Content uploaded by Nur Izdihar Muhd Amir
Author content
All content in this area was uploaded by Nur Izdihar Muhd Amir on Jan 23, 2023
Content may be subject to copyright.
978-1-7281-7523-2/21/$31.00 ©2021 IEEE
Development of Fall Detection Device Using
Accelerometer Sensor
Nur Izdihar Muhd Amir
Ubiquitous Broadband Access
Networks Lab (U-BAN)
Razak Faculty of Technology and
Informatics (FTIR)
Universiti Teknologi Malaysia
Kuala Lumpur, Malaysia
nizdihar4@live.utm.my
Rudzidatul Akmam Dziyauddin
Ubiquitous Broadband Access
Networks Lab (U-BAN)
Razak Faculty of Technology and
Informatics (FTIR)
Universiti Teknologi Malaysia
Kuala Lumpur, Malaysia
rudzidatul.kl@utm.my
Norliza Mohamed
Ubiquitous Broadband Access
Networks Lab (U-BAN)
Razak Faculty of Technology and
Informatics (FTIR)
Universiti Teknologi Malaysia
Kuala Lumpur, Malaysia
norlizam.kl@utm.my
Liza A. Latiff
Ubiquitous Broadband Access Networks Lab (U-BAN)
Razak Faculty of Technology and Informatics (FTIR)
Universiti Teknologi Malaysia
\Kuala Lumpur, Malaysia
liza.kl@utm.my
Nor Syahidatul Nadiah Ismail
Faculty of Computing
Universiti Malaysia Pahang
Pahang, Malaysia
nadiahismail@ump.edu.my
Abstract—This paper presents the development of Fall
Detection System in the form of a wearable device integrating
an ADXL335 accelerometer as a fall detection sensor. The
system means to detect fall occurrence of the wearer and able to
distinguish the activity of daily livings and fall events. In this
study, the fall detection system consists of two-part, the
transmitter and receiver. Transmitter will be attached to the
user’s g arment, then the Arduino microcontroller will process
the acceleration reading from the sensor and transmit the data
via wireless Zigbee networks to the receiver. The receiver which
is connected to a computer will display the acceleration graph
reading through the monitor. The working hardware of
wearable fall detection system in this work is validated through
a comparison of data acquisition with previous research, SisFall.
The results obtained in this work correspond well to SisFall thus
indicate the sensor reliability satisfied the benchmark.
Keywords—Fall Detection System, wearable device, ADXL
335, fall dataset.
I. I
NTRODUCTION
In 2018, the total world population was 7.59 billion people
and according to global population forecast by United Nation,
the human population will continue to grow in the coming
decades and is expected to exceed approximately 10.87 billion
by 2100 [1]. The well congenial living space and environment
amongst the contributions to this increase of life expectancy.
As a result, the number of elderly who are living alone also
increases accordingly [2]. These people are vulnerable to
adverse events such as any health-related issues as well as
accidents resulting in body instability and eventually
hazardous falls. The statistic shows that more than 25% of
people aged over 65 years old experience falls every year and
this figure grows to 32%–42% for those over 70. Especially
those who without caretaker could turn this event into fatality.
Along with the advancement of medical technologies and
healthcare systems [3], researchers are working to aid this
ageing population by developing Fall Detection systems
(FDS) with means to reduce the unfavourable consequences
of falls. Various type of FDS has been introduced throughout
the research. Among the evolving one is wearable-type
detection devices, it is considered as cost-effective and also
less intrusive as the detection method only involve the
movement sensing rather than visual-monitoring. The non-
movement restricted feature of the wearable device also
allows continuous monitoring of the user [4].
To achieve high accuracy detection, integrating powerful
algorithms method are crucial, such as machine learning
(ML). By engaging high-performance detection algorithms,
reliable data is concerned. Based on the literature review, a
state-of-art dataset for fall detection is SisFall [5]. The dataset
was established by researchers in Columbia and the data
collected are from local participants there, by using own-
developed hardware integrating ADXL345 accelerometer
sensor. Currently, SisFall is deemed as the most reliable
publicly available fall detection data. However, FDS in this
work is meant to be developed and used for local elderly in
Malaysia. Thus, the dataset in [5] showed attributes of people
in Columbia may be different from Malaysian, the sensors
reading from both countries might be conflicting in
acceleration steeps and movements. To overcome this
problem, own developed FDS with data collected from local
people is preferred in this work.
Another matter, wearable types of FDS are mostly in the
form of wrist-watch and pendant-looks style. In this work,
sensor located at the torso, particularly chest area is favoured
due to stable reading emitted compare to other places [6]. In
terms of the design, existing fall detector could be improved
to be lighter and smaller for ease of handling [7]. Besides, this
work opted device with round shape design to reduce harmful
event such that when the device accidentally presses onto the
skin while user is doing certain movements, no pointy or sharp
edges could injure the user. Thus, in this paper, we present the
development of a custom design FDS. ADXL335
accelerometer is selected as a fall detection sensor in this work
due to its higher sensitivity compared to ADXL345. The
system consists of two-part, the transmitter (FDS-Tx) and
receiver (FDS-Rx), overall system name as FDS-X. Data
acquisition will be done by FDS-Tx by reading the user’s
acceleration, and then transmit the data over a wireless
medium using XBee module to the FDS-Rx for offline
analysis. The reliability of the hardware working systems is
validated through comparison with SisFall as the benchmark.
Section II explains the system design and architecture of
FDS-X. Then, the development of FDS-X is described in
Section III. Next, the experimental setting is explained in
2021 IEEE 7th International Conference on Smart Instrumentation, Measurement and Applications (ICSIMA2021)
Section IV followed by results and discussion in Section V.
Last but not least, Section VI concludes the paper.
II. S
YSTEM
D
ESIGN AND
A
RCHITECTURE
Fig. 1. illustrates the system architecture of FDS-X, FDS-
Tx is attached to the user, while FDS-Rx is connected to a
computer for data monitoring. Upon start, FDS-Tx will
transmit data to FDS-Rx wirelessly and FDS-Rx will display
the data in sequential graph on the computer connected.
Fig. 1. Illustration of FDS-X system architecture
Further explanation of FDS-X working system is as shown
in Fig. 2. block diagrams. Three main components involved in
the system are ADXL 335 accelerometer sensor, Arduino Pro
mini microcontroller and XBee Pro module. In FDS-Tx part,
Arduino will process the data reading from the accelerometer
sensor, then trigger the wireless communication module,
XBee to transmit the data to receiver board, FDS-Rx. In FDS-
Rx, XBee module will capture the wireless transmission data
and then Arduino that connected to the computer will display
the data reading in the form of sequential graph on the
monitoring screen.
Fig. 2. Block diagram of FDS-X working system
The process flow in Fig. 3. explained the development
details of the system. First step, the components specifications
and their working system are studied, particularly the
specifications of ADXL335 Accelerometer. Initial
connections between components are performed on the
prototype board before moving to software for designing the
Printed Circuit Board (PCB). For the development, the PCB
is then soldered. Afterward, the soldered PCB working
condition is justified before it being used for data acquisition
experiment. The data collected will undergone comparative
validation with SisFall work.
Fig. 3. Flow chart of the hardware development
A. ADXL335 sensor
ADXL335 sensor has a precise and full sensing range of
±3g. Note that the output is ratiometric, thus, sensitivity varies
proportionally to the supply voltage. Since there is 3.3V
regulator available onboard, 0g measurement output is
nominally equal to half of the 3.3V supply voltage (1.65V), -
3g is at 0v and 3g is at 3.3V with full scaling in between. To
interpret raw data reading into G-force unit (g), the equation
on the code program is as follow:
= (, , ,
−3000, 3000)
(1)
The data is initially in long type to allow big size data
storage up to 8 bytes. The data named as xScale mapped the
x-axis (Xaxis) raw minimum data reading (Rawmin) and
maximum data reading (Rawmax) onto its corresponding
milli-Gs values. The term map in this context means direct
conversion of x-axis’s raw values reading into gravitational
force precise values. To obtain Rawmin is when x-axis of
accelerometer in -1g position which in this work is 133 and
Rawmax
is when in +1g position that gives 528 reading, these
values then will be converted into -3000 and 3000 respectively
for precise conversion before rescaling to fractional Gs:
=
1000.0
(2)
Start
Study ADXL335
specifications
Design FDS-Tx using
EasyEDA software
Validate and compare
data with SisFall
End
Troubleshoot
No
Conduct
experimental setup
Yes
Develop FDS-X
hardware
Verify hardware using
multimeter
ADXL
335
Arduino
Pro mini
XBee
PRO
Display
data
Arduino
Pro mini
XBee
PRO
FDS-Rx
FDS-Tx
2021 IEEE 7th International Conference on Smart Instrumentation, Measurement and Applications (ICSIMA2021)
Equation (2) shows the final conversion of acceleration
reading in g-unit. The data represented in float type, 4 bytes
size that enough to store and display precise reading of the
converted acceleration and help ease the data analysis process.
Equation (1) also applicable for finding y-axis and z-axis data,
just change the into or and into
or . The same goes for equation (2), change
and into respective y-axis or z-axis
acceleration and scale to get the output.
a) Three-axis Pointing Direction
Fig. 4. The direction of pointing axes of the sensor with change in user’s
body position
In fall data involve acceleration reading that consists of
three-axis named x-axis, y-axis and z-axis. In recording and
analysis of data, the pointing of axes is important to identify
the type of movement and body position. In this study, the
position of FDS-Tx which embedded with ADXL335 sensor
is attached to the user with a fixed direction of each three-axis.
Fig. 4. visualise the fixed position of the sensor on user with
direction of pointing axes when user in different positions.
B. Arduino Pro mini
Fig. 5. Arduino Pro mini
As the brain of the system, Pro mini [8] is chosen mainly
due to its small figure and fairly available features that suit this
work requisite. In this work, Arduino Pro mini is a 5V type
running the 16MHz bootloader. The output voltage is fairly
enough to supply power for both ADXL335 sensor and XBee
module which each require a typical supply voltage of 3.3V
with a voltage regulator. The wiring connection between
components will be explained further in Section III.
III. FDS-X
D
EVELOPMENT
The development of FDS-X hardware covered both FDS-
Tx and FDS-Rx parts.
A. Transmitter (FDS-Tx)
FDS-Tx was designed using EasyEDA software. The
schematic diagram of the components wiring connections for
FDS-Tx is as shown in Fig. 9. Other than Arduino Pro mini
microcontroller, XBee Pro and ADXL 335 accelerometer
sensor, components of FDS-Tx also consist of 5V voltage
regulator as a filter from high voltage power supply that may
damage the components on-board. Schematic diagram then
converted into PCB design; design of the desired PCB is
visualised into 3D view before submitting for final printing.
Fig. 6. Front (left) and back (right) view of fabricated PCB of FDS-Tx
FDS-Tx was designed to look simple to reduce the size of
the PCB as well as the power consumption. As seen in Fig. 6.,
the fabricated PCB for FDS-Tx is designed in a round shape
to reduce any inconvenience to the user as the normal PCB’s
square or rectangle board with pointing edge design are likely
to cause harm by jabbing into the skin when doing certain
movements. After the PCB is received, the next step is
soldering. Soldered FDS-Tx as seen in Fig. 7. Some soldered
components cannot be seen physically as they are hidden
underneath bigger component. The purpose of such
arrangement is to best utilise the available space on PCB.
Fig. 7. Final look of FDS-Tx, front(left) and back after soldered
Next, to ensure the correct connection, each component
will be validated using multimeter such as in Fig. 8. After
connecting the power supply, the voltage difference is
measured for the main four components, Arduino Pro mini,
ADXL 335, XBee module and output of voltage regulator to
make sure the components functioning at optimal voltage.
Fig. 8. Validating component operation voltage using multimeter
2021 IEEE 7th International Conference on Smart Instrumentation, Measurement and Applications (ICSIMA2021)
Fig. 9. Schematic diagram of FDS-Tx
The Arduino Pro mini is pre-programmed, the working
function of ADXL 335 is validated based on the datasheet.
The communication modules, XBee PRO S1 for both FDS-
TX and FDS-Rx also has been pre-programmed by XCTU
software, and tested their connection via the simulation option
available in the XCTU software. Next, the connection test of
the communication module from FDS-Tx to FDS-Rx. Prior to
that, the details of research design for receiver are discussed
in the following sub-section.
B. Receiver (FDS-Rx)
The receiver board or FDS-Rx consist of the same
microcontroller as the FDS-Tx, Arduino Pro mini,
communication module XBee PRO and microSD slot as
additional storage to store the obtained ADXL data. FDS-Rx
PCB board is designed by previous work in [9], the similarity
in the list of components to be used in this project for the
receiver part with the mentioned work are being used for good
by incorporating the same design of the PCB board as shown
in Fig. 10.
Fig. 10. The Fabricated FDS-Rx PCB; front(left) and back view
The same next step as FDS-Tx, the PCB board then being
soldered. The connection between components are tested, and
validated the FDS-Rx working function by testing the
connection between XBee module from FDS-TX and XBee
module of FDS-Rx. After the sensor’s data reading is received
accurately by FDS-Rx, both FDS-Tx and FDS-Rx has
successfully function as a complete fall detection data
acquisition device. To proceed to the next step of data
acquisition, details are in the next section.
IV. E
XPERIMENTAL
S
ETTING
To validate the hardware functioning system, the
experiment must be carried out. The data obtained from the
experiment then analysed. Further details are explained in the
next respective Sections.
A. Setup
For preparation, the experiment follow a protocol that
required test scripts for reference of activities to perform, a
number of participants and a venue with some equipment to
conduct the experiment.
a) Test Script
TABLE I and TABLE II list the ADLs and falls events
respectively. These are the four activities selected to validate
the hardware working reliability, they are equally divided into
two ADLs and two fall events. The difference in activities
duration is due to the period during fall occurs rapidly,
therefore 15 seconds is revised enough to cover before, during
and after a fall. Compared to the easier ADLs activities which
decided to take 100 seconds is solely for data repository
purpose and might be useful for further enhancement of the
system.
b) Volunteers
Participants in this study are three individuals. Details of
the participants are recorded in terms of gender, weight and
height as shown in TABLE III.
TABLE I. ACTIVITIES OF DAILY LIVING (ADL) TRANSCRIPT
Code Activity Trials Duration (s)
D01 Walking slowly 1 100
D02 Walking quickly 1 100
TABLE II. FALLS TRANSCRIPT
Code Activity Trials Duration (s)
F01 Fall forward while walking caused
by a slip 3 15
F02 Fall backward while walking
caused by a slip 3 15
2021 IEEE 7th International Conference on Smart Instrumentation, Measurement and Applications (ICSIMA2021)
Participants are given a unique subject name each, such
A01 indicate adult number one, A02 and A03 are adults
number two and three respectively. These data are recorded
for reference factors that might influence the acceleration
reading of each participant.
c) Location and Equipment
User will wear the FDS-Tx on their right chest, placing
sensor on the chest is less obtrusive compare to the other
favourable place: waist, both waist and chest hold the same
status of steady sensor reading [10], nevertheless, chest is
chosen based on the ease of placing it onto the garment. The
position of the device is such the backside facing the user and
the front side which has transmission module facing out from
the user. The orientation of the sensor (see Fig. 11.) presents
the positive z-axis toward user direction, the positive y-axis to
the gravity direction and the x-axis pointing to the user’s left
side. The experiment is held in Universiti Teknologi Malaysia.
Fig. 11. Location of FDS-Tx to user
B. Sensors reading Comparison
Sample data collected from validation test of FDS-X and
equivalent SisFall dataset are compared in the same graph for
clear analysis. The subject of comparison is acceleration
reading pattern between these two works. The similarities and
distinctions results will help identify leading factors of certain
behaviour of hardware since both works are using own-
customised hardware. Further results and discussion based on
the graphs are explained in the next section.
V.
R
ESULTS
A
ND
D
ISCUSSION
A. Activities of Daily Living (ADLs)
Fig. 12. and Fig. 13. tabulate the graphical readings of
accelerometer sensor of selected ADLs for FDS-X validation
process, two activities of ADLs performed were D01-Walking
slowly and D02-Walking quickly. The experiment was
performed several times for each activity to ensure stable
readings. Raw data from the sensor reading are converted into
gravitational force unit (g) to ease the graph reading.
Calibration of sensor was performed following the datasheet
by adjusting value on the code program as explained in
Section II.A. during the experiment process until the desired
outcome is reached.
Each of Fig. 12. and Fig. 13. is tabulated with readings
from this work which labelled as X, Y, Z, and the comparative
reference, SisFall Dataset, labelled SF-X, SF-Y, SF-Z. The
gravitational reading indicates the pointing direction of each
axis: x-axis, y-axis, and z-axis during experiment.
Fig. 12. Acceleration reading for D01-Walking slowly
Fig. 13. Acceleration reading for D02-Walking quickly
In Fig. 12. and Fig. 13., the graph reading shows all the
three-axis presented were stabilised around the same axis
throughout the whole duration. Since walking is a repetitive
activity, no significant changes occur thus resulting in almost
steady reading. The same reading pattern can be seen in both
this work and SisFall.
B. Fall Events
The same procedure was applied for fall events
experiment. Two activities performed were F01-Fall forward
while walking caused by a slip and F02-Fall backwards while
walking caused by a slip. Other specifications as described in
previous sub-Section. Fig. 14. and Fig. 15. shows the
occurrence of fall. The moment of falls was labelled on the
graph.
In Fig. 14. and Fig. 15., changing of axis occur during fall
process indicating the changing position of the wearer during
fall from vertical to horizontal. Moreover, the axis reading in
both Fig. 14. and Fig. 15. form a spark before instant changing
of directions. Differences observed from both works are the
spark in Fall Events graph from SisFall are more protrusive
than in this work, probably due to the different physical traits
of subjects that perform this experiment from both works.
Some The differences in people characteristics also contribute
to the variety of data reading patterns. The spark demonstrated
hard fall of the wearer that causes high acceleration when
TABLE III. DETAIL OF PARTICIPANTS
Subject Age Height(cm) Weight(kg) Gender
A01 22 167 70 F
A02 26 164 57 F
A03 29 170 77 M
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Acceleration (g) vs time (s)
SF-X SF-Y SF-Z X Y Z
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
15
Acceleration (g) vs time (s)
SF-X SF-Y SF-Z X Y Z
x
z
y
2021 IEEE 7th International Conference on Smart Instrumentation, Measurement and Applications (ICSIMA2021)
slipping occur. Both graphs of Fall Events end with constant
reading after the fall, signalling subjects were static or lying
flat on the floor.
Fig. 14. Acceleration reading for F01- Fall forward while walking caused by
a slip
Fig. 15. Acceleration reading for F02-Fall backwards while walking caused
by a slip
Note that since the direction of fall in both figures is
opposite of each other, thus final constant reading of axes in
both graphs also shows different settling readings. Besides, in
this work, the final reading after falls in both Fig. 14. and Fig.
15. is different compare to SisFall. The significance in final
constant reading can be seen in z-axis wherein Fig. 14., the
volunteers fall forward thus z-axis final reading in this work
shows 1g whereas SisFall shows -1g. Vice versa for Fig. 15.
when volunteers fall backwards. This due to the placement of
sensor on the wearer as mentioned in IV.A and the fixed
pointing axes direction of sensor as described in II.A which
are quite different compare to SisFall.
Another issue, the graph pattern in SisFall (200Hz)
tabulate more reading in second compared to this work
(100Hz). Nevertheless, as long as the data reading is
continuous throughout the duration, differences in working
frequency do not affect the validation process of this work, the
main purpose is to match the graph reading pattern and work
out for any differences on the comparative reference
benchmark.
Furthermore, it can be seen that the falls in both Fig. 14.
and Fig. 15. for this work and SisFall occur at different times
because there is no restriction in the exact time of fall during
experimenting, participants can act the fall within the duration
given which is in this case anytime within 15 seconds.
Moreover, to train ML, the data will be further divided into
certain sizes of windows with fall or ADLs labelled on them
that will enable ML to distinguish between fall and ADLs
easily.
Since this experiment validates the hardware of FDS-X
working system, the data presentation also characterises its
feasibility to be used for datasets in ML. Thus, based on the
fall graphs, commonly, to recognize a fall, a noticeable spark
will form followed by constant data reading afterwards, this
trait will be used to train ML to identify fall for fall events of
F01 and F02 type. Nevertheless, different fall gives different
characteristic of graph reading such as fall forward and fall
backwards have different final constants reading for specifics
axes.
Overall, all the reading presented satisfied the desired
output as referred to SisFall protocol. Each activity was
performed several times to obtain constant reading from the
sensor. The constant reading indicates the reliability of the
sensor device.
VI. C
ONCLUSION
This paper discussed the development of wearable Fall
Detection System, which integrating ADXL335
accelerometer sensor as data collector. The data transmission
occurs wirelessly from FDS-Tx to FDS-Rx via the XBee PRO
modules. Experiment on data acquisition was performed with
three volunteers according to protocol stated. The working
hardware was validated with the SisFall dataset. In addition to
that, this work wearable device has agreed with the reference
work. Future work is to propose ML model for the proposed
FDS system.
A
CKNOWLEDGE
This research was funded by Universiti Teknologi
Malaysia (Vot number: R.K130000.7356.4B443) and the
Universiti Malaysia Pahang under Internal Research Grant
(RDU192304) for financial support and advice.
R
EFERENCES
[1] United Nations Department of Economic and Social Affairs Population
Division, World Population Prospects 2019:, in Highlights. 2019.
[2] Jahanjoo, A., M. Naderan, and M.J. Rashti, Detection and multi-class
classification of falling in elderly people by deep belief network
algorithms. Journal of Ambient Intelligence and Humanized
Computing, 2020: p. 1-21.
[3] Ramachandran, A. and A. Karuppiah, A survey on recent advances in
wearable fall detection systems. BioMed research international, 2020.
2020.
[4] Luna-Perejón, F., M.J. Domínguez-Morales, and A. Civit-Balcells,
Wearable fall detector using recurrent neural networks. Sensors, 2019.
19(22): p. 4885.
[5] Sucerquia, A., J.D. López, and J.F. Vargas-Bonilla, SisFall: A fall and
movement dataset. Sensors, 2017. 17(1): p. 198.
[6] Özdemir, A.T., An analysis on sensor locations of the human body for
wearable fall detection devices: Principles and practice. Sensors, 2016.
16(8): p. 1161.
[7] Jung, S., et al., Wearable fall detector using integrated sensors and
energy devices. Scientific reports, 2015. 5(1): p. 1-9.
[8] Sebastian, S., A SURVEY ABOUT POWER CONSUMPTION FOR
ARDUINO. Carpathian Journal of Electrical Engineering, 2020. 14(1):
p. 105-109.
[9] Ismail, M.I.M., et al., An RSSI-based Wireless Sensor Node
Localisation using Trilateration and Multilateration Methods for
Outdoor Environment. arXiv preprint arXiv:1912.07801, 2019.
[10] Krupitzer, C., et al. Hips do lie! A position-aware mobile fall detection
system. in 2018 IEEE International Conference on Pervasive
Computing and Communications (PerCom). 2018. IEEE.
-13
-11
-9
-7
-5
-3
-1
1
3
5
7
9
11
13
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
15
Acceleration(g)vs time (s)
SF-X SF-Y SF-Z x y z
Fall
(FDS-X)
-4
-3
-2
-1
0
1
2
3
4
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
15
Acceleration(g) vs time (s)
SF-X SF-Y SF-Z x y z
Fall
(FDS-X)
Fall
(SisFall)
Fall
(SisFall)