Year
24 / Nº 38 / 2022 /
DOI:
https://doi.org/10.36995/j.recyt.2022.38.002
Electronic device for gait analysis
Dispositivo electrónico para el estudio de la marcha
Alejandro A., Vistorte
Salgado1, *; Fidel E., Hernández Montero1; Gianna,
Arencibia Castellanos1
1- Universidad Tecnológica
de La Habana (CUJAE). Cuba.
* E-mail: alejandrovistorte@gmail.com
Received: 06/08/2021; Accepted: 20/01/2022
Abstract
In
order to automate the determination of gait parameters, a system capable of
acquiring data from inertial units, exploiting their maximum sampling
frequency, was developed. The study of gait is one of the fundamental
indicators for the evaluation of physical performance. It allows the estimation
of the functional deterioration of the elderly in an objective way, so several
tests have been designed to evaluate it. The system developed has two
fundamental elements: an electronic device and a desktop application. The
electronic device has the function of collecting data from the MPU-9255 sensor
using an ESP32 to set the sampling rate, transmitting the data via WiFi to the
computer and monitoring the system's battery. The desktop application allows
the electronic device to be configured and controlled, as well as receiving,
displaying and storing the data. As a result, a prototype capable of operating
at a sampling frequency of 1 kHz was built. Tests carried out on the system
demonstrate its reliability and allow the limits of sampling frequency and
working distance to be set.
Keyworks: Server-Client Application; ESP32; Gait Analysis; MPU-9255; Data
Acquisition Systems.
Resumen
Palabras claves: Aplicación cliente-servidor; ESP32; Estudio de la
marcha; MPU-9255; Sistema de Adquisición de Datos.
1. Introduction
Population ageing is currently one of the most important problems
in society at large. This is a phenomenon caused in particular by the increase
in life expectancy and the decline in the birth rate, which has led to
significant changes in the age structure of the world's population [1].
Cuba has experienced an increase in life expectancy, with 19.8% of
the population now aged 60 and over, with approximately 78 years of life
expectancy. It is also among the oldest countries in the region, with
indicators equivalent to those of developed countries [1].
At the same time, there is a trend towards increasing disability
in the elderly population, but there is an increase in the prevalence of
chronic conditions, which can lead to functional limitation, disability and
dependence [2], increasing health and care costs for both the family and the
health system.
With respect to the above, early assessment of signs and symptoms
related to functional decline would contribute to the detection of frail
elderly, who could receive therapeutic interventions, both in primary and
secondary care, with the aim of minimising the occurrence of adverse outcomes
[3].
Physical performance assessment allows the estimation of
functional impairment in the elderly in an objective, simple, easily
reproducible and also cost-effective manner. Various tests have been designed
to assess physical performance, and the most commonly used parameters include
gait, balance and muscle strength [4].
The traditional method of gait testing involves a person walking a
known distance, counting the number of steps and measuring the elapsed time
with a stopwatch [5]. In these experiments it is assumed that the step lengths
are equal, which is not correct, and the time measurement is subject to errors
associated with the operation of the stopwatch.
In Cuba, there is a precedent for the study of physical
performance indicators in the elderly that has allowed for their correct
characterisation and the identification of frail older adults, in order to
prioritise their specialised care in the National Health System. This work
consists of an evolutionary study of the performance of older adults who attend
Grandparents' Circles in the Plaza de la Revolución municipality. In
particular, gait, among other aspects, is measured, for which data on speed,
step amplitude and cadence are recorded [6].
With the aim of automating this process, in [6] an experimental
electronic device was developed to measure the physical variables necessary to
determine the gait parameters of elderly people. This system consists of an
Arduino Nano, MPU-6050 sensor with accelerometers and gyroscopes in the three
coordinate axes, a MicroSD card reader for an offline working mode and a
Bluetooth module for real-time data transmission. This system was able to store
the samples in a file for later processing, obtaining a sampling frequency of
144 Hz.
The electronic device described above is used in [7] in order to
determine the orientation. For this purpose, a data fusion method using
conventional complementary filters is applied and a new variant of Kalman
filter was proposed. The proposed algorithm showed some improvement over
conventional algorithms, but obtaining the phase using only the gyroscope
output signal from the angular velocity integration does not give accurate
results due to the deviation that accumulates over time. This result can be
improved with the inclusion of a magnetometer.
Based on [7] and with the aim of determining position, in [8] an
algorithm is developed using a windowing method for position estimation by
detecting stability periods in a walking process. The final result in the
application of the algorithm is translated as the stride length. The different
experiments performed were obtained with a certain degree of effectiveness,
indicating the need for sensors, such as the magnetometer, for better accuracy.
It was shown that it is necessary to increase the number of samples per window
to improve the effectiveness of the algorithm, this implies that the sampling
frequency must be increased which is not possible using the existing hardware.
This work aims to develop an electronic device capable of
operating with IMUs (Inertial Measurement Units) containing magnetometer at the
maximum sampling rate.
2. Design requirements and system structure
The
following aspects were taken into account when designing the system:
·
Electronic
device capable of reliably sampling at a maximum frequency of 1 kHz, in order
to exploit the sensor bandwidth to the maximum and to achieve higher accuracy
in the determination of gait parameters.
·
Real-time
data transmission and display.
·
Simple
interface design, capable of setting the system's working frequency and
controlling the data acquisition process.
·
Portable
and lightweight electronic device.
The
system will have two fundamental elements: an electronic device for sample
acquisition and a desktop application for system control, storage,
visualisation and processing of the data obtained. Both parts will communicate
via WiFi without the need to use a separate module because the microcontroller
used has an internal adapter.
3. Electronic device
An
electronic device was developed, which is able to acquire signals from an IMU
which can be stored for further processing. If the device is used in gait
testing, with these signals it is possible to obtain different parameters such
as: step length, number of steps and cadence, among others.
Figure
1 shows the design of this system, which is mainly composed of an ESP32
microcontroller [9], [10], an MPU-9255 sensor [11], [12], a voltage regulator
and a battery.
Figure
1. Design of the electronic device.
In
this design, the ESP32 will be in charge of establishing the sampling frequency
of the sensor signals and their transmission via WiFi to the computer. The
battery will allow portability, while the regulator will allow monitoring to determine
its charge. A button with two functionalities will be used: when not
transmitting, when pressed, the LEDs will show the battery status, and when
transmitting, it will be used to stop the process and wait for new
instructions.
3.1. Software on the microcontroller
The
acquisition system has three main tasks: acquiring data from the sensors,
measuring the battery level and transmitting this data to the PC. Figure 2 shows the flowchart of the
implemented algorithm.
To
meet these specifications, the system starts by initialising the sensor, the
WiFi adapter and will connect to the socket created on the PC by sending the
battery level that powers the system. The program will be waiting for the
command from the PC to start taking data. The received frame will contain some
"header" bytes and the information about the sampling frequency to be
set for sample acquisition; if this is zero, it will indicate that the sampling
process has been stopped by the control system and sampling will stop until
another non-zero sampling frequency (fs) value is received. When a non-zero fs
value is received, a timer is set and activated, which, each time it
interrupts, will take a sample and form the frame to be transmitted.
The
MPU-9250.h library, developed by the programming community to facilitate the
work with the sensor, was used to acquire the samples. The process starts by
configuring the sensor, i.e. the range of the accelerometer, gyroscope and
magnetometer. A study of the range effect of the accelerometer and gyroscope in
measurements of this type using an MPU-6050 (includes accelerometer and
three-axis gyroscope) was performed in [13] and it was determined that for
these scenarios it is necessary for the sensors to work at full scale; for this
reason, these sensors were configured in this way.
While
the interruption caused by the timer allows the samples to be taken, the main
flow of the program will wait for a block of 250 samples. When the block of
samples is built, the header will be set and sent to the PC. The sending of
samples per block is done in order to save battery power as the WiFi adapter
has a high consumption.
Figure 2.
Algorithm implemented in the electronic device.
4. Desktop application
The desktop application was developed in Python
and has 3 main windows (see Figure 3):
·
Main
window: where the graphs of the acceleration, gyroscope and magnetometer values
(3 axes), as well as the temperature are displayed. In addition, it has the
buttons to control the system and to display the configuration menu and the
registration menu.
·
Registration
window: allows new users to enter the system. Data such as identity card (ID),
name, surname and gender are registered. In case a user is already registered,
an alert will be displayed to indicate this.
·
Configuration
window: here the user who is going to run the tests is selected and the values
of the sampling frequency and the number of IMUs to be used are set. In case
the user is not registered, an alert will appear indicating that he/she has to
register before starting the tests.
Figure
3. Desktop application interface.
The
main window is in charge of controlling the system in a general way by giving
instructions to the ESP32 and displaying the graphs of the samples taken,
which, at the end of the process, will be saved in a file.
The
flow of the application is handled by two threads: a main thread in charge of
maintaining the visual interface and controlling the events that are executed
on it; and a secondary thread, using the Threading module (Python module that
allows the concurrent execution of code), to carry out the other functions,
such as graphing, updating the battery level on the screen, detecting the
disconnection of the ESP32, etc. Figure 4 shows the algorithm of the desktop
application.
When
running the application, an access point will be created for the connection to
the ESP32, and then a socket is created on the IP address of the gateway of the
access point, which will run in the background of the application and will
always be waiting for data from the electronic device controlled by the ESP32.
As
the network is of the point-to-multipoint type, i.e., the access point created
to connect the ESP32 can be accessed by other devices, and taking into account
the problem of identity theft caused by the PC where the application is running
being connected to the internet, a Symmetric Key authentication protocol was
implemented [14].
If
the battery does not exceed 5%, it will not be possible to start the process
and an alert will indicate that the current battery must be charged or replaced
by another one. This action is performed to guarantee the correct functioning
of the electronic device and to ensure the reliability of the data. If all
conditions are in place to start the sampling process, i.e. the ESP32 is
connected, its battery level is sufficient and the system has been configured,
the process can be started; a frame with the selected sampling rate is sent and
data reception and display is enabled.
Once
the data collection is completed, the Stop button must be pressed and if the
test was successful, the data is stored in a file.
Figure
4. Algorithm of the desktop application.
5. Results and discussion
The
proposed system was tested in two different scenarios: indoors in a house and
outdoors in an open space. In each of the tests, the system was pushed a bit
further than necessary in order to leave a margin for future needs when
upgrading some of its functionalities or adding some new ones.
The
tests performed checked two fundamental aspects: finding the limit of the
reliable distance for future measurements and determining the maximum frequency
of operation that guarantees the reliability of the data. In general, two types
of tests were designed: transmitting a known value to check the performance of
the system and transmitting real data taken by the sensor.
5.1. Sampling frequency analysis
The
first test performed was to check the work at the sampling frequency of 1 kHz.
First, the magnetometer data were validated. For this, one of the
magnetometer's axes was oriented towards magnetic north and data collection was
started. Then, the sensor was moved in known directions. To validate the
accelerometer and gyroscope data, the sensor was placed in the rest position
and movements in known directions were started, but this time, varying the
acceleration of the movement to check the state of the gyroscope. The results
were as expected. Figure 5 shows the graphs of the tests performed.
Figure
5. System response operating at a sampling
frequency of 1 kHz.
To
determine the maximum sampling frequency at which the system is reliable, the
movement described in Figure 6 was performed. First, the system was started at
a frequency of 1 kHz to determine the system response to this movement, and
then the frequency was increased until the graph began to distort. As a result,
the maximum sampling frequency at which the system operated satisfactorily was
found to be 1.5 kHz. Figure 7 shows the results of tests performed at 1 kHz and
2 kHz, which show the distortion of the graphs obtained when the higher
sampling frequency is applied.
Figure
6. Movement made to determine the maximum sampling
frequency.
Figure
7. Tests performed at sampling frequencies of: a)
1 kHz and b) 2 kHz.
5.2. Working distance limit
A
second test was performed to check the maximum distance at which the electronic
device and the PC can be located in such a way that the communication between
the two points is reliable.
The
working distance was determined in two different scenarios: indoors and
outdoors. For the indoor test, the electronic device was placed at various
positions in the house and at each position, data transmission and graphical
behaviour was started. For this test, a function was added to the software to
determine the amount of data received and thus verify the amount of packets
lost. At each position, 3 tests were performed, while the data transmission was
carried out in 10-second intervals.
Table
1 shows the results of this test, i.e. the number of packets lost at each of
the positions.
Table
1. Indoor test results.
|
Position |
Distance [m] |
Number of lost packets |
||
|
Test 1 |
Test 2 |
Test 3 |
||
|
1 |
7.5 |
0 |
0 |
0 |
|
2 |
5.25 |
0 |
0 |
0 |
|
3 |
7 |
463 |
4876 |
0 |
|
4 |
10 |
10000 |
10000 |
10000 |
According
to Table 1, in positions 1 and 2, the system functioned correctly as no samples
were lost in the tests. In the case of position 4, the connection could not be
established even though the system was moved around the room. In fact, a scan
was carried out with a mobile phone, which also failed to find the access point
created by the computer. In the case of position 3, the results showed high
variability as the system was apparently at the limit of coverage; already in
test 3 all transmitted packets were received.
For
the outdoor test, the dynamics were different. The computer was placed at the
entrance of a building and the system was configured to sample at 1 kHz. The
desktop application was configured in such a way that once it detected that no
data was being received, it would stop the process and display an alert that the
connection had been lost. In the first scenario, there were not many obstacles
between the PC and the electronic device. In the second scenario, there was no
direct visibility due to the presence of bushes. In the third scenario, there
was direct visibility between the electronic device and the PC. Table 2 shows
the distance ranges obtained in each of the scenarios.
Table
2. Outdoor test results.
|
Scenario |
Distance
range [m] |
|
1 |
33 – 35 |
|
2 |
23 – 24 |
|
3 |
49 – 52 |
6. Conclusions
In
this work, a system capable of acquiring, visualising and storing inertial
sensor data by exploiting its bandwidth to the maximum was developed. The
system consists of an electronic device and a desktop application.
The
electronic device takes data at a configured sampling rate from the desktop
application and sends it to the PC. The desktop application is responsible for
displaying the data on the screen and storing it in files for later use. The
data is stored in a structured way to form a database on which data analysis
and computational learning algorithms can be applied.
An
authentication mechanism was added to the system to ensure the security of the
system and to ensure that the desktop application only receives data from the
ESP32.
The
tests carried out on the system show its correct functioning and allowed the
working limits to be established in terms of sampling frequency and distance.
In this way, it was determined that it is possible to operate the system with a
sampling frequency of 1 kHz, thus fulfilling the objectives set for this work.
7. Acknowledgements.
Esta investigación ha recibido financiamiento de la
OGFPI, referencia PN305LH13-050.
References
[1] “ENCUESTA
NACIONAL DE ENVEJECIMIENTO DE LA POBLACIÓN ENE-2017,” Oficina Nacional de
Estadística e Información, Sitio en Actualización. [Online]. Available:
http://www.onei.gob.cu/node/14725. [Accessed: 19-Feb-2020].
[2] J. E. Morley, “Frailty and Sarcopenia: The New Geriatric
Giants,” Rev. Investig. Clin. Organo Hosp. Enfermedades Nutr., vol.
68, no. 2, pp. 59–67, Apr. 2016.
[3] D. García Agustín, S. Morejón, C. de Dios,
and Z. Rodríguez Pérez, “Las pruebas de desempeño físico en el pronóstico de
desenlaces adversos en los ancianos,” MEDISAN, vol. 22, no. 6, pp.
466–470, Jun. 2018.
[4] J. M. Guralnik, L. Ferrucci, E. M. Simonsick, M. E. Salive, and
R. B. Wallace, “Lower-extremity function in persons over the age of 70 years as
a predictor of subsequent disability,” N. Engl. J. Med., vol. 332, no.
9, pp. 556–561, Mar. 1995.
[5] G. Rosas, “Velocidad de la marcha en
ancianos de la comunidad de la ciudad de Pasto,” ResearchGate, 2015.
[6] J. Méndez, “Sistema electrónico experimental
para la medición de variables físicas necesarias en la determinación de
parámetros de marcha en ancianos,”RECyT, vol 33, p. 108, 2020.
[7] G. Arencibia, “Evaluación de métodos de
estimación de orientación, basados en filtro complementario y filtro de Kalman,
usando unidades de medición inercial,” Universidad Tecnológica de La Habana
“José Antonio Echeverría”, CUJAE, La Habana, Cuba, 2019.
[8] A. Perez, “Estimación de la posición
mediante la detección de periodos de estabilidad durante la marcha humana,”
Maestría, Universidad Tecnológica de La Habana “José Antonio Echeverría”,
CUJAE, La Habana, Cuba, 2020.
[9] Espressif Systems, “ESP32 Datasheet.” 2019.
[10] Espressif Systems, “ESP32 Technical Reference Manual.” 2019.
[11] InvenSense, “MPU-9255 Product Specification.” 14-Sep-2014.
[12] InvenSense, “MPU-9255 Register Map.” 10-Oct-2014.
[13] M. Potter, L. Ojeda, N. Perkins, and S. Cain, “Effect of IMU
Design on IMU-Derived Stride Metrics for Running,” MDPI, vol. 19, p. 19,
2019.
[14] Omar G, Abood, and S. K. Guirguis, “A Survey on Cryptography
Algorithms,” Int. J. Sci. Res. Publ., 2018.