: Radar conﬁguration.

Start Frequency (GHz) 60.6

Slope (MHz/µs) 54.725

Samples per chirp 96

Chirps per frame 288

Frame duration (ms) 50

Sampling Rate (Msps) 2.950

Bandwidth (MHz) 2249

Range Resolution (m) 0.084

Velocity resolution (m/s) 0.17

Number of Rx antennas 4

Number of Tx antennas 3

4 EXPERIMENTS AND RESULTS

4.1 Data Collection Protocols

Two participants are recruited to repeat the TUG test

at three different speeds (normal, fast, slow) for two

different positions of the sensor (Positions 1 and 2),

which resulted in 12 experiments in total (see Ta-

ble 2). The TUG test procedure explained in the

section 2 is applied. The TUG completion time is

obtained manually using a timer, as this is the clini-

cal standard (Time-Manual column of Table 2). The

experiments were recorded on camera to obtain the

ground truth, which was deﬁned as the TUG comple-

tion time gathered from the video (Time-Video col-

umn of Table 2). The ground truth was calculated by

determining the total number of frames captured dur-

ing the movement, and the video frame rate was set

at 30 frames per second (fps). To assess the radar

implementation to the TUG test, the completion time

is also inferred from the radar signals with two dif-

ferent methods, using velocity (micro-Doppler signa-

tures) and distance measures.

4.2 Distance-Based Proposal Analysis

The distance graph presents data on the temporal vari-

ation of the distance between the subject and the

ICT4AWE 2024 - 10th International Conference on Information and Communication Technologies for Ageing Well and e-Health

164

Table 2: Experiments. ’Time-Video*’ shows the ground truth of the TUG completion time gathered from video recordings,

’Time-Manual’ shows the version measured manually as in clinical practice, ’Time-Doppler’ shows the results of velocity-

based analysis and ’Time-Distance’ shows the results of distance-based analysis.

Experiment Participant Position TUG Test Time-Video* (s) Time-Manual (s) Time-Doppler (s) Time-Distance (s)

1 1 2 Normal 11.6 11 11.88 11.83

2 1 2 Fast 8.6 8 9.68 9.08

3 1 2 Slow 17.5 18 18.48 17.88

4 2 2 Normal 10.4 11 10.56 10.73

5 2 2 Fast 6.6 7 6.60 6.88

6 2 2 Slow 14.5 15 14.96 14.85

7 1 1 Normal 9.6 9 11.88 9.90

8 1 1 Fast 7.5 7 8.36 7.98

9 1 1 Slow 22.3 22 23.32 22.00

10 2 1 Normal 10.4 10 11.00 10.73

11 2 1 Fast 7.9 8 8.36 8.53

12 2 1 Slow 21.3 21 21.56 21.18

radar. For example, Figure 2 provides the distance

graphs of the experiments 4,5 and 6. These are nor-

mal, fast and slow versions of the TUG test with the

radar placed at the front. During these tests, the dis-

tance initially decreases as the subject walks towards

the sensor, and then increases when the subject re-

turns to the chair. The variations of the tests can be

observed from the kurtosis of their curves. While the

fast TUG test has a more pointed structure, the slow

TUG test has the widest curve.

Figure 2: Distance graphs of experiments 4 (normal), 5

(fast) and 6 (slow).

When the radar position changes, it is expected

that the curve gets mirrored for the distance measure.

Figure 3 depicts the distance graphs of experiments 6

and 12, which are slow TUG tests from different po-

sitions. The slow version of the tests are selected here

as examples, for a better distinction of boundaries.

In both distance plots, instead of a perfectly smooth

curve, small curve variations are observed at the be-

ginning and end. These represent the movements of

getting up and sitting down at the beginning and end

of the TUG test. This is attributable to the inclina-

tion of the body leaning forward when rising from the

chair, and conversely, a backward lean when seated.

These movements change the distance to the sensor,

but more slowly than the walking movement. Hence,

different slopes are observed at the beginning and end

of the curve.

Figure 3: Distance graphs of experiments 6 (top) and 12

(bottom), corresponding to positions 2 (radar in front) and 1

(radar behind) respectively.

The TUG completion time is inferred from the dis-

tance parameter by determining start and end points

of the curve, as depicted with dashed lines in Fig-

ure 3. The results for all experiments are provided

in the Time-Distance column of Table 2.

4.3 Velocity-Based Proposal Analysis

The micro-Doppler effect captures the subtle move-

ments and rotations of body segments, providing dis-

tinct signatures for different activities. Figure 4 de-

picts the micro-Doppler signatures of the experiments

6 and 12, with the axes of velocity (m/s) vs time. It

is observed that the TUG test results in a characteris-

tic S-shaped micro-Doppler curve, which is due to the

subject walking away from the radar (positive veloc-

ities), turning (approximately zero velocity) and then

walking back to the radar (negative velocities). The

S-shape can be reversed based on the position of the

sensor. The action of sitting and standing also result in

A Falls Risk Screening Tool Based on Millimetre-Wave Radar

165

Doppler responses. Similarly to the distance graphs,

the test boundaries can be easily detected. We note

that the micro-Doppler signatures are more prominent

when the radar is in front of the chair than when it

is behind (experiment 6 compared to experiment 12),

which is likely due to the chair obstructing the radar

signal. This can be mitigated against by increasing

the height of the radar. Nevertheless, test boundary

timings can still clearly be detected.

Figure 4: Micro-Doppler signatures of experiments 6

and 12.

Using the micro-Doppler signature, the time be-

tween the start and end of the movement is calculated.

The results for all experiments are provided in Time-

Doppler column of Table 2.

4.4 Performance Evaluation

The TUG completion time gathered from the video

recording is selected as the ground truth, as the video

provides the most accurate timings through frame-by-

frame analysis to identify precise activity time seg-

ments. The TUG completion times measured man-

ually and inferred from radar data with two differ-

ent methods are compared to the ground truth. For

the evaluation, root mean square error (RMSE), mean

absolute error (MAE), mean absolute percent error

(MAPE) and correlation coefﬁcient are calculated.

Table 3 provides the performance evaluation mea-

sures.

Current clinical practice uses manual timing of

Table 3: Performance evaluation measures.

Pe.Measure Manual Micro-Doppler Distance

RMSE 0.4743 0.9156 0.3695

MAE 0.4500 0.7033 0.3479

MAPE 4.26% 6.49% 3.48%

Correlation 0.9960 0.9936 0.9996

the TUG test. It is important to recognise that man-

ual timing does not guarantee 100% accuracy, and

so we use camera recordings to measure our ground

truth. Nonetheless, consultations with healthcare pro-

fessionals have indicated a permissible margin of er-

ror of 10% in timing, and clinicians have expressed a

preference for manual timing due to its simplicity and

cost-effectiveness.

Based on the outcomes, the TUG completion time

deduced from the distance measurement of radar sig-

nals demonstrated superior performance compared to

manual timing. The TUG completion time inferred

from the micro-Doppler signature exhibited slightly

poorer performance. Nevertheless, it is noteworthy

that all measures yielded an error of less than 10%.

5 DISCUSSIONS

There might be several underlying reasons for the dif-

ference in performance of the two radar measures. To

observe the inﬂuence of the sensor’s position on the

results, the differences from the ground truth (the er-

rors) are calculated for manual, micro-Doppler and

distance timings. The mean of the errors are com-

pared for two different positions of the radar. There

was no signiﬁcant difference between the mean errors

from the two positions for any of the methods. There-

fore, the superior performance of distance-based anal-

ysis over velocity-based analysis is not contingent on

the sensor’s position. From closer inspection of the

results in Table 2 it is evident that the velocity-based

(Time-Doppler) measurements consistently overesti-

mate the duration of the TUG sequence. This can be

explained by the sensitivity of the radar in detecting

small movements, such as moving around in a chair,

which might occur just before the motion of stand-

ing up. The inclusion of these additional movements

therefore increases the detected sequence duration.

While the TUG completion times inferred from

the micro-Doppler signature exhibit less accuracy

than the ones from the distance measure, the perfor-

mance of both methods remains acceptable to health-

care professionals. The distance measure yielded an

even better performance than the manual timing. The

outcomes of the ﬁrst experiments are promising for

the implementation of radar as a falls risk screen-

ICT4AWE 2024 - 10th International Conference on Information and Communication Technologies for Ageing Well and e-Health

166

ing tool. Furthermore, the sensitivity of the micro-

Doppler measurements will be combined, in a future

study, with the distance measurements for automatic

boundary detection, to more precisely determine the

TUG sequence activity boundaries and duration.

Despite the promising outcomes, certain limita-

tions should be acknowledged that provide insights

into the boundaries of our study and contribute to a

nuanced understanding of the ﬁndings. The partici-

pants of this study are not older adults and they have

no known gait or balance problems. The inherent dif-

ferences in walking patterns between young individ-

uals and older adults with gait and balance problems

may introduce variations not explicitly addressed in

this study.

The experiments were performed in a cluttered en-

vironment – an ofﬁce space with multiple desks and

moving occupants – which shows the robustness of

the technique to environmental noise and movements.

Nevertheless, the TUG test’s actions were controlled,

which would not reﬂect the complexities of natural

movements during daily activities. Additionally, the

sample size in our study was limited to only two par-

ticipants, which may constrain the broader applicabil-

ity of our results. A larger and more diverse partici-

pant pool could provide a better understanding of the

effectiveness and reliability of the radar-based timing

of TUG test across various demographic groups and

conditions.

6 CONCLUSIONS

This study was designed to observe the innovative

application of radar technology in conjunction with

the TUG test as a means of falls risk screening. To

develop a real-time falls risk screening tool, we ex-

plored the ways to automate the process of capturing

the TUG completion time through non-intrusive am-

bient sensing.

The TUG completion time inferred from the dis-

tance measurements of radar has achieved a level of

accuracy surpassing that of manual timing. Although

micro-Doppler timings were less accurate than dis-

tance, due to the radar’s high sensitivity in detecting

small movements, both techniques had sufﬁcient per-

formance to satisfy the requirements set by healthcare

professionals. This suggests a viable and effective in-

corporation of radar technology into falls risk screen-

ing protocols.

The results of this study not only underscore the

feasibility of integrating radar technology into falls

risk screening but also highlights the importance of

continuous innovation in healthcare technology. The

superior performance of radar-based measurements

hints at the transformative role this technology can

play in healthcare. The implementation of radar as

a falls risk screening tool represents just one facet of

its potential application. The ability to capture move-

ments and timings with precision, without interfering

with privacy, opens avenues for enhanced diagnostics,

monitoring, and personalised care.

A noteworthy ﬁnding emerged from this study

by examining the impact of environmental clutter

on measurement outcomes. The over-estimation of

TUG sequence duration by the velocity-based (micro-

Doppler) technique, due to the high sensitivity of the

radar, highlights a potential area for reﬁnement. Fu-

ture research endeavours could expand its assessment

of radar placement, or encompass use of multiple

radars to measure gait and posture. Another possi-

bility is to expand the participant pool to include in-

dividuals with varying levels of mobility and balance

challenges. Finally, future research could also explore

a more generalised model of subjects’ activities than

the TUG test, so that a radar-based falls risk assess-

ment may more adaptively ﬁt continuous measure-

ment during day-to-day living.

ACKNOWLEDGEMENTS

This work was supported by the UK Engineering and

Physical Sciences Research Council (EPSRC) grant

EP/W03199X/1, “Smart sensors for a wearable-free

and contactless virtual ward at home”.

REFERENCES

Alanazi, M., Alhazmi, A., Alsattam, O., Gnau, K., Brown,

M., Thiel, S., Jackson, K., and Chodavarapu, V.

(2022). Towards a low-cost solution for gait analysis

using millimeter wave sensor and machine learning.

Sensors, 22(5470).

Ayena, J., Chioukh, L., Otis, M., and Deslandes, D. (2021).

Risk of Falling in a Timed Up and Go Test Using an

UWB Radar and an Instrumented Insole. Sensors,

21(3).

Beck Jepsen, D., Robinson, K., Ogliari, G., Montero-

Odasso, M., Kamkar, N., Ryg, J., Freiberger, E., and

Masud, T. (2022). Predicting falls in older adults:

an umbrella review of instruments assessing gait,

balance, and functional mobility. BMC Geriatrics,

22(1):615.

Cardillo, E., Li, C., and Caddemi, A. (2022). Millimeter-

wave radar cane: A blind people aid with moving hu-

man recognition capabilities. IEEE Journal of Elec-

tromagnetics, RF and Microwaves in Medicine and

Biology, 6(2):204–211.

A Falls Risk Screening Tool Based on Millimetre-Wave Radar

167

Choi, J., Parker, S. M., Knarr, B. A., Gwon, Y., and Youn, J.-

H. (2021). Wearable Sensor-Based Prediction Model

of Timed up and Go Test in Older Adults. Sen-

sors, 21(20):6831. Number: 20 Publisher: Multidis-

ciplinary Digital Publishing Institute.

Diao, Y., Lou, N., Liang, S., Zhang, Y., Ning, Y., Li, G.,

and Zhao, G. (2021). A Novel Environment-Adaptive

Timed Up and Go Test System for Fall Risk Assess-

ment With Wearable Inertial Sensors. IEEE Sensors

Journal, 21(16):18287–18297.

Dierick, F., Stoffel, P., Sch

utz, G., and Buisseret, F.

(2022). High Speciﬁcity of Single Inertial Sensor-

Supplemented Timed Up and Go Test for Assessing

Fall Risk in Elderly Nursing Home Residents. Sen-

sors, 22(6).

Frenken, T., Brell, M., G

overcin, M., Wegel, S., and Hein,

A. (2013). aTUG: technical apparatus for gait and bal-

ance analysis within component-based Timed Up &

Go using mutual ambient sensors. Journal of Ambient

Intelligence and Humanized Computing, 4(6):759–

778.

Fudickar, S., Hellmers, S., Lau, S., Diekmann, R., Bauer,

J., and Hein, A. (2020). Measurement System for

Unsupervised Standardized Assessment of Timed ”Up

& Go” and Five Times Sit to Stand Test in the

Community-A Validity Study. Sensors, 20(10).

Gambi, E., Ciattaglia, G., Santis, A. D., and Senigagliesi, L.

(2020). Millimeter wave radar data of people walking.

Data in Brief, 31(105996).

Greene, B., Doheny, E., O’Halloran, A., and Kenny, R.

(2014). Frailty status can be accurately assessed using

inertial sensors and the TUG test. Age and Ageing,

43(3):406–411.

Kataoka, Y., Saito, Y., Takeda, R., Ishida, T., Tadano, S.,

Suzuki, T., Nakamura, K., Nakata, A., Osuka, S., Ya-

mada, S., Samukawa, M., and Tohyama, H. (2023).

Evaluation of Lower-Limb Kinematics during Timed

Up and Go (TUG) Test in Subjects with Locomotive

Syndrome (LS) Using Wearable Gait Sensors (H-Gait

System). Sensors, 23(2).

Laved

an, A., Viladrosa, M., J

urschik, P., Botigu

e, T., Nu

ın,

C., Masot, O., and Laved

an, R. (2018). Fear of falling

in community-dwelling older adults: A cause of falls,

a consequence, or both? PLoS one, 13(3):e0194967.

Li, T., Chen, J., Hu, C., Ma, Y., Wu, Z., Wan, W., Huang,

Y., Jia, F., Gong, C., Wan, S., and Li, L. (2018).

Automatic Timed Up-and-Go Sub-Task Segmenta-

tion for Parkinson’s Disease Patients Using Video-

Based Activity Classiﬁcation. IEEE Transactions

on Neural Systems and Rehabilitation Engineering,

26(11):2189–2199.

Moncada, L. V. V. (2011). Management of Falls in Older

Persons: A Prescription for Prevention. American

Family Physician, 84(11):1267–1276.

NICE (2013). National Institute for Health and Care Ex-

cellence. Falls in older people: assessing risk and pre-

vention. Clinical guideline [CG161]. Technical report,

NICE.

NodeNs Medical Ltd. (2023). Millimetre Wave Radar.

https://nodens.eu.

Rajab, K. Z., Wu, B., Alizadeh, P., and Alomainy, A.

(2021). Multi-target tracking and activity classiﬁca-

tion with millimeter-wave radar. Applied Physics Let-

ters, 119(3):034101.

Richards, M. A. (2014). Fundamentals of Radar Signal

Processing. McGraw-Hill Education, New York, 2nd

edition edition.

Savoie, P., Cameron, J. A. D., Kaye, M. E., and

Scheme, E. J. (2020). Automation of the Timed-

Up-and-Go Test Using a Conventional Video Camera.

IEEE Journal of Biomedical and Health Informatics,

24(4):1196–1205. Conference Name: IEEE Journal

of Biomedical and Health Informatics.

Soubra, R., Mourad-Chehade, F., and Chkeir, A. (2023).

Automation of the Timed Up and Go Test Using a

Doppler Radar System for Gait and Balance Analysis

in Elderly People. Journal of Healthcare Engineering,

2023:e2016262. Publisher: Hindawi.

Spina, S., Facciorusso, S., D’ascanio, M. C., Morone, G.,

Baricich, A., Fiore, P., and Sandamato, A. (2022).

Sensor based assessment of turning during instru-

mented Timed Up and Go Test for quantifying mo-

bility in chronic stroke patients. European Journal of

Physical and Rehabilitation Medicine, 59(1):6–13.

Sun, R. and Sosnoff, J. J. (2018). Novel sensing technology

in fall risk assessment in older adults: a systematic

review. BMC Geriatrics, 18(1):14.

Takayanagi, N., Sudo, M., Yamashiro, Y., Lee, S.,

Kobayashi, Y., Niki, Y., and Shimada, H. (2019).

Relationship between Daily and In-laboratory Gait

Speed among Healthy Community-dwelling Older

Adults. Scientiﬁc Reports, 9(1):3496. Number: 1 Pub-

lisher: Nature Publishing Group.

Taylor, K. I., Staunton, H., Lipsmeier, F., Nobbs, D., and

Lindemann, M. (2020). Outcome measures based

on digital health technology sensor data: data- and

patient-centric approaches. npj Digital Medicine,

3(1):1–8. Number: 1 Publisher: Nature Publishing

Group.

WHO (2021). World Health Organization fact sheet on

falls providing key facts and information on scope

of the problem, who is at risk, prevention, who

response. Falls, https://www.who.int/news-room/fact-

sheets/detail/falls.

Yu, Z., Taha, A., Taylor, W., Zahid, A., Rajab, K., Hei-

dari, H., Imran, M. A., and Abbasi, Q. H. (2022).

A Radar-Based Human Activity Recognition Using a

Novel 3-D Point Cloud Classiﬁer. IEEE Sensors Jour-

nal, 22(19):18218–18227.

ICT4AWE 2024 - 10th International Conference on Information and Communication Technologies for Ageing Well and e-Health

168