Gait Phases Detection in Elderly using Trunk-MIMU System

Elisa Digo

, Valentina Agostini

, Stefano Pastorelli

, Laura Gastaldi

and Elisa Panero

Department of Mechanical and Aerospace Engineering, Politecnico di Torino, Turin, Italy

Department of Electronics and Telecommunications, Politecnico di Torino, Turin, Italy

Department of Mathematical Sciences “G.L. Lagrange”, Politecnico di Torino, Turin, Italy

Department of Surgical Sciences, Università degli Studi di Torino, Turin, Italy

Keywords: Gait Phases, Elderly Population, MIMU, Walking Condition, Accuracy.

Abstract: The increasing interest towards wearable Magnetic Inertial Measurement Units (MIMUs) for gait analysis is

justified by their low invasiveness, confirmed repeatability and complete independence from laboratory

constraints. However, some crucial doubts about the identification of a suitable sensor set-up and algorithm

in different gait conditions and populations still exist. In this context, the principal aim of the present study

was to investigate the effect of different walking conditions on the accuracy of gait phases detection with a

trunk-MIMU system. Eleven healthy elderly subjects performed gait trials in four different walking conditions

(fast speed, normal speed, slow speed and normal speed with dual-task). A stereophotogrammetric system

was adopted as gold standard. The accuracy of the estimation of stance and swing phases was evaluated from

the comparison of trunk-MIMU to the stereophotogrammetric system. Mean error values smaller than 0.03 s

confirmed the accuracy of the trunk-MIMU algorithm for an elderly population. Consequently, trunk-MIMU

system can be considered suitable for the characterization of gait phases in elderly subjects regardless of

walking conditions.

1 INTRODUCTION

During the last decades, different applications

highlighted the central role of locomotion in human

daily activities, generating a strong interest towards

gait analysis. Several studies have been directed to

assess standard gait patterns (Davis 1997), to identify

the conditioning factors (Hebenstreit et al. 2015), to

select systems and set-ups (Benndorf, Gaedke, and

Haenselmann 2019), as to characterize human gait

phases and kinematics (Kadaba et al. 1989). In

particular, clinical gait analysis is usually aimed at

monitoring rehabilitation processes (Moon et al.

2017), characterizing normal and pathological

locomotion (Prakash, Kumar, and Mittal 2018;

Shirakawa et al. 2017) and verifying therapeutic

treatments (Gastaldi et al. 2015). The objective

measurement of gait parameters supports clinical

https://orcid.org/0000-0002-5760-9541

https://orcid.org/0000-0001-5887-1499

https://orcid.org/0000-0001-7808-8776

https://orcid.org/0000-0003-3921-3022

https://orcid.org/0000-0002-5555-5818

experts during the observational assessment of gait.

Human locomotion can be mainly described by the

identification of two gait events: the heel strike (HS)

and the toe off (TO). In detail, the detection of gait

events allows first to divide each walking trial into

consecutive cycles, then to estimate different gait

phases. The gait cycle (GC) of each limb can be

mainly divided in stance and swing phases. The first

one starts with the load acceptance from the foot and

lasts the entire time the foot is in contact with the

ground, while correspondingly the limb bears part or

whole human weight. The swing phase depicts the

time period of foot oscillation without floor contact.

Durations of stance and swing phases are expressed

as percentages of the GC duration. Generally, in

healthy adults the stance phase represents

approximatively the 60% of the GC, while the swing

phase the 40% of the GC.

Digo, E., Agostini, V., Pastorelli, S., Gastaldi, L. and Panero, E.

Gait Phases Detection in Elderly using Trunk-MIMU System.

DOI: 10.5220/0010256400580065

In Proceedings of the 14th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2021) - Volume 1: BIODEVICES, pages 58-65

ISBN: 978-989-758-490-9

Stance and swing phases can be crucially

influenced by gait velocities, external disturbs or

dual-tasks (Liu et al. 2014). In addition, previous

studies highlighted the aging effect on gait phases

(Aboutorabi et al. 2016). Healthy elderly people

demonstrated a compensatory strategy to overcome

instability and loss of control through the variation of

spatio-temporal parameters. The percentage duration

of the stance phase is increased, entailing a reduced

percentage duration of the swing phase. More in

general, in clinics, altered patterns of locomotion are

assessed by a different percentage distribution of time

in the two phases (Trojaniello et al. 2014). Another

important aspect of pathological gait is the symmetry

between right and left limbs. However, reduced

symmetry is not clearly associated with age in healthy

elderly populations (Aboutorabi et al. 2016).

During past decades, several tools have been used

for the analysis of human locomotion, especially to

add an objective measure to the observational gait

evaluation (Akhtaruzzaman, Shafie, and Khan 2016).

Literature confirms optoelectronic systems as the

gold standard technology thanks to their high

accuracy and precision. Several improvements,

methodologies and innovative biomechanical models

are proposed nowadays to be implemented with

optoelectronic systems for deeper kinematic and

dynamic investigations (Panero, Gastaldi, and Rapp

2018). However, these systems have some crucial

disadvantages, as the cost, the restriction to the

laboratory environment and the required expert

operation.

Recently, wearable sensor technologies such as

Magnetic Inertial Measurement Units (MIMUs) have

shown promising results in measuring human body

motion with limited cost and invasiveness, with a

good reliability and without laboratory constraints

(Cereatti, Trojaniello, and Croce 2015; Digo et al.

2020; Petraglia et al. 2019; G. Yang et al. 2019). The

use of wearable systems may be more suitable for

monitoring the subject for longer observation periods

and during daily activities. However, some open

issues related to MIMUs still exist, such as the

definition of a suitable and reliable set-up (S. Yang

and Li 2012) and the implementation of a robust

algorithm for gait phases identification (Caldas et al.

2017) that can be used in different conditions. Several

previous studies have proposed MIMUs set-ups and

algorithms to assess gait parameters both in healthy

and pathological subjects.

A previous pilot study has been conducted with

three healthy young subjects performing gait trials for

the evaluation of two MIMUs set-ups and associate

algorithms for gait events detection (Panero et al.

2018). In the first set-up one MIMU was positioned

on the trunk, while in the second set-up two MIMUs

were fixed on heels. Results have demonstrated the

suitability of the two MIMUs set-ups and algorithms,

but the set-up involving the trunk-MIMU showed the

best accuracy and simplest usage. Considering these

results and concentrating on the trunk-MIMU set-up,

the analysis has been extended to a larger population

of healthy elderly subjects, in order to validate the

robustness of the algorithm in different walking

conditions.

Consequently, the aim of the current study deals

with the analysis of gait speeds and conditions effects

on the accuracy of gait phases detection with a trunk-

MIMU system. Eleven healthy subjects over 65 years

old performed gait trials in four different walking

conditions. Stance and swing phases have been

monitored as outcomes of interest. Accuracy and

error quantification, obtained from the comparison of

trunk-MIMU results with an optoelectronic reference

system, are analysed.

2 MATERIALS & METHODS

2.1 Participants

Eleven healthy elderly subjects (4 males and 7

females) participated in the research after giving their

written informed consent. Four inclusion criteria were

considered: (i) age over 65 years old, (ii) no declared

neurological disorders, (iii) no musculoskeletal

diseases in the last five years and (iv) no internal

prostheses. The study was approved by the Local

Institutional Review Board. All procedures were

conformed to the Helsinki Declaration. Mean and

standard deviation values of subjects’ age, height,

weight and Body Mass Index (BMI) are reported in

Table 1.

Table 1: Subjects’ data (mean ± standard deviation).

Age

ears

)

Height

(

)

Weight

(

)

BMI

(

)

68.8 ± 5.0 1.6 ± 0.1 70.3 ± 14.9 25.8 ± 3.1

2.2 Instruments

Two motion capture systems were adopted for the

study: an inertial system consisting of one MIMU and

a stereophotogrammetric system composed of six

infrared cameras and nine passive reflective markers.

Gait Phases Detection in Elderly using Trunk-MIMU System

2.2.1 Inertial System

One MTx MIMU (Xsens, The Netherlands)

containing a tri-axial accelerometer (range ±5 G), a

tri-axial gyroscope (range ±1200 dps) and a tri-axial

magnetometer (±75 μT) was used for the test. The

MIMU was fixed on trunk (TRN) of participants at

the level of T12-L1 vertebrae, with an elastic band

provided by the Xsens kit. The sensor was oriented

with

the vertical x-axis pointing downward, the

medio-lateral y-axis directed to the right side of

participants and the anterior-posterior z-axis pointing

in the opposite direction of the gait (Figure 1A). The

MIMU was connected to the Xbus Master, the control

unit able to send data to the PC via Bluetooth. Data

were acquired through the Xsens proprietary software

(MT Manager) with a sampling frequency of 50 Hz.

2.2.2 Stereophotogrammetric System

The stereophotogrammetric system adopted for the

test was composed of two V120:Trio tracking bars

(OptiTrack, USA) and nine passive reflective markers

with a diameter of 14 mm. Each bar was self-

contained, pre-calibrated and equipped with three

cameras able to detect infrared light.

Six markers were fixed on feet of participants

with adhesive tape (Figure 1B): two on toes (right toe

= TOR, left toe = TOL), two on malleolus (right

malleolus = MAR, left malleolus = MAL) and two on

heels (right heel = HER, left heel = HEL). Other three

markers (A, B and C) were placed on the floor in

order to define the Global Coordinate System (GCS)

in which to report data recorded by the bars (Panero

et al. 2018). Each bar was connected to a separate PC.

Data acquisition was made with the OptiTrack

proprietary software (Motive) with a sampling

frequency of 120 Hz.

2.3 Protocol

The experimental test was conducted indoor. The two

OptiTrack bars were located one in front of the other

parallel to a 6-meters linear walking path traced on

the floor. Consequently, the obtained captured area

was 2.5 m x 3.5 m, to guarantee the acquisition of at

least three steps for each transition in front of the

cameras. A static recording was made to obtain the

coordinates of the three fixed markers A, B, C on the

floor (Figure 2).

Participants were first asked to hit their right heel

on the floor to define an external event to synchronize

the stereophotogrammetric system and the inertial

system. Subsequently, subjects walked barefoot on

the linear path in four conditions. In the first three

conditions, they were asked to walk at different self-

selected speeds: fast, normal and slow. In the fourth

condition, participants were involved in a dual-task

condition at self-selected normal speed. While

walking, they were asked many questions about their

lives and habits. For each walking condition, all

subjects performed 26 transitions in front of the

cameras. The order of the four sets of walking

conditions was randomized for all subjects.

Coordinates of markers and signals of MIMUs were

acquired at the same time with the two motion capture

systems.

Figure 1: Configuration of trunk-MIMU (A) and markers

(B) on body of participants.

2.4 Signal Processing and Data

Analysis

Signal processing and data analysis were conducted

with customized Matlab routines. Considering the

static recording of markers on the floor, a

transformation matrix was built and used to express

in the GCS all markers trajectories collected during

gait sessions. Afterwards, the temporal

synchronization of data from the two motion capture

systems was guaranteed through the initial impact of

the right foot on the floor (Panero et al. 2018). Gait

events were then separately identified from data

BIODEVICES 2021 - 14th International Conference on Biomedical Electronics and Devices

acquired by the MIMU system and the optoelectronic

system. This detection was made with two algorithms

inspired by previous literature works. Considering the

optoelectronic system, HSs and TOs were identified

from horizontal and vertical coordinates of heels and

toes markers, respectively (Panero et al. 2018;

Veilleux et al. 2016). Since each bar captured the

lateral view of one side of the body, markers on

malleolus were used to distinguish between right and

left sides during gait. As regards the MIMU system,

gait events were identified from the anterior-posterior

acceleration signal of trunk-MIMU. More in detail,

HSs and TOs were detected as maximum and

minimum peaks of this signal, respectively (Panero et

al. 2018; Zijlstra and Hof 2003). In addition, the

distinction between right and left gait events was

made by considering the alternation sign of trunk-

MIMU angular velocity signal around the vertical

axis (McCamley et al. 2012; Panero et al. 2018).

For each subject, a total number of gait cycles

between 150 and 300 was collected. First, for each

participant in each testing condition, the average

walking velocity was calculated as the ratio between

the total gait path and the travel time. Then, for each

testing condition, inter-subjects mean and standard

deviation of walking speed values were estimated.

Afterwards, using gait events obtained with both

algorithms, spatio-temporal parameters of stance and

swing times were assessed for each gait cycle of each

participant in all walking conditions. For both stance

and swing times, mean and standard deviation values

were calculated intra- and inter-subjects for both right

and left sides. Moreover, the symmetry of participants

was evaluated by estimating the limp index as the

ratio between right and left stance times. According

to this confirmed symmetry, values of stance and

swing times were averaged between right and left

sides and represented through bar diagrams. In

addition, stance and swing durations were estimated

as percentages of the GC, in order to evaluate the

effect of age on gait phases distribution.

The accuracy of the MIMU algorithm was

evaluated as the relative error between the mean value

estimated with the optoelectronic system and the

mean value obtained with the MIMU system, for each

participant. Subsequently, inter-subjects mean values

of errors were calculated in all walking conditions.

The sign of the error allowed the differentiation

between overestimation (negative sign) and

underestimation (positive sign) with respect to the

reference value. Finally, a stem graph representation

was adopted in

order to compare errors for both

stance and swing times in different walking

conditions.

Figure 2: Top view of the setting with distance between

OptiTrack bars, measures of the capture volume, length of

walking path and GCS definition.

3 RESULTS

Table 2 depicts average and standard deviation values

of walking speed (m/s) for the tested population in all

the four conditions.

Table 2: Inter-subjects mean and standard deviation values

of walking speed (m/s) in four conditions.

Speed (m/s) Mean ± St. Dev.

Fast 1.16 ± 0.16

Normal 0.87 ± 0.12

Slow 0.74 ± 0.14

Dual 0.82 ± 0.15

Figure 3 shows inter-subjects mean and standard

deviation values of stance and swing times (s)

estimated with both OptiTrack and trunk-MIMU in

all walking conditions. In Figure 4, two stem graphs

represent mean errors for stance and swing times in

all walking conditions (red circle for fast speed, green

diamond for normal speed, blue square for slow speed

and black pentagram for dual-task).

Table 3 contains inter-subjects mean and standard

deviation values of limp index, stance duration (%

GC) and swing duration (% GC) obtained from the

two algorithms in all walking conditions.

Walking path

6 m

6.5 m

V120:Trio

Capture volume

3.5 m

2.5 m

GCS

Gait Phases Detection in Elderly using Trunk-MIMU System

Figure 3: Stance time and swing time estimated in different

walking conditions with OptiTrack (blue) and trunk-MIMU

(orange) systems.

4 DISCUSSIONS

The main aim of the current study was to evaluate

how the accuracy and robustness of a trunk-MIMU

algorithm in gait phases identification are influenced

by four different walking conditions (speeds and

dual-task). In order to fulfill this purpose, inter-

subjects mean and standard deviation values of

walking speeds (Table 2) were calculated. As

reported by Aboutorabi and colleagues, a walking

speed of 1.30 m/s can be considered the standard

reference value for normal walking in healthy adults

(Aboutorabi et al. 2016). Moreover, they referred to

previous studies showing a loss of gait speed based

on age (1.2%/year). In the present work, inter-

subjects mean walking speed in normal condition

(0.87 m/s) confirms this reduction provoked by age.

Moreover, even the registered walking speed in fast

condition (1.16 m/s) is lower than the reference value

of normal walking speed in healthy adults. In the

dual-task condition, walking speed of subjects (0.82

m/s) was lower than the one of normal condition (0.87

m/s), but higher with respect to the slow speed

condition (0.74 m/s). This aspect could be justified

considering that participants were involved in

answering questions and consequently were less

focused on walking. Deeper investigation comparing

normal and dual-task conditions with a larger

population might demonstrate the significance of this

difference.

Figure 4: Errors of trunk-MIMU algorithm with respect to

OptiTrack for both stance time and swing time in the four

walking conditions.

The effect of age on symmetry has been

previously investigated by different studies

(Aboutorabi et al. 2016). In the present work, the

symmetry of participant was evaluated by estimating

the limp index in all walking conditions both with

trunk-MIMU and OptiTrack (Table 3). Since inter-

subjects mean values of limp index were always

around 1 as expected in a healthy gait, symmetry of

participants was confirmed. Consequently, right and

left values of stance time and swing time (Figure 3)

and percentage durations (Table 3) were averaged.

OptiTrack - MIMU (s)

Fast Norm Slow Dual

-0.05

0.05

OptiTrack - MIMU (s)

Stance time error

BIODEVICES 2021 - 14th International Conference on Biomedical Electronics and Devices

Table 3: Limp index, stance duration (%GC) and swing duration (%GC) estimated by OptiTrack and trunk-MIMU systems

in all walking conditions (inter-subjects mean ± standard deviation).

Fast Normal Slow Dual

OptiTrack

Trunk-

MIMU

OptiTrack

Trunk-

MIMU

OptiTrack

Trunk-

MIMU

OptiTrack

Trunk-

MIMU

Limp

index

1.01

(0.03)

1.01

(0.03)

1.00

(0.01)

1.01

(0.03)

1.01

(0.03)

1.00

(0.04)

1.00

(0.02)

1.01

(0.04)

Stance

duration

(%GC)

61.76

(1.42)

60.44

(1.61)

63.20

(1.66)

62.09

(2.40)

64.57

(1.70)

62.77

(2.01)

64.31

(1.71)

62.39

(2.13)

Swing

duration

(%GC)

38.24

(1.42)

39.56

(1.61)

36.80

(1.66)

37.91

(2.40)

35.43

(1.70)

37.23

(2.01)

35.69

(1.71)

37.61

(2.13)

Low cost, low invasiveness and confirmed

repeatability of inertial sensors make them a suitable

alternative to optoelectronic systems for gait analysis.

Despite large investigations and many applications,

some crucial gaps still exist for the identification of a

robust and accurate sensor set-up configuration and

algorithm that can be applied in different gait

conditions and populations. Considering young

subjects, the trunk-MIMU solution resulted to be the

most suitable one (Panero et al. 2018). In the present

study, stance time and swing time have been selected

as outcomes of interest for the validation of accuracy

and robustness of the trunk-MIMU algorithm and set-

up on an elderly population. As Figure 3 shows, both

stance time and swing time increase with the

reduction of gait speed. In the dual-task condition,

values of stance time and swing time are halfway

between the correspondent ones of normal and slow

speed conditions. Moreover, small standard deviation

values depict a repeatability of the measure inside the

tested sample of elderly subjects (Pacini Panebianco

et al. 2018). Considering the accuracy in gait phases

detection with the trunk-MIMU system with respect

to the OptiTrack one, bar diagrams of Figure 3 show

strong accordance between values of both stance time

and swing time in all walking conditions. This

correspondence could be evaluated with stem graphs

in Figure 4. Smaller errors were obtained for

conditions at fast (+0.01 s for stance time, -0.01 s for

swing time) and normal speeds (+0.01 s for stance

time, -0.01 s for swing time). Stance time error is

greater in dual-task condition (+0.03 s), while the

greater error for swing time was registered in slow

speed condition (-0.03 s). However, in all walking

conditions, errors are lower than 0.03 s for both

parameters. In addition, stance time is always

overestimated (positive sign of errors), while an

underestimation interests the swing time (negative

signs of errors). This aspect might be justified by the

later detection of toe off performed with the trunk-

MIMU, probably caused by less clear minimum

peaks of the signal. Nevertheless, the overestimation

of stance time and the underestimation of swing time

demonstrate the constancy of the gait cycle duration.

Better performance at fast and normal speeds could

be explained by an easier identification of peaks of

interest in acceleration and angular velocity signals

used for HSs and TOs detection. Despite this aspect,

the trunk-MIMU algorithm could be considered

accurate for gait phases detection also in elderly

subjects.

Considering Table 3, values of stance duration

and swing duration obtained as percentages of GC

were observed. Reference values of stance duration

and swing duration in normal gait are 60% and 40%

of the GC, respectively. The current elderly

population shows an increased stance duration

(around 63% GC for OptiTrack and 62% GC for

trunk-MIMU) and a consequent reduction of swing

duration (around 37% GC for OptiTrack and 38% GC

for trunk MIMU) in normal walking condition. In

faster walking speed, the reduction of stance duration

with respect to normal speed can be underlined with

both OptiTrack (around 62% GC) and trunk-MIMU

(around 60% GC), with a resulting increase of swing

phase duration. In slow walking speed, the increase

of stance duration with respect to normal speed can

be underlined with both OptiTrack (around 65% GC)

and trunk-MIMU (around 63% GC), with a resulting

reduction of swing phase duration. Finally, the

walking condition with dual-task shows percentage

times distribution similar to the slow speed condition,

both for OptiTrack and trunk-MIMU.

5 CONCLUSIONS

In conclusion, the presented analysis confirms that

the trunk-MIMU system is suitable for the

characterization of gait phases not only in healthy

young subjects (Panero et al. 2018), but also in an

healthy elderly population. The trunk-MIMU system

depicts small errors of stance time and swing time

Gait Phases Detection in Elderly using Trunk-MIMU System

calculation at different walking conditions, revealing

its accuracy and robustness. Moreover, the singular

MIMU configuration might reveal advantages in

terms of ease of use, limited cost and reduced

invasiveness. For all these reasons, the trunk-MIMU

system demonstrates to be a strategical and potential

alternative to traditional stereophotogrammetric

systems to evaluate gait phases.

The principal limitation of this study consists in

the involvement of a small sample of participants.

However, this limit is expected to be overcome in the

future, by testing a larger number of elderly subjects

and by considering the possibility to identify

subgroups based on gender, healthy conditions and

specific age.

Future perspectives will concentrate first on the

evaluation of additional spatio-temporal parameters,

including symmetry indices. Then, plans are to test

the same MIMU set-up and algorithm on pathological

populations, in order to define a complete protocol for

the evaluation of rehabilitation progress and

therapeutic treatments benefits.

REFERENCES

Aboutorabi, Atefeh et al. 2016. “The Effect of Aging on

Gait Parameters in Able-Bodied Older Subjects: A

Literature Review.” Aging Clinical and Experimental

Research 28(3): 393–405.

https://link.springer.com/article/10.1007/s40520-015-

0420-6 (September 29, 2020).

Akhtaruzzaman, Md, Amir Akramin Shafie, and Md

Raisuddin Khan. 2016. “GAIT ANALYSIS:

SYSTEMS, TECHNOLOGIES, and IMPORTANCE.”

Journal of Mechanics in Medicine and Biology 16(7).

Benndorf, Maik, Martin Gaedke, and Thomas

Haenselmann. 2019. “Towards Gait Analysis - Creating

a Setup for the Analyses under Laboratory Conditions.”

In ACM International Conference Proceeding Series,

New York, New York, USA: Association for

Computing Machinery, 1–5.

http://dl.acm.org/citation.cfm?doid=3341325.3342005

(September 29, 2020).

Caldas, Rafael et al. 2017. “A Systematic Review of Gait

Analysis Methods Based on Inertial Sensors and

Adaptive Algorithms.” Gait and Posture 57: 204–10.

Cereatti, Andrea, Diana Trojaniello, and Ugo Della Croce.

2015. “Accurately Measuring Human Movement Using

Magneto-Inertial Sensors: Techniques and

Challenges.” In 2nd IEEE International Symposium on

Inertial Sensors and Systems, IEEE ISISS 2015 -

Proceedings, Institute of Electrical and Electronics

Engineers Inc.

Davis, Roy B. 1997. “Reflections on Clinical Gait

Analysis.” In Journal of Electromyography and

Kinesiology, Elsevier, 251–57.

Digo, Elisa, Giuseppina Pierro, Stefano Pastorelli, and

Laura Gastaldi. 2020. “Evaluation of Spinal Posture

during Gait with Inertial Measurement Units.”

Proceedings of the Institution of Mechanical Engineers,

Part H: Journal of Engineering in Medicine 234(10):

1094–1105.

Gastaldi, Laura et al. 2015. “Effects of Botulinum

Neurotoxin on Spatio-Temporal Gait Parameters of

Patients with Chronic Stroke: A Prospective Open-

Label Study.” Eur J Phys Rehabil Med 51(5): 609–18.

Hebenstreit, Felix et al. 2015. “Effect of Walking Speed on

Gait Sub Phase Durations.” Human Movement Science

43: 118–24.

Kadaba, M. P. et al. 1989. “Repeatability of Kinematic,

Kinetic, and Electromyographic Data in Normal Adult

Gait.” Journal of Orthopaedic Research 7(6): 849–60.

https://onlinelibrary.wiley.com/doi/full/10.1002/jor.11

00070611 (September 29, 2020).

Liu, Yancheng et al. 2014. “Gait Phase Varies over

Velocities.” Gait and Posture 39(2): 756–60.

McCamley, John, Marco Donati, Eleni Grimpampi, and

Claudia Mazzà. 2012. “An Enhanced Estimate of Initial

Contact and Final Contact Instants of Time Using

Lower Trunk Inertial Sensor Data.” Gait and Posture

36(2): 316–18.

Moon, Sang Bok et al. 2017. “Gait Analysis of Hemiplegic

Patients in Ambulatory Rehabilitation Training Using a

Wearable Lower-Limb Robot: A Pilot Study.” Int J

Prec Eng and Manufacturing 18(12): 1773–81.

Pacini Panebianco, Giulia, Maria Cristina Bisi, Rita Stagni,

and Silvia Fantozzi. 2018. “Analysis of the

Performance of 17 Algorithms from a Systematic

Review: Influence of Sensor Position, Analysed

Variable and Computational Approach in Gait Timing

Estimation from IMU Measurements.” Gait and

Posture 66(April): 76–82.

https://doi.org/10.1016/j.gaitpost.2018.08.025.

Panero, E., E. Digo, V. Agostini, and L. Gastaldi. 2018.

“Comparison of Different Motion Capture Setups for

Gait Analysis : Validation of Spatio-Temporal

Parameters Estimation.” In MeMeA 2018 - 2018 IEEE

International Symposium on Medical Measurements

and Applications, Proceedings, Institute of Electrical

and Electronics Engineers Inc.

Panero, E, L Gastaldi, and W Rapp. 2018. “Two-Segments

Foot Model for Biomechanical Motion Analysis.” In

Mechanisms and Machine Science, Springer

Netherlands, 988–95.

Petraglia, Federica et al. 2019. “Inertial Sensors versus

Standard Systems in Gait Analysis: A Systematic

Review and Meta-Analysis.” Eur J of Phy and Rehab

Med 55(2): 265–80.

Prakash, Chandra, Rajesh Kumar, and Namita Mittal. 2018.

“Recent Developments in Human Gait Research:

Parameters, Approaches, Applications, Machine

Learning Techniques, Datasets and Challenges.”

Artificial Int Review 49(1): 1–40.

Shirakawa, Tomohiro et al. 2017. “Gait Analysis and

Machine Learning Classification on Healthy Subjects

BIODEVICES 2021 - 14th International Conference on Biomedical Electronics and Devices

in Normal Walking.” Int J Parallel, Emergent and

Distributed Systems 32(2): 185–94.

Trojaniello, Diana et al. 2014. “Estimation of Step-by-Step

Spatio-Temporal Parameters of Normal and Impaired

Gait Using Shank-Mounted Magneto-Inertial Sensors:

Application to Elderly, Hemiparetic, Parkinsonian and

Choreic Gait.” J NeuroEng and Rehab 11(1): 1–12.

Veilleux, Louis Nicolas et al. 2016. “Agreement of Spatio-

Temporal Gait Parameters between a Vertical Ground

Reaction Force Decomposition Algorithm and a

Motion Capture System.” Gait and Posture 43: 257–64.

http://dx.doi.org/10.1016/j.gaitpost.2015.10.007.

Yang, Gelan et al. 2019. “Review Wearable Sensing

System for Gait Recognition.” Cluster Computing

22(2): 3021–29.

Yang, Shuozhi, and Qingguo Li. 2012. “Inertial Sensor-

Based Methods in Walking Speed Estimation: A

Systematic Review.” Sensors 12(5): 6102–16.

Zijlstra, Wiebren, and At L. Hof. 2003. “Assessment of

Spatio-Temporal Gait Parameters from Trunk

Accelerations during Human Walking.” Gait and

Posture 18(2): 1–10.

Gait Phases Detection in Elderly using Trunk-MIMU System