Data Acquisition System for a Wearable-Based Fall Prevention

Raul Kaizer

1 a

, Leonardo Sestrem

1 b

, Tiago Franco

1 c

, Jo

ao Gonc¸alves

1 d

ao Paulo Teixeira

1,2 e

, Jos

e Lima

1,2 f

, Jos

e Augusto Carvalho

1,2 g

and Paulo Leit

1,2 h

Research Center in Digitalization and Intelligent Robotics (CeDRI), Instituto Polit

ecnico de Braganc¸a,

Campus de Santa Apol

onia, 5300-253 Braganc¸a, Portugal

Associate Laboratory for Sustainability and Technology (Sustec), Instituto Polit

ecnico de Braganc¸a,

Campus de Santa Apol

onia, 5300-253 Braganc¸a, Portugal

Keywords:

Wearable Health Monitoring, Data Acquisition, Fall Detection.

Abstract:

Reliable ways to treat and monitor patients remotely have been researched and proposed by numerous people.

Many of these propositions are under the wearable category due to it usually not requiring deep knowledge

to be handled and its durability. Among the many applicable ways, fall monitoring has gained importance as

the world population ages and countries aim to increase the quality of life. For it to be possible, there are

many ways such as analyzing muscle response, body position, or brain activities, but for most of them, the

result ends up being expensive and or inaccurate. With this in mind, this paper brings the development of an

acquisition system for electromyography, electrocardiography, body position and temperature. The acquired

data is transmitted to the smartphone through Bluetooth Low Energy (BLE) and then sent to a secure cloud

to be provided to the physician. In future works, artiﬁcial intelligence codes will analyze the data patterns to

predict fall occurrences and establish functional electrical stimulation (FES) routines to prevent falls and or

treat the patients according to their necessities.

1 INTRODUCTION

Isolated in their homes, elderly people with mobil-

ity issues are mostly unable to receive common treat-

ment from clinics, since constant consultations may

become a ﬁnancial burden, and prolonged consulta-

tions are more demanding on therapists and increase

patient discomfort. A quick solution to this kind of

situation is requiring the health professionals to have

a more direct approach and move to each patient’s

house, however, is fated to become an obstacle to the

health system. That is because the number of people

requiring treatment increases at a higher rate than the

amount of specialized caregivers (WHO, 2022).

It is difﬁcult to hope for the elderly to keep exer-

cise or treatment routines by themselves, which just

https://orcid.org/0000-0001-9273-2257

https://orcid.org/0000-0002-9344-3075

https://orcid.org/0000-0001-8574-4380

https://orcid.org/0000-0002-3502-7444

https://orcid.org/0000-0002-6679-5702

https://orcid.org/0000-0001-7902-1207

https://orcid.org/0000-0002-6074-8112

https://orcid.org/0000-0002-2151-7944

increases the burden on professionals and relatives.

Besides, even with proper treatment, the high amount

of time they are left alone in their houses increases the

chances of them falling or suffering any injury, and

not be found by anyone for a long time. Even when

considering home treatment, visits may prove them-

selves of no help if taken into account that the cause

of the problem can be an isolated event, that may not

occur or goes unnoticed during the visit (Jeong et al.,

2021).

When dealing speciﬁcally with the fall problem,

it is difﬁcult to consider the early diagnosis as an

alternative. This is due to the diverse and complex

fall causes, and also individual susceptibility (Rakugi

et al., 2022). This problem can be easily overcome by

including non-invasive monitoring and rehabilitation

devices. These devices can be implemented in wear-

ables and can be used to remove the need for constant

human interference.

Up to date, there are wearables with capabilities

to analyze muscle, heart, and brain activity, or other

data signals such as body temperature and position. It

is also possible to monitor other parameters, but these

may require invasive sensors which, in addition to be-

Kaizer, R., Sestrem, L., Franco, T., Gonçalves, J., Teixeira, J., Lima, J., Carvalho, J. and Leitão, P.

Data Acquisition System for a Wearable-Based Fall Prevention.

DOI: 10.5220/0011926500003414

In Proceedings of the 16th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2023) - Volume 5: HEALTHINF, pages 701-710

ISBN: 978-989-758-631-6; ISSN: 2184-4305

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

701

ing uncomfortable, make it difﬁcult for patients with-

out the necessary expertise to use them. Other models

present treatment features like electrotimulation to as-

sist physical therapy, an example is Functional ELec-

tric Stimulation (FES), but they usually also require

speciﬁc knowledge to some degree.

The number of signals that make it possible to ac-

cess the patient’s current situation allows many pos-

sible combinations, but implementing all of them to-

gether is a problem. Since it would increase both the

number of components and the price. That way, this

work brings the development and implementation of

a modular biosignal acquisition system specially de-

veloped to be integrated together with FES actuators.

The aim is to analyze the data with AI and generate

personalized FES routines for treatment or impulses

to prevent muscle imminent failure that would lead

to a fall. For it to be possible, the stabilizing mus-

cles located in the abdomen and lumbar are monitored

and stimulated individually. So each one of the target

muscles require a pair of EMG sensors and a pair of

FES actuators.

However, the description and analysis of the stim-

ulation system and the AI structure are outside this pa-

per’s scope due to their complexity requiring a more

in depth study. Therefore, this work will focus on the

acquisition portion of the system, i.e., electromyog-

raphy (EMG) from the stabilizing muscles located in

the abdomen and lumbar, electrocardiography (ECG),

body temperature, and position change (IMU). Fig-

ure 1 (Muharrem Adak, Vecteezy, 2022) exempliﬁes

the placement of EMG and ECG sensors, keeping in

mind that the placement may present slightly varia-

tions according to the patients’ anatomy. The tem-

perature and IMU sensors present a much more ﬂexi-

ble position range, being highly dependant on comfort

and practicality.

Figure 1: Placement of EMG (red) and ECG (black) sensors

from the front (a) and the back (b).

Acquiring EMG signals allow the visualization

of the stabilizing muscles’ behavior, i.e. when they

work properly in a coordinated way to maintain the

patient stabilized or when there is some abnormal-

ity. The same applies to ECG signals; if the car-

diac waves present abnormalities or sudden changes,

the patient may have had a problem and needs help.

The IMU helps in monitoring the angular displace-

ment of the upper body that, together with the EMG

signals, allows the physician team to determine if the

change was voluntary or a fall. To complete the ac-

quired data, body temperature will be constantly ver-

iﬁed since it often indicates health problems or other

factor changes.

The data acquired from both the electrodes and the

other sensors (IMU and thermometer) will be wire-

lessly transmitted to the smartphone with Bluetooth

Low Energy (BLE) without being treated to reduce

the delay (Qian et al., 2017; Marin-Pardo et al., 2020).

The smartphone then serves as a mediator to the cloud

where the data stays available to the physician to con-

sult and make decisions and to AI algorithms to anal-

yse the patterns and predict the falls.

The paper is separated as follows: Section 2

presents the related work. Section 3 describes the

proposed architecture and system requirements while

Section 4 focus on the implementation of the pro-

posed acquisition system and explains the functional-

ities and integration of the circuit modules. Section 5

analyses the prototype experimental results and Sec-

tion 6 summarizes the conclusions and points out the

future work.

2 RELATED WORK

With higher living standards, well-developed coun-

tries tend to research and develop systems to improve

the quality of life of the elderly, being fall detection

one of the most researched (Wang et al., 2020). The

problem is that for most systems, the detection of-

ten results in false alarms due to imprecise data in-

put and analysis. Recent studies indicate that fusing

the signals of different sensors is ideal to lower the

false alarm occurrences of fall detection and predic-

tion system (Siwadamrongpong et al., 2022). Besides

the increase in accuracy, analyzing multiple signal

responses improves the robustness of such systems,

since it needs more indicative responses to trigger the

alarm.

Examples of devices with only one type of sig-

nal input such as EMG are often found in the litera-

ture (Zhu et al., 2021; Ali et al., 2021; Steinberg et al.,

2019). Even for the works that develop one device

capable of acquiring different signals, the inability to

treat and analyze these signals in an integrated way is

a problem (Wearable Sensing, 2022). This increase

both the cost and the amount of components patients

need to deal with to be able to monitor their daily

life. Some works reported simpler technologies with

WHC 2023 - Special Session on Wearable HealthCare

702

higher accuracy such as computer vision and doppler

radar (Ren and Peng, 2019) but these tend to be ex-

pensive. Regardless of their precision, both these sys-

tems and the less accurate ones present the same prob-

lem, they only detect the fall after it occurs or during

the fall.

Aiming to predict the fall and with improvements

in electronics and materials, medical diagnosis de-

vices started to be integrated into wearables without

compromising the patient’s ﬂexibility, easy mobility,

and capacity of executing their daily tasks. These

wearables can be found in various forms and shapes,

such as watches, t-shirts, armbands, etc. Most de-

vices are projected to monitor vital signals (Lou et al.,

2020) and, recently, there is a tendency to invest

in wearables with extra functionalities (Bruce-Brand

et al., 2012; Qian et al., 2017) like life support sys-

tems.

There are many different ways to detect a fall with

sensors adaptable to wearables, such as analyzing the

muscle response through EMG signals. (Biometrics

Ltd, 2021) is one of the many sellers of EMG sensors

and systems ready to be used for research and other

purposes. But, as said before, to increase the preci-

sion of the systems, fusing other signal responses with

EMG is necessary.

Combining EMG and IMU responses, (ZiYing

et al., 2021) uses machine learning techniques to treat

the data and, based on previously known patterns,

decide if the current combination of inclination and

muscle response is leading toward a fall. All this

data is sent via Bluetooth to a computer, which de-

creases the practicality of the device by having to stay

near a computer during the monitoring. Other works

also see the importance of implementing some type

of artiﬁcial intelligence (especially machine learn-

ing) to better perceive the EMG patterns and risks of

falling (Jeong et al., 2021; Rescio et al., 2018).

A different approach is shown by (ZiYing et al.,

2021) where the system’s main routine is to de-

tect Movement Related Cortical Potential (MRCP)

through an electroencephalogram (ECG). When there

is an intention of movement, the EMG sensors are ac-

tivated and the force threshold is veriﬁed to decide

if there is enough force to realize the movement. If

the threshold is below the necessary, the patient is in-

structed to wait for aid and the caregiver receives a

warning about the situation.

Just like EMG, there is still the application of ECG

as the main parameter for fall risk assessment (Shim-

mer Discovery in Motion, 2022). Since the ECG ac-

quires the response from the heart, analyzing anoma-

lies in the cardiac waves may reveal a possible fall

event. The number of works using mainly ECG or

a fusion of cardiac signals and some other parame-

ter increases each year but the tendency is for them

to integrate some type of artiﬁcial intelligence into

it (Butt et al., 2021; Melillo et al., 2015). This in-

tegration of ECG and AI tends to show good results

when predicting fall risk and lessens the burden on

specialists (Queralta et al., 2019).

3 SYSTEM DEVELOPMENT

By combining the responses from diverse biosensors,

this work’s aim is to analyze a modular system to

acquire, process and transmit the necessary biosig-

nals. These signals include the EMG signal of the

abdomen and lower back, ECG, IMU (by verifying if

the subject is inclined or upright) and body tempera-

ture. Later, these signals are to be used as input for AI

codes to create personalized FES routines.

3.1 System Requirements

Each biosignal that must be acquired has different

properties and so, presents different system parame-

ters requirements, all of which must be handled by

the same microcontroller. Starting with EMG acqui-

sition, to properly acquire the muscles’ responses the

sampling rate of the acquisition must be at least 1 kHz

on a frequency bandwidth of 18−480 Hz. To acquire

the response of four pairs of electrodes (two pairs at

the abdomen and two at the lower back), the micro-

controller was conﬁgured to receive data through mul-

tichannel.

Regarding the ECG signal acquisition, the fre-

quency bandwidth can vary from 0.5 Hz up to 200 Hz

and the best sampling rate frequencies are around

120 Hz and 200 Hz (Kwon et al., 2018; Ajdaraga

and Gusev, 2017). The system also must include ba-

sic signal treatment features to eliminate power line

noise which alters the ECG signal acquired around

the 60 Hz.

Unlike the previous two signals which were sig-

nals acquired directly from the body, the response of

the IMU comes from an accelerometer and a gyro-

scope, each generating three responses, one for each

axis (x, y and z). These two devices work with a dif-

ferent communication protocol called Inter-Integrated

Circuit, or I2C what prevent the IMU acquisition rate

to be managed by the ESP-32’s timer, and instead,

works based on code routines. With this, the sampling

rate was set to 100 Hz (Zhou et al., 2020) since it is

the minimum necessary to accompany simple daily

movements

The developed system also must be prepared to

Data Acquisition System for a Wearable-Based Fall Prevention

703

handle a body temperature sensor. This sensor will

acquired the body’s temperature with a sampling rate

of 10 Hz in the same way the IMU, following the code

routine.

Since the purpose of this work is to develop the

necessary microelectronic to acquire and transmit

biosignals from inside a wearable, there are a few re-

quirements it must follow besides the ones speciﬁc

to each signal and circuit. To not inﬂuence the daily

activities of the patient, the ﬁnal circuit must be light

and small without compromising comfort when wear-

ing the T-shirt.

Another factor to be taken into account when

thinking about not interfering in the daily tasks is the

absence of wires, i.e., the circuit must be wireless both

for energy source and data transmission. To avoid bat-

tery discharge in the middle of the session, energy-

efﬁcient batteries and a battery monitor are also req-

uisites included in the scope of the project.

Fulﬁlling all these requirements while using high-

end materials with their durability and quality would

be an easier task, but the cost would turn exponen-

tially higher. To ensure that most people have the pos-

sibility of owning one exemplar, the cost of the ﬁnal

product must be kept accessible while using durable

and trustworthy materials.

3.2 System Architecture

The proposed architecture, illustrated in Fig-

ure 2, constitutes an advanced technological sys-

tem/medical device with biofeedback response, since

the built-in sensors/actuators will be able to monitor

muscle and vital signs and, at the same time, act by

functional electrical stimulation (FES). Furthermore,

the device allows to work as an emergency system, in

situations of imminent fall, in which the sensors will

trigger a muscle activation, remotely controlled, of

the abdomen and/or low back muscles.

The designed system offers the possibility for the

patient to carry out the therapy comfortably at home,

allowing the reduction of social and economic bur-

dens, as well as the opportunity for hospital centers

to reduce the number of falls of users in the context

of hospitalization. The proposed wearable device is

responsible for ensuring the acquisition and condi-

tioning (ampliﬁcation, ﬁltering and conversion) of the

EMG, ECG, IMU and temperature data properly, and

subsequently send the acquired information to a mo-

bile application running on a smartphone/tablet.

The EMG sensors are destined to capture the mus-

culoskeletal signal, which measures the potential dif-

ference between the muscle ﬁbers recruited during a

muscle contraction. In this way, the designed archi-

Figure 2: System architecture focusing the acquisition, sig-

nal conditioning and stimulation system.

tecture employs four pairs of EMG sensors, two for

the abdomen and two for low back muscles to monitor

the signals aiming to assist the fall detection system.

The ECG electrodes allow monitoring of the

user’s heart rate, which can be useful to check if a fall

has occurred or even alarm if another heart disease

is detected. This allows us to understand the level

of physiological arousal that someone is experienc-

ing, but it can also indicate a better understanding of

the psychological state (Berkaya et al., 2018). There-

fore, recording heart rate data gives access to sev-

eral parameters, such as Heart Rate (HR) that reﬂects

arousal, Inter-Beat Interval (IBI) and Heart Rate Vari-

ability (HRV), that are closely related to emotional

arousal (Berkaya et al., 2018).

Temperature sensor affords to measure the pa-

tient’s body temperature, which allows to monitor

and alarm if any anomaly occurs. The sensor is po-

sitioned in the core, that refers to the temperature

of the body’s organs. This metric ﬂuctuates follow-

ing physiological processes, such as circadian cy-

cles, menstruation, illness or physical activity (Mina,

2021). Regarding remote patient monitoring, varia-

tions in core body temperature often provide insight

on health-related problems prior to the appearance

of other symptoms (Dolson et al., 2022). Therefore,

accurately monitoring the patient’s body temperature

can help to identify issues at an early stage.

The IMU provides to measure the acceleration and

angle, which allows jointly with the EMG to perform

the biofeedback for the stimulation module in the fall

detection system. Therefore, IMU sensor is the most

signiﬁcant in the wearable its metrics will be the base

WHC 2023 - Special Session on Wearable HealthCare

704

for the fall detection system where according to the

patient’s core angle and the acceleration value, then

the respective actuator array will be powered on in

order to compensate this undesirable motion and re-

establishing the patient corporal posture.

The FES actuators perform the electrostimulation

through electric pulses, aiming to assist in the user

posture correction and preventing imminent falls reg-

istered by the monitoring system. These actuators are

crucial in physiotherapy and monitoring contexts aim-

ing to strengthen the muscles of the abdomen and low

back, that are the main muscular group recruited for

posture stabilization and elderly people usually have

atrophy in these muscles, which contributes to an in-

creased risk of falls.

After collected each information from their re-

spective transducer, it is forwarded to a signal con-

ditioning block that provides a properly ampliﬁcation

to a higher amplitude aiming to decrease the trans-

mission losses, in addition, this same block allows to

ﬁlter the signal to attenuate undesired frequencies and

noise.

Next, all the sensor data is converted from ana-

logue to digital through an Analog/Digital Converter

(ADC) to be able to transmit all content to the micro-

controller unit (MCU). The MCU will receive all the

information, store it and process locally the data by

executing control rules to perform the muscle stimu-

lation function. These control rules triggered accord-

ing to the biofeedback from the EMG and IMU col-

lected data (providing a loop between the acquisition

and stimulation systems). Therefore, the MCU will

send commands to the stimulation driver according to

the IMU angle and acceleration values, to avoid im-

minent falls and guaranteeing the user integrity and

health.

The mobile application is responsible for manag-

ing the information that deﬁnes the treatment sessions

in real-time, that is, receiving the data collected by

the set of sensors and sending the stimulation rules

considering the biofeedback to the wearable system.

This management is done in the mobile app due to its

greater computational power, enabling more complex

algorithms to be explored to provide adequate stimu-

lation adaptation for each patient.

In this way, before a treatment session starts, the

initial conﬁguration parameters deﬁned by the physi-

cian are downloaded from the cloud to the mobile

app. During the session, the mobile app will only

communicate with the wearable, requiring no internet

connection to perform the session. When the session

ends, all the data collected and the stimulation rules

applied will be sent to the cloud and be available for

the physician to evaluate and prepare the next treat-

ment session.

4 SYSTEM HARDWARE

DEVELOPMENT

Considering the aforementioned project requirements

and the designed system architecture, a prototype of

an integrated acquisition system was developed as il-

lustrated in Figure 3 (note that the stimulation sys-

tem, the mobile app, the fall detection system and the

AI algorithms are not detailed in this paper). The

proposed solution comprises the data acquisition of

EMG, ECG, IMU and temperature, which operates in

real-time and integrated with the mobile app.

Figure 3: Schematic of prototype data acquisition and stim-

ulation.

4.1 Bio-Signal Acquisition Module

The designed acquisition prototype employs commer-

cial and low-cost sensors, namely MPU 6050, EBZ-

AD8233, SI 7021 and AD8232 Heart Monitor, as pre-

sented in Figure 3.

Seeking to measure the signals produced by the

abdomen and low back muscles, the evaluation board

AD8233 was selected to perform the signal condition-

ing (amplifying and ﬁltering). Originally, this device

was designed by the Analog Devices, Inc. (ADI) as

a heart rate monitor front end to acquire ECG sig-

nals providing a frequency bandwidth of 7 − 25 Hz.

Then, aiming to measure EMG signals an adaptation

in the components of the High-pass (HPF) and Low-

pass (LPF) ﬁlters circuits were necessary, resulting on

a new frequency bandwidth of 15 − 480 Hz. Those

changes allow to acquire and amplify musculoskele-

tal signals properly since the biosignals have low am-

plitude levels, ranging from µV to a few mV.

The AD8232 Heart Monitor is a low-cost devel-

Data Acquisition System for a Wearable-Based Fall Prevention

705

opment kit speciﬁed for ECG signals, which offers

complete voltage interval for reading (0 − 3.3 V) to

sample the heart rate signal amplitude and provides

a frequency spectrum of 7 − 500 Hz. ECGs can be

extremely noisy, the AD8232 acts as an operational

ampliﬁer (op amp) to assist obtain a clear signal from

the PR and QT intervals easily. Furthermore, this in-

tegrated signal conditioning block is designed to ﬁl-

ter small biopotential signals in the presence of noisy

conditions, such as those created by motion or remote

electrode placement.

The temperature sensor SI 7021 offers a 400 kHz

C bus communication, high accuracy for a low-cost

device ± 0.4

◦

C providing a temperature ranging of

−10

◦

C to 85

◦

C, and low-power sensor presenting

150 µA of active current.

The IMU sensor adopted is the MPU 6050, which

offers a 400 kHz I

C bus communication, a low-

cost sensor, low power consumption 3.6 mA and

500 µA operating currents of the gyroscope and ac-

celerometer respectively, and high-performance re-

quirements for wearable devices. This sensor model

allows tracking 6-axis motion, being 3-axis for the

gyroscope and 3-axis the accelerometer. Moreover,

the MPU 6050 enables to follow fast and slow

motions providing a user-programmable scale for

the gyroscope (±250,±500,±1000, ±2000)

◦

/s and

accelerometer(±2g,±4g,±8g,±16g) m/s

4.2 Microcontroller Unit

To be possible to read data from all the sensors,

a MCU was employed to lead all acquisition and

transmission functionalities, as presented in Figure 3.

Therefore, the whole system architecture was based

on the ESP32 microcontroller which offers low-cost,

wireless communication (Wi-Fi and BLE) and capa-

bility to operate with dual core.

ESP32 is a MCU with enough computational

power to acquire sensor data from all the sensors, be-

ing capable to operate at high frequencies since the

clock of a standard model, such as ESP32-WROOM-

32D, is above 150 MHz. In addition, the ESP32 al-

lows separating through timer tasks the distinct ac-

quisition routines for each sensor, respecting their

sampling frequencies while avoiding delays and data

losses.

The ESP32 has an ADC with 12 bits of resolu-

tion, permitting to convert analogue voltage values

from 0.8 mV while e.g. the Arduino microcontroller

only has 10 bits of resolution that allows to measure

a signal from 4.89 mV. Lastly, this device provides

a dual-core functionality that is essential to guaran-

tee the correct acquisition and data transmission inte-

grated, being able to execute separately each function

without delays or interference among the tasks.

4.3 Data Transmission

The data transmission between the microcontroller

and the smartphone was an important part of the

project having in mind that data loss during the com-

munication could lead to wrong or incomplete infor-

mation being displayed to the medical team. Aiming

to implement a reliable transmission technology with

enough security and customizable, Bluetooth Low

Energy (BLE) was implemented.

Besides the items aforementioned, BLE is sup-

ported by most smartphones, tablets and similar de-

vices nowadays. The coverage range of BLE 4.2 also

reaches greater distances, even being able to commu-

nicate at around 100 meters. Power consumption that

is already low for this version can be lowered even

more with simple conﬁgurations, what is an advanta-

geous point for a wireless wearable powered by bat-

teries.

Serving as the central, the mobile application is

responsible for sending requests to peripheral device

(the wearable) to read and send the acquired data or

to write new routines. In the meantime, the periph-

eral device can only send indications, responses to

requests and notiﬁcation about characteristics previ-

ously selected during the conﬁguration. Since the

ESP-32 employed has to deal with vast amounts of

data being transferred regarding all the signals ac-

quired, the signals responses pass through different

buffers while the responses required by the client are

sent directly. To better handle this task one of ESP-

32’s threads is exclusively to acquire, manage and

transmit the data to the smartphone in real-time.

Expanding the maximum throughput to 517 bytes

was another method for improving the transmission

performance since with this the computational cost

associated to constantly sending and receiving the

overhead of packages is lowered.

This necessity becomes even more visible sup-

posing the mobile application will present some type

of play/pause/stop command. With this in mind, the

transmission code was developed in a way that even

if the transmission is paused, biosignals acquisition

keeps running on the background while being stored

locally to avoid data loss until the treatment is re-

sumed.

WHC 2023 - Special Session on Wearable HealthCare

706

5 EXPERIMENTAL RESULTS

In order to verify the accomplishment of the require-

ments, the prototype was used to perform several ex-

perimental tests. The acquired signals were derived

from a healthy person executing a simple set of move-

ments during the tests routine. While standing, bend

the upper body forward, backward, to the left and ﬁ-

nally to the right holding the position for a couple

seconds in each one. Afterwards, the subject would

strongly contract the biceps once followed by three

weaker contractions.

5.1 Data Acquisition Test

Auto-adhesive wet Ag-AgCl electrodes were chosen

to ensure that the noise from the sensor/skin interface

was the minimum possible. Besides being the golden

standard, these sensors present common connection

characteristics that allow them to be used with other

acquisition devices which makes possible to compare

the signals acquired by commercial devices and the

prototype.

Although the acquisition boards used are commer-

cially available and already validated, the EMG ac-

quisition boards had some components changed. The

resulting acquisition system was then compared with

Bitalino’s output as can be seen in Figure

refﬁg:comp.

Figure 4: EMG comparison between Bitalino and the pro-

totype.

During the tests, it was veriﬁed that simply mov-

ing the cables resulted in high noise inputs, some-

times requiring the test to be remade. Even so, the

best signal-to-noise ratio when using the prototype to

acquire EMG signals achieved 14.68 dB in contrast

with 18.15 dB obtained when using commercial ac-

quisition devices.

Figure 5 and Figure 6 are examples of data ac-

quired with the proposed systems.

As can be seen, both signals still have a great

amount of noise present, even so, both the biceps

contractions and the PQRST complex are clearly de-

tectable. Since the IMU sensor does not need to stay

in contact with the body, it can be kept closer to the

Figure 5: EMG acquired using the prototype.

Figure 6: ECG acquired using the prototype prototype.

microcontroller, requiring shorter cables and reduc-

ing the movement and instability artifacts. This can

be seem in Figures 7 and 8.

Figure 7: IMU response to bending the body to the front

and the back.

Figure 8: IMU response to bending the body to the left and

the right.

Both the EMG and ECG signal, as well as the

six IMU responses and the temperature were acquired

and transmitted to the microcontroller via cables. For

the duration of the tests (approximately one hour and

a half), the ESP was able to handle all data without

problems.

5.2 Data Transmission Test

Besides acquiring the signals and handling the rou-

tines, the microcontroller was responsible for trans-

mitting the data through BLE. This type of communi-

cation presents the risk of losing some data be it due

to some glitch at the smartphone or the ESP not be-

ing able to keep track of the packets sent resulting in

packets being skipped.

To closely verify how many packets of each type

Data Acquisition System for a Wearable-Based Fall Prevention

707

of data was received, the prototype was put to work

with simulated values as acquisition’s responses. The

time between the received packets was monitored for

ﬁfty packets for each of the four types of signals. This

being said, it is important to note that comparison

between two types of signals may cause wrong con-

clusions as the transmission system works based on

buffers, so signals with greater acquisition rates, like

EMG, will present smaller time differences than sig-

nals with smaller acquisition rates (i.e. temperature).

Figure 9 shows the time between consecutive

packets for the acquired data. The ideal response

would be to present the same delay between all the

packets of one type of signal. To illustrate, if all the

EMG packets presented zero and 100 ms consecu-

tively between two packets, it would be proved that,

for this signal, the routine is being well handled by the

ESP. But as can be seem, the microcontroller presents

some problems to handle the acquisition and trans-

mission at the same time around the thirtieth packet

which cause irregularities in all four signals transmis-

sion.

Even with some irregularities in the transmission,

at long term tests the mean time is kept close to the

settings so there is no major prejudice to the ﬁnal data

set. Since the ESP has two cores and just one of them

is being used to this task, employing other microcon-

troller with more cores or with higher performance

is bound to solve the discrepancies in delay between

packets.

6 CONCLUSIONS

Employing wearable systems in daily life is a reliable

way to improve healthcare while maintaining the bur-

den on the healing system at acceptable levels. The

possibility of remote monitoring of vital signals is

an important factor for both health professionals and

concerned relatives.

The proposed innovative system aims at revolu-

tionizing the method for detecting fall occurrences

with a low-cost modular device embedded into a

wearable T-shirt. Acquiring a set of important signals

such as EMG, ECG, body temperature and body posi-

tion allows the usage of this data for not only detect-

ing and preventing falls but to detect improvements

resulting from treatments. To provide the acquired

data to the medical team this system presents connec-

tion to the smartphone, which safely sends the data to

the cloud. To lower the burden on the ESP, data treat-

ment and storage are done on the smartphone, which

reduces the need for more expensive microcontrollers.

Preliminary experimental tests show that the

main acquisition system requirements were fulﬁlled,

mainly regarding the acquisition rates and identiﬁca-

tion of each signal in the buffers. The microcontroller

also proved itself capable of handling the data trans-

mission while keeping the acquisition routines on.

There is a need to improve the transmission routines

since some packets presented higher delay, although

the frequency with which this occurred and the mag-

nitude of the delay did not have great inﬂuence in long

term.

During the experiments, it was clear that the con-

nections between the boards created instability among

the components resulting in motion artifacts and noise

increase. Also, isolating one core of the ESP reduced

the processing capabilities so using both cores to ac-

quire and transmit the data may present a great posi-

tive impact on the result. Implementing the routines

in other microcontroller models may be an option as

well.

Since the acquisition system was developed to be

integrated with the stimulation system, future works

are summarized in implementing the necessary elec-

tronics and verifying if the microcontroller is able

to handle all tasks, taking into account the complex-

ity of the stimulation routines. With both stimula-

tion and acquisition working together, a new printed

circuit board (PCB) will be designed to reduce even

more connection problems and circuit size. With this,

the analysis of the system’s integration viability into

wearables and battery consumption can be easily and

properly done.

Then, the PCB will be used for tests with volun-

teers for validation purposes and the acquired signals,

used to populate the database. With this data, the AI

will be trained to develop customized stimulation rou-

tines for each muscle according to the necessary treat-

ment or to prevent some fall occurrence.

ACKNOWLEDGEMENTS

This work was supported by the European Regional

Development Fund (ERDF) through the Operational

Programme for Competitiveness and International-

ization (COMPETE 2020), under Portugal 2020 in

the framework of the NanoID (NORTE-01-0247-

FEDER-046985) Project.

This work has been supported by the Founda-

tion for Science and Technology (FCT, Portugal)

through national funds FCT/MCTES (PIDDAC) to

CeDRI (UIDB/05757/2020 and UIDP/05757/2020)

and SusTEC (LA/P/0007/2021).

WHC 2023 - Special Session on Wearable HealthCare

708

Figure 9: Received Data Packets ∆ Time.

REFERENCES

(2022). Ageing and health. https://www.who.int/news-

room/fact-sheets/detail/ageing-and-health. Accessed:

2022-05-12.

Ajdaraga, E. and Gusev, M. (2017). Analysis of sam-

pling frequency and resolution in ecg signals. In 2017

25th Telecommunication Forum (TELFOR), pages 1–

4. IEEE.

Ali, H., Naing, H. H., and Yaqub, R. (2021). An iot assisted

real-time high cmrr wireless ambulatory ecg monitor-

ing system with arrhythmia detection. Electronics,

10(16):1871.

Berkaya, S. K., Uysal, A. K., Gunal, E. S., Ergin, S., Gunal,

S., and Gulmezoglu, M. B. (2018). A survey on ecg

analysis. Biomedical Signal Processing and Control,

43:216–235.

Biometrics Ltd (2021). Datalog MWX8.

http://www.biometricsltd.com/datalog.htm. Ac-

cessed: 26.11.2021.

Bruce-Brand, R. A., Walls, R. J., Ong, J. C., Emerson, B. S.,

O’Byrne, J. M., and Moyna, N. M. (2012). Effects

of home-based resistance training and neuromuscular

electrical stimulation in knee osteoarthritis: a random-

ized controlled trial. BMC musculoskeletal disorders,

13(1):1–10.

Butt, F. S., La Blunda, L., Wagner, M. F., Sch

afer, J.,

Medina-Bulo, I., and G

omez-Ullate, D. (2021). Fall

detection from electrocardiogram (ecg) signals and

classiﬁcation by deep transfer learning. Information,

12(2):63.

Dolson, C. M., Harlow, E. R., Phelan, D. M., Gabbett, T. J.,

Gaal, B., McMellen, C., Geletka, B. J., Calcei, J. G.,

Voos, J. E., and Seshadri, D. R. (2022). Wearable sen-

sor technology to predict core body temperature: A

systematic review. Sensors, 22(19):7639.

Jeong, J.-W., Lee, W., and Kim, Y.-J. (2021). A real-time

wearable physiological monitoring system for home-

based healthcare applications. Sensors, 22(1):104.

Kwon, O., Jeong, J., Kim, H. B., Kwon, I. H., Park, S. Y.,

Kim, J. E., and Choi, Y. (2018). Electrocardiogram

sampling frequency range acceptable for heart rate

variability analysis. Healthcare informatics research,

24(3):198–206.

Lou, Z., Wang, L., Jiang, K., Wei, Z., and Shen, G. (2020).

Reviews of wearable healthcare systems: Materials,

devices and system integration. Materials Science and

Engineering: R: Reports, 140:100523.

Marin-Pardo, O., Laine, C. M., Rennie, M., Ito, K. L.,

Finley, J., and Liew, S.-L. (2020). A virtual reality

muscle–computer interface for neurorehabilitation in

chronic stroke: A pilot study. Sensors, 20(13):3754.

Melillo, P., Castaldo, R., Sannino, G., Orrico, A., De Pietro,

G., and Pecchia, L. (2015). Wearable technology and

ecg processing for fall risk assessment, prevention and

detection. In 2015 37th Annual International Confer-

ence of the IEEE Engineering in Medicine and Biol-

ogy Society (EMBC), pages 7740–7743. IEEE.

Mina, S. M. (2021). Chronobiological assessment. In Refer-

ence Module in Neuroscience and Biobehavioral Psy-

chology. Elsevier.

Muharrem Adak, Vecteezy (2022).

https://pt.vecteezy.com/foto/9213086-

3d-renderizacao-de-tronco-humano and

https://pt.vecteezy.com/foto/9211711-3d-

renderizacao-de-tronco-humano. Accessed:

25.10.2022.

Qian, Q., Hu, X., Lai, Q., Ng, S., Zheng, Y., and Poon, W.-

S. (2017). Early stroke rehabilitation of the upper limb

assisted with an electromyography-driven neuromus-

cular electrical stimulation-robotic arm. Frontiers in

Neurology, 8.

Queralta, J. P., Gia, T. N., Tenhunen, H., and Westerlund, T.

(2019). Edge-ai in lora-based health monitoring: Fall

detection system with fog computing and lstm recur-

rent neural networks. In 2019 42nd international con-

ference on telecommunications and signal processing

(TSP), pages 601–604. IEEE.

Rakugi, H., Sugimoto, K., Arai, H., Kozaki, K., Matsui, Y.,

Mizukami, K., Ohyagi, Y., Okochi, J., and Akishita,

M. (2022). Statement on falls in long-term care fa-

cilities by the japan geriatrics society and the japan

association of geriatric health services facilities.

Data Acquisition System for a Wearable-Based Fall Prevention

709

Ren, L. and Peng, Y. (2019). Research of fall detection

and fall prevention technologies: A systematic review.

IEEE Access, 7:77702–77722.

Rescio, G., Leone, A., and Siciliano, P. (2018). Supervised

machine learning scheme for electromyography-based

pre-fall detection system. Expert Systems with Appli-

cations, 100:95–105.

Shimmer Discovery in Motion (2022). Shimmer Sensors.

https://shimmersensing.com. Accessed: 25.10.2022.

Siwadamrongpong, W., Chinrungrueng, J., Hasegawa, S.,

and Nantajeewarawat, E. (2022). Fall detection and

prediction based on imu and emg sensors for el-

ders. In 2022 19th International Joint Conference

on Computer Science and Software Engineering (JC-

SSE), pages 1–6. IEEE.

Steinberg, C., Philippon, F., Sanchez, M., Fortier-Poisson,

P., O’Hara, G., Molin, F., Sarrazin, J.-F., Nault, I.,

Blier, L., Roy, K., et al. (2019). A novel wearable de-

vice for continuous ambulatory ecg recording: proof

of concept and assessment of signal quality. Biosen-

sors, 9(1):17.

Wang, X., Ellul, J., and Azzopardi, G. (2020). Elderly fall

detection systems: A literature survey. Frontiers in

Robotics and AI, 7:71.

Wearable Sensing (2022). Multimodal Acquisition Sys-

tems. https://wearablesensing.com/multimodal-2/.

Accessed: 25.10.2022.

Zhou, L., Fischer, E., Tunca, C., Brahms, C. M., Ersoy, C.,

Granacher, U., and Arnrich, B. (2020). How we found

our imu: Guidelines to imu selection and a compar-

ison of seven imus for pervasive healthcare applica-

tions. Sensors, 20(15).

Zhu, L., Mao, G., Su, H., Zhou, Z., Li, W., L

u, X., and

Wang, Z. (2021). A wearable, high-resolution, and

wireless system for multichannel surface electromyo-

graphy detection. IEEE Sensors Journal, 21(8):9937–

9948.

ZiYing, F., Neo, D., Jue, W., YiCheng, Z., and Lam, Y. Y.

(2021). An integrated fall prevention system with

single-channel eeg and emg sensor. In 2021 4th In-

ternational Conference on Circuits, Systems and Sim-

ulation (ICCSS), pages 183–189. IEEE.

WHC 2023 - Special Session on Wearable HealthCare

710