Content uploaded by Sampath Dakshina Murthy Achanta
Author content
All content in this area was uploaded by Sampath Dakshina Murthy Achanta on Oct 26, 2020
Content may be subject to copyright.
A wireless IOT system towards
gait detection technique
using FSR sensor and
wearable IOT devices
Sampath Dakshina Murthy Achanta and Karthikeyan T.
Department of Electronics and Communication Engineering,
K L University, Guntur, India, and
Vinoth Kanna R.
Department of Electronics and Communication Engineering,
Vivekanandha College of Engineering for Women,
Tiruchengode, India
Abstract
Purpose –The recent advancement in gait analysis combines internet of things that provides better
observations of person living behavior. The biomechanical model used for elderly and physically challenged
persons is related to gait-related parameters, and the accuracy of the existing systems significantly varies
according to different person abilities and their challenges. The paper aims to discuss these issues.
Design/methodology/approach –Deployment of wearable sensors in gait analysis provides a better
solution while tracking the changes of the personal style, and this proposed model uses an electronics system
using force sensing resistor and body sensors.
Findings –Experimental results provide an average gait recognition of 95 percent compared to the existing
neural network-based gait analysis model based on the walking speeds and threshold values.
Originality/value –The sensors are used to monitor and update the predicted values of a person for
analysis. Using IoT a communication process is performed in the research work by identifying a physically
challenged person even in crowded areas.
Keywords Gait analysis, Body wearable sensors, Force sensing resistors
Paper type Research paper
Introduction
The interest in analyzing the physically challenged person in a particular area is to
provide preference to that person and predict the nature of the walking style that will
provide an immediate medical facility when they need. Many researchers’works in human
identification systems, face recognition and gait recognition model for identifying a
person but compared to human gait analysis had evident advantages over the others since
it works without any interruption or interface of the subject cooperation. The basic model
of gait analysis uses human walking styles so that it can be used in detection process more
effectively from a long distance and can be used in surveillance application also in the
public places. Figure 1 presents a basic walking style of a person from load response to
pre-swing, while they walk the nodes used in walking styles are pointed in Figure 1(b),
which is used for analysis.
The consequences in the analysis include the mobility, disability and, in particular, the
disability in walking involves effects in a person walking style. The risk factor involving a
person can be categorized into three categories such as fall detection, fall prediction, and
fall prevention system. The fall detection system is used to notify the user in the case of fall
occurrence, and it cannot provide any suggestion or notification to help users before they fall.
The fall in this model is compulsory, and it just provides the information, so it is less effective
than the other models. Many fall prediction models are available to estimate the nature of fall
International Journal of Intelligent
Unmanned Systems
Vol. 8 No. 1, 2020
pp. 43-54
© Emerald Publishing Limited
2049-6427
DOI 10.1108/IJIUS-01-2019-0005
Received 19 January 2019
Revised 16 February 2019
Accepted 16 March 2019
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/2049-6427.htm
43
Wireless
IOT system
and the risk after fall based on the clinical assessment of the gait and other factors. Using some
threshold value, the person is categorized into fallers and non-fallers. Real-time systems need a
continuous monitoring system for the user so that a prevention system predicts the fall in real
time and alerts a person, thereby reducing the health consequences after fall. The prediction
and prevention system has common terms since it collects the data from the sensors and
compute a feature set before the risk of fall occurs through classification algorithms. The use of
sensors in investigating the parameters is common and it is considered as an easy process as it
collects data without the need of any other factors, unlike the other systems such as camera
and sound. Accelerometers and gyroscopes are considered as some common systems that
provide data based on the nature of the person. Embedding smart phones to the feature
extraction process has increases the ability of system in IoT, which has made it much
convenient for the users.
The presented fall prediction system uses posture variables to analyze the movement
of a person and a mathematical calculation to predict the fall. Since the fall time is very
less and the system must compute the mathematical process and it must alert the user so
that using complex algorithms in analyzing the performance reduces the system
efficiency. Therefore, an effective algorithm is needed for analyzing the performance with
a better accuracy and time reflex nature for consideration. A significant development of
IoT in many application helps to improve the communication; the proposed model
handles nonlinear motion of a person using gait analysis and combining IoT through
the body wearable sensors and force sensors that is integrated into the mobile
application. Handling the user acceleration, it predicts the behavior of gait through
threshold values for a person with abnormal walking style. The research work is
organized as follows: The second section describes about the related works in gait
analysis and IoT models. The third section provides the proposed model mathematical
formulation and the fourth section provides the summary about the experimental results
along with discussion.
Related works
Many research works have been done and the process still continues; this section provides a
summary about the existing research works in gait analysis and IoT applications. Research
work (Qi et al., 2018) describes about the sensor-based activity recognition system in health
care applications related to IoT. The process involved in the research work combines the
health care and IoT so that an effective data transfer is possible from one place to another
place and the human cost is reduced. The usage of cloud networks in medical applications
can provide suitable medical treatment from one place to another place without the need of
physical presence. Research work (Qi et al., 2017) describes about the various health care
systems related to IoT as a survey. It provides a keen knowledge about integrating health
Load
Response
Heelstrike HeelstrikeToe off
Single Support
Stance Phase Swing Phase
Pre-
Swing
Figure 1.
(a) Walking style
of person; (b)
nodes points
44
IJIUS
8,1
care applications into IoT that is related to this proposed research model. Implementing gait
analysis in IoT is based on this health care applications, and this literature provides the
merits and demerits in the existing research works. The literature (Godfrey, 2017) reports
about the issues present in the gait analysis prediction system and the advantage of using
wearable models for an independent living of elderly persons. The monitoring system used
in the model works based on the electro-mechanical movement based values, and there is a
threshold level for each process. If the wearable sensor notices the level crossing the
threshold value, then a warning or alert is provided to the person so that they can change
the current process, which is much suitable for aged persons.
Research work (Gadaleta and Rossi, 2018) provides a smart phone-based recognition
model that has recognition process obtained through neural network. The process involves
smart phone application that monitors the data from the person and provides suitable
communication or alert to the connected persons. The literature (Llamas et al., 2017)
provides a sensor-based platform in gait recognition using an open source hardware. The
proposed work utilizes the sensors for accessing the information about the person walking
behavior so that analysis identifies the person. As an alternative to normal authentication
system the gait nature is used for the authentication process in some applications. Research
works (Nweke et al., 2018; Llamas et al., 2016) describe about the wearable sensor application
interconnected to mobile for human activity monitoring. The process has multiple sensors
and each provides data from each location so that all of them are consolidated into
application in the mobile providing an alert on unusual detection. Research work (Ing-Jr and
Chang, 2017) provides details about the recognition system based on the gesture commands.
The extraction process based on the features compares the present data and the existing
data for better analysis and it also provides a better recognition system using data stream.
Theliterature(Zhanget al., 2018) also describes about the gait analysis model based on
sensors as a health care application. In order to provide a smart heal application, the proposed
model uses sensor signals for processing and then a predictive analysis generates a report to the
corresponding user. A research model (Gravina et al., 2017) reports about the issues and solution
in a cloud-based human activity process in mobile computing. The utilization of cloud in human
detection system increases the accuracy and also reduces the errors. The literature (García et al.,
2019) describes about the health care application for detecting the stroke, and the results are
transmitted through cloud network. The process provides a simple detection mechanism, and
the issues present in this research work do not provide any prediction analysis. The literature
(Gavrilova et al., 2018) reports about the sensor-based activity recognition application for
cognitive systems. The process differs from other activity monitoring systems. Due to its
cognitive nature, the application is limited to a specific monitoring process. Research work
(Stack et al., 2018) describes about the wearable sensors and video-based identification system
for the person having Parkinson’s diseases. The findings from article (Kaur and Sood, 2017;
Paraskevopoulos et al., 2017) discuss about the resource constrained routing and scheduling in
the service industry in allocation of scarce resources to various locations. The survey paper
discusses various issues in the constrained routing of various authors and highlights the
benefits of scheduling algorithms with respect to qualification, services and other modules of
problems. Based on the above survey work, it is observed that earlier gait recognition model
performs only either fall detection or fall prediction. They lag in efficiency if it locates the person
in crowded places so that the accuracy level also decreases. The proposed model provides a
better classification algorithm and an optimized routing model for fall prediction and detection
analysis along with the communication model through Internet of things (IoT).
Proposed work
The proposed work includes an ANFIS model for classification and ant colony optimization
for optimizing the efficient routing for the IoT network. This section provides the
45
Wireless
IOT system
mathematical of ANFIS and ACO and how it is related to the proposed work. Initially, the
sensors and the data are used to identify the person with disabilities, and the information
related to the user is passed into the cloud network for the transmission process. A simple
illustration of an ANFIS model is depicted in Figure 2, and it includes five layers of
architecture. The input layer and the second layer are used to validate the information
received through the sensors and based on the threshold value that is set in
the system model, and the classification is performed by processing the data set.
This process is performed in the consecutive layers, and the final result is a summarization
of all over data from the person wearable sensors and pressure sensors.
The mathematical analysis is performed based on IF and THEN rules, and it is given as:
Rule 1 :If xis A1
ðÞand yis A1
ðÞthen f1p1xþq1yþr1
ðÞ
Rule 2 :If xis A2and yisB
2
ðÞðthen f2p2xþq2yþr2
ðÞ
):(1)
The independent information from the sources is considered as xfor the body wearable sensors
and yfor the pressure sensor, and the fuzzy set is given as A
i
and B
i
and the parameters are
defined based on the prediction procedure of fuzzy governance. In this a five-level neuron
an identical layer which includes functional relations andan adaptive input layer considered as
node terminals as the data collected from the sensors. The result after the fuzzy relationship is
given as:
r1
i¼mAixðÞi¼1;2;(2)
r1
i¼mBi2yðÞi¼3;4:(3)
The fuzzy membership function is formed in bell-shaped function, and the values are
presumed as mAixðÞ,mBi2yðÞ. The value of mAixðÞis given as:
mAixðÞ¼ 1
1þxci
ai
2
bi:(4)
where a
i
,b
i
and c
i
are the factors of the member ship function that is used in bell-shaped
function, and the second layer, third layer and fifth layer in the ANFIS are given as:
L2
i¼Mi¼mAixðÞmBiyðÞ;(5)
Layer 1
Layer 2 Layer 3
Layer 4
Layer 5
x
x
x
y
y
y
A1
A2
N
N
w1
w2 f2
w1 f1
f
w1
w2
w2
B1
B2
Figure 2.
Illustration of five
layer ANFIS model
46
IJIUS
8,1
L3
i¼wi¼wi
w1þw2
;(6)
L4
i¼~
wifi¼~
wipixþqiyþri
ðÞ:(7)
The final result of the model is a summary of the all the layer results and it is given as:
L5
i¼X
2
i¼1
~
wifi¼P2
i¼1oifi
o1þo2
:(8)
The membership function defining the ANFIS model and the fuzzification is used to
produce data from the inputs that is used to train the propagation network in the
proposed classification model. The architecture includes the entire layer and its
identical outputs in practical classification cases. The factors that can be modifiable
a
i
,b
i
and c
i
are linked with the input function and the polynomial function. The learning
process in the ANFIS is used to build the results based on the data from the sensors
such as body wearable sensors and foot pressure sensors. Once the classification
process is completed the data from the prediction system shares the information
into the cloud so that the person with disability can be identified easily and a preference
will be provided along with the information sharing system to the related persons.
For efficient communication in IoT, the system must provide end-to-end delivery
system from one network to other network through cloud. Many routing models are
available in present situation, and the proposed model uses ant colony optimization
for efficient routing in IoT. The reason to choose ant colony optimization in efficient
routing is due to the behavior of ant path selection for the food searching process.
The mathematical model is obtained based on the nature of ants pheromones deposit
methodology to obtain shortest path to reach the destination. The mathematical model of
ACOisilluminatedbyconsideringthepathsxand yobtained from the classification
results.ThenumberofpathsoftheantsisconsideredalongwiththetimestepfunctionT
and the probability of path choosing is considered as Px and Py. The step function-based
probability of choosing the correct path is given as:
Tþ1ðÞ¼ zþnx TðÞ
½
zþnx TðÞ
½
þzþny TðÞ
½½
;(9)
Tþ1ðÞ¼1Py Tþ1ðÞ:(10)
The degree of attraction is given as zand the bias toward pheromone deposits provides the
decision process, and the updating process of new position from the existing position is
given based on the random distribution process as:
lj;t
!¼l
!
i;tþReal Rand int 1;1
½
;(11)
lj;t
!Tþ1ðÞ¼l
!
i;tTðÞþlj;t
!TðÞl
!
i;tTðÞ
Rand int 0;1
½
:(12)
The process is repeated until it reaches the shortest path from the nest to food resource
location, and for every displacement, the present position and the new position are updated.
47
Wireless
IOT system
The step movement is given as:
lj;t
!Tþ1ðÞ¼l
!
i;tTðÞþ lj;t
!TðÞl
!
i;tTðÞ
ji;t
!
Rand int 0;1
½
:(13)
li
!position of AF ifollows the best lb
!. Figure 3 gives an illustration of ACO model path
selection strategy that is related in the proposed model routing process.
The standard process in the IoT includes the size of information packet that is used in the
routing process and the parameters used in finding the shortest path. The routing table is
used in the optimization model in order to maintain the changes the routing process also to
analyze the quality of services.
Figure 4 gives the proposed model process starting from the data from the sensors
to the ANFIS classification model and the ACO optimization model. The proposed model
provides efficient routing based on the factors such as route discovery and maintenance.
A short summary of the proposed model routing efficiency is described as follows.
xyChosen one
N
FFF
N
N
123
Figure 3.
Illustration of ant
colony optimization
Feature
Extraction of
Data Base
Image
Database
Feature
Database
Comparison
Extraction
from Video
Frame
Video
Input
Data from
Pressure
Sensors
ANFIS
Classification
Model
Data from
Body Wearable
Sensors Person
Status
ACO based
Routing model
Cloud
OptimizationClassificationGait analysis
Figure 4.
Proposed
ANFIS-ACO model
48
IJIUS
8,1
Algorithm for efficient routing in IoT Based on ant colony optimization
Result and discussion
The proposed model is experimented in MATLAB 2014 and the data set is used for gait
recognition that is obtained from the person with disability, and the movement of twenty
participants includes ten males and ten females and the experimentation is held at various
locations. The participants are aged between 45 and 60 years and the experimentation is
computed based on the observations and data collected. There are five sensors in the each
leg nodes, and a pressure sensor is used in the foot base. The data which are collected from
the sensors are processed through a processing unit and then transmitted to the
classification section for analysis. After predicting the actual status of the person, an alert is
provided to the user and also the other persons if there is chance to meet any critical
situations. Table I provides the characteristic summary used in the proposed model
classification process and optimization process.
S. No. Type Unhealthy gait
1 Gender 10 males 10 females
2 Age 45–60
3 Height 1.6–1.7
4 Weight 45–80
Table I.
Subject characteristics
summary
49
Wireless
IOT system
Placing sensors on the person plays a vital role while obtaining the desired classification
output. The process of placing sensors on the foot and body is illustrated in Figure 5(a)
and (b), respectively. The foot section is sub-divided into three sections as forefoot,
mid-foot and hind-foot. Sensors are placed in all the three sections, and the in particular
mid-foot contains the bones to support the walking style. The hind foot includes the heel
bone and direct ankle joint. The body wearable sensors are placed on the legs and hip area
and also in the hangs. Challenged persons use their sticks for balancing these sensors that
are in contact with the physical movement sensors providing the information based on the
movement and sending the collected information from five different places along with
the information from the foot sensors. All the eight types of information are given to the
ANFIS model for classification process.
The classification results from the ANFIS model are given in Figure 6 based on the
sensor data from foot pressure. It is observed that the detection range varies for each time as
the person movement changes and a sharp detection is obtained in the time period of 8 s.
The output of unhealthy gait and their feet distance between the legs is observed
in Figure 7. The range varies from one point to another point with short duration of time
3
2
4
1
5
Figure 5.
(a) Placement of
foot sensor; (b)
wearable sensor
0
18
16
14
12
10
8
6
4
2
0246810
Time (sec)
Angle Detected (Degree)
Foot Sensor
Figure 6.
Analysis of
unhealthy gait using
foot sensor data
50
IJIUS
8,1
since the distance between the physically disabled person legs are close to each other and
the reflex action takes place immediately from one position to another position.
Using the sensors in the proposed model, the recognition rate is measured in percentage
and the sensors are placed in right arm, left arm, left leg, right leg, pelvis region. It is
observed that the detection rate reaches maximum for the right leg and left leg compared to
other portions (Figure 8).
Based on the walking speed, the detection rate changes and it is plotted in Figure 9 for
the proposed model and the neural network model. It is observed that the proposed
model has better recognition rate than the neural network model as the speed increases.
The performance of neural network model lags as the speed crosses the 6–8 km/h.
The recognition rate based on the gait cycles is described in Figure 10 compared with the
neural network model. The plot describes that the proposed model has better recognition
rate than the existing model that is 20 percent higher recognition rate (Figure 11).
0
50
40
30
20
Distance Detected
10
0246810
Time (sec)
Figure 7.
Analysis of unhealthy
gait Feet distance
between legs
100
90
70
60
50
Detection Rate (%)
40
30
10
0
20
Left Arm Right Arm Left Leg Right Leg Pelvis Region
Galt Recognition Rate
80
Figure 8.
Gait recognition rate
for different
placements of the
sensor nodes
51
Wireless
IOT system
Conclusion
The proposed model is a combination of adaptive neuro fuzzy inference system and ant
colony optimization for a better recognition of physically disabled person identification
system in a crowded region for analyzing the behavior of a particular person. Different
sensors are used to collect the information from the person and based on that information
the ANFIS model provides classification results as prediction mechanism if there
is a change in the walking style. The corresponding classification results are forwarded to
a team of doctors or the other persons through the cloud IoT network that is suitable
for elderly persons and physically disable persons in order to avoid fall. If changes
are found in the walking style, then the information is passed to the corresponding
caretakers so that medication facilities can be provided easily at correct time without
any delay.
7.576.565.5
Neural Network Model
Proposed ANFIS Model
54.543.532.521.5
100
90
80
70
60
50
Recognition Rate (%)
40
30
20
10
0
Speed (km/h)
Figure 9.
Recognition rate vs
walking speed
120
10 20 30 40 50 60
Gait Cycles
70 80 90 100
Neural Network Model
Proposed Model
110 120
100
80
60
Recognition Rate (%)
40
20
0
Figure 10.
User recognition
rate vs the number
of gait cycles
52
IJIUS
8,1
References
Gadaleta, M. and Rossi, M. (2018), “IDNet: smartphone-based gait recognition with convolutional
neural networks”,Pattern Recognition, Vol. 74, pp. 25-37.
García, L., Tomás, J., Parra, L. and Lloret, J. (2019), “An m-health application for cerebral stroke
detection and monitoring using cloud services”,International Journal of Information
Management, Vol. 1 No. 45, pp. 319-27.
Gavrilova, M.L., Wang, Y., Ahmed, F. and Paul, P.P. (2018), “Kinect sensor gesture and activity
recognition: new applications for consumer cognitive systems”,IEEE Consumer Electronics
Magazine, Vol. 7 No. 1, pp. 88-94.
Godfrey, A. (2017), “Wearables for independent living in older adults: gait and falls”,Maturitas,
Vol. 100 No. 2017, pp. 16-26.
Gravina, R., Ma, C., Pace, P., Aloi, G. and Fortino, G. (2017), “C-b activity-aaService cyber–physical
framework for human activity monitoring in mobility”,Future Generation Computer Systems,
Vol. 75, pp. 158-171.
Ing-Jr, D. and Chang, Y.-J. (2017), “HMM with improved feature extraction-based feature parameters for
identity recognition of gesture command operators by using a sensed Kinect-data stream”,
Neuro Computing, Vol. 262, pp. 108-119.
Kaur, N. and Sood, S.K. (2017), “An energy-efficient architecture for the Internet of Things (IoT)”,
IEEE Systems Journal, Vol. 11 No. 2, pp. 796-805.
Llamas, C., González, M.A., Hernández, C. and Vegas, J. (2016), “Open source platform for collaborative
construction of wearable sensor datasets for human motion analysis and an application for gait
analysis”,Journal of Biomedical Informatics, Vol. 63, pp. 249-258.
Llamas, C., González, M.A., Hernández, C. and Vegas, J. (2017), “Open source hardware based sensor
platform suitable for human gait identification”,Pervasive and Mobile Computing, Vol. 38 No. 1,
pp. 154-165.
Nweke, H.F., Teh, Y.W., Al-garadi, M.A. and Alo, U.R. (2018), “Deep learning algorithms for human
activity recognition using mobile and wearable sensor networks: state of the art and research
challenges”,Expert Systems with Applications, Vol. 105, pp. 233-261.
Paraskevopoulos, D.C., Laporte, G., Repoussis, P.P. and Tarantilis, C.D. (2017), “Resource constrained
routing and scheduling: review and research prospects”,European Journal of Operational
Research, Vol. 263 No. 3, pp. 737-754.
100.00%
98.00%
96.00%
94.00%
92.00%
90.00%
88.00%
86.00%
84.00%
Average Accuracy Maximum Accuracy
Neural Network Model
Proposed Model Figure 11.
Performance
comparison
53
Wireless
IOT system
Qi, J., Yang, P., Min, G., Amft, O. and Xu, L. (2017), “Advanced internet of things for personalized
healthcare systems: a survey”,Pervasive and Mobile Computing, Vol. 41, pp. 132-149.
Qi, J., Yang, P., Waraich, A., Deng, Z. and Yang, Y. (2018), “Examining sensor-based physical activity
recognition and monitoring for healthcare using Internet of Things: a systematic review”,
Journal of Biomedical Informatics, Vol. 87, pp. 138-153.
Stack, E., Agarwal, V., King, R., Burnett, M. and Kunkel, D. (2018), “Identifying balance impairments in
people with Parkinson’s disease using video and wearable sensors”,Gait & Posture, Vol. 62,
pp. 321-326.
Zhang, Y., Xhafa, F., Ruiz, C. and Yao, L. (2018), “Wearable sensor signal processing for smart health”,
Smart Health, Vols 5/6, pp. 1-3.
About the authors
Sampath Dakshina Murthy Achanta received his BTech Degree in Electronics and Communication
Engineering Degree in 2013 and MTech Degree in Digital Electronics and Communication Engineering
2015. He is pursuing PhD in Electronics and Communication Engineering from Koneru Lakshmaiah
Education Foundation Deemed University in Guntur, Andhra Pradesh. His research interest includes
image and video processing, fuzzy logic and neural networks. He has published 18 papers in
international journals. He is Life Member of ISRS and MIET. Sampath Dakshina Murthy Achanta is
the corresponding author and can be contacted at: sampathdakshinamurthy@gmail.com
Dr Karthikeyan T. received his BE Degree in Electronics and Communication Engineering in
2006 and ME Degree in Embedded System Technologies in 2012 from Anna University and PhD
Degree in Wireless Sensor Networks in 2018 from KAHE, Coimbatore. He has total of 11 years of
experience out of which 2 years in the field of software testing. He is currently working as Associate
Professor in the Department of Electronics and Communication Engineering (DST-FIST sponsored)
at KLEF, Guntur. He is Life Member of ISTE. He has published more than eight research papers in
the leading journals and conferences. His area of interest includes wireless sensor networks,
MANET, Wi-MAX, Wi-Fi and IoT.
Dr Vinoth Kanna R. completed PhD Degree in 2014 at Faculty of Information and Communication
Engineering, Anna University, Chennai. He is working as Professor at the Department of Electronics
and Communication Engineering, Koneru Lakshmaiah Education Foundation, Guntur, Andhra
Pradesh. His research interest includes biometrics, image and video processing, pattern recognition,
embedded systems, fuzzy logic and neural networks. He has published 19 papers in International
Journals and 10 papers in international/national conferences. He is Life Member of ISTE, New Delhi
and Member of IEEE. He has 14 years of teaching and 9 years of research experience.
For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: permissions@emeraldinsight.com
54
IJIUS
8,1