Detection of Gait Events and Assessment of Fall Risk Using

Accelerometers in Assisted Gait

A. Tereso

, M. Martins

, C. P. Santos

M. Vieira da Silva

, L. Gonçalves

and L. Rocha

Industrial Electronics Dep., Minho University, Guimarães, Portugal

Braga Hospital, Braga, Portugal

Keywords: Walker, Accelerometer, Assisted-Gait, Stability, Fall Risk, KOA (Knee Osteoarthritis), TKA (Total Knee

Arthroplasty).

Abstract: The use of the walker in rehabilitation has increased in the past few years. Therapists evaluate patient’s

rehabilitation by observation and subjective tests. Thus, it is necessary the use of an assistive tool which can

measure and quantify the patient’s walker-assisted movement and stability, providing an objective clinical

assessment. The aim of this study is to detect differences in assisted gait when using the assistive devices

(ADs) – crutches, standard walker and rollator (4-wheeled walker) with forearm supports (RFS) - in patients

with knee osteoarthritis (KOA) that suffered the surgery - Total Knee Arthroplasty (TKA). Additionally, it

is to verify the link between gait parameters and acceleration signals.

The evaluation is reached by the use of two 3 axis-accelerometers. The signals extracted from the sensors, at

the ankle and trunk, are related to gait events and evaluation of fall risk, respectively. Results show that

despite the differences between the signals obtained with the three ADs and with the subjects in this study, it

is possible to identify effectively the gait parameters and prove the stability that the RFS provides.

1 INTRODUCTION

Pain relief and the improvement of knee function are

the two main reasons for total knee arthroplasty

(TKA) in cases of Knee Osteoarthritis (KOA). KOA

patients suffer by pain, stiffness and decreased range

of motion of the knee, which provokes the reduction

of their mobility (Kaufman et al., 2001).

Precise motor function evaluation in

rehabilitation programs is a major challenge in

clinical practice and has gained widespread interest

with recent technologies. Nowadays, in assistive

device rehabilitation, therapists evaluate patient’s

rehabilitation by observation and subjective tests.

Such information is qualitative and final clinical

decisions are strongly empirical and subjective. This

evaluation can be more objective and quantitative, if

it applies gait techniques that allow a systematic

study and characterization of the human locomotion

like accelerometers - low cost wearable sensor

systems. These devices are easy to use, can be

positioned closed to the places that are supposed to

be, are portable and have several biomedical

applications (Watanabe et al., 2011).

In this study, it is proposed to assess gait parameters

(stance, swing, stride time, etc.) and their variability

in assisted gait with three different assistive devices

(ADs) (crutches, standard walker and rollator with

forearm supports (RFS)) with KOA patients that

suffered the surgery TKA. This evaluation was

reached with two accelerometers placed at the ankle

to detect gait events (toe-off and heel strike) and at

the trunk to assess the centre of mass (COM)

displacement of the subject. These will provide

information about the stability provided by the ADs,

as well as estimation of fall risk. The choice of the

spatiotemporal (stride, swing and stance time,

velocity, cadence and step length) over the

kinematics parameters, it was because these

parameters provide an objective measurement tool

and can help in evaluating KOA severity,

effectiveness of treatment and might help in disease

management (Debi et al. 2011). It was not possible

to compare in this study the assisted gait with

unassisted gait, because at the moment of this

evaluation, the patients were in recovery from the

surgery (between 3

and 5

day after surgery), so

they only could walk with the help of ADs. The

788

Tereso A., Martins M., P. Santos C., Vieira da Silva M., Gonçalves L. and Rocha L..

Detection of Gait Events and Assessment of Fall Risk Using Accelerometers in Assisted Gait.

DOI: 10.5220/0005117507880793

In Proceedings of the 11th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2014), pages 788-793

ISBN: 978-989-758-039-0

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

KOA patients are characterized by slow speed,

shorter step length and shorter single limb support

(Debi et al. 2011). Nowadays, the recovery of KOA

patients is made with the help of crutches. However,

this type of AD provides an unnatural gait

performance and the patients cannot alleviate their

pain while walking. Thus, it is intended on this study

to find a better solution for the recovery of KOA

patients, providing a better gait performance in

terms of cadence, speed, comfort and safety. The

authors hypothesized that the RFS is a better

solution.

The localization of the accelerometer depends on

the purpose of the study in terms of gait assessment.

In (Sabatini et al. 2005; Doheny et al. 2012), they

only used 1 accelerometer positioned at the centre of

the foot of the subject to identify gait parameters,

but the signal had too much noise and was very

irregular. Also, the sensor can be located at the trunk

and at the ankle (Lee et al. 2010) for gait evaluation.

However, to detect gait events (toe-off and heel

strike) it is preferable to place the sensor at the

ankle, over the trunk, since it is more sensitive to

changes on the lower limbs, providing more

information about gait events (Lee et al. 2010). In

order to detect automatically such events, the

selected method of this study will be based on (Lee

et al. 2010). Thus, assisted-gait evaluation with an

accelerometer still remains to be validated and in

this study it will be performed with the sensor at the

ankle. In this case, since the subjects have the knee

injured, it would make sense the placement of the

sensor at the knee, to better assessment. However, in

this study we intended to evaluate the gait and detect

gait events, so, since the magnitude of acceleration

increases from the head to the ankle (Mathie et al.

2004), the signal in the ankle will be more precise

and reliable.

To assess the risk of fall, a sensor should be

located near the COM since it is the best place to

evaluate with accuracy (Vaughan et al. 1999). Thus,

in this work, an accelerometer is placed at the level

of the trunk (sacrum), closed to the COM to evaluate

its displacement. The assessed COM displacement

parameters are based in (Doheny et al. 2012).

However, the evaluation performed in (Doheny et al.

2012) was done for the standing position and not

during walk. Thus, this study aims to verify the

potential of using an accelerometer placed on the

trunk to assess fall risk in assisted gait.

Overall, the goal of this study is to detect

differences between three ADs in assisted gait by

analysing which parameters are most affected by the

use of the crutches, standard walker and RFS in

patients with KOA, considering gait events and the

trunk parameters. As far as the authors know, there

are no references on the use of accelerometers in

assisted gait with any ADs, only in non-assisted gait.

Furthermore, it is intended to verify and validate if

the data extracted from the accelerometers is able to

detect gait events and changes on the variability of

the parameters in assisted gait. The authors expect

that the RFS produce the most stable and less

variable gait, because of the support provided by the

forearms, relatively to the others ADs.

The article is organized as follows. In section 2 it

is presented the algorithm, the processing

implemented and the parameters analysed. Section 3

briefly reveals the results that were obtained.

Sections 4 and 5 are referenced to the discussion and

conclusions of these results, respectively.

2 METHODS

2.1 Subjects

A group (N=7) of subjects (3 men and 4 women)

aged 67.3 5.06 years that were diagnosed with

KOA and suffered the surgery TKA were selected to

the trials. The study was conducted at Hospital of

Braga, approved by the Ethical Committee, and all

the patients signed the informed consent. All trials

were filmed with a video camera.

2.2 Test Procedure

In order to assess the effect of the ADs on gait, tests

are conducted using crutches, standard walker and

RFS (ASBGO walker developed by the authors’

team). All the ADs are shown in Figure 1. In these

tests, subjects had to walk approximately 10m with

the ADs, along a corridor. 3 walking trials for each

subject and AD are realized. Then, the mean and

standard deviation are estimated for each gait

parameter. For each patient the height of the Ads is

adjusted. To measure the accelerations of the lower

limb and the trunk, two inertial sensors are used.

These sensors (SMI, MP6000 of InvenSense, which

include an accelerometer and a gyroscope, both of

them are 3-axial) need a computer and a base station

(CC2530 of Texas Instrument).

In this study only the accelerometer is used. Two

sensors are used, for simplicity, attached to the ankle

of the leg with the injured knee and at the sacrum

(trunk). The used system configuration and the

coordinates of reference for the ankle and trunk are

shown in Figure 2. The x-axis, y-axis and z-axis

DetectionofGaitEventsandAssessmentofFallRiskUsingAccelerometersinAssistedGait

789

correspond to the medio-lateral (ML), vertical (V)

and anterior-posterior (AP) accelerations,

respectively.

Figure 1: ADs used in this study. Left image: Crutches;

centre image: standard walker; and right image: RFS

ASBGo walker).

Figure 2: Description of the axis of the accelerometer at

the right ankle (left image) and at the trunk (right image).

2.3 Data Acquisition and Processing

2.3.1 Detection of Gait Parameters

The algorithm implemented in this study for the

detection of gait events (heel strike and toe-off) is

based on (Lee et al. 2010). These two events are

essential for the calculation of gait parameters like

stance and swing phase.

The implementation consists on the detection of

the time peak of Heel Strike (HS) and Toe-Off (TO)

events. First, at each instant of time, the data of each

axis is summed and transformed to produce the

‘Signal Vector Machine’, represented by s:























(1)

Where a

, a

and a

are the ML, V and AP

accelerations, respectively.

This step is applied since acceleration is highly

influenced by the position of the sensor and the 3

axis have significant information. Second, s is

filtered by a low pass filter (f

pass

=6Hz, f

stop

=10Hz) to

extract features related to the gait cycle,



∑











(2)

where b

corresponds to the coefficients of the

filter. These coefficients are obtained by running the

fdatools interface in MATLAB. Third, a least-square

polynomial derivative approximation filter

eliminates noise (points that could be considered

wrongly as peaks),













1











1







3



24



(3)

After this processing, the final step consists on

the peak detection. For each gait cycle there are two

peaks, each of them corresponding to a gait event

(HS and TO). Before this last step, it was necessary

to remove some sample points from the start and end

of each test, which correspond to the period of

acceleration and deceleration in gait, respectively.

the duration of these periods is irregular. To validate

the detection of such events, it was used one FSR

(Force Sensitive Resistor) under the right heel,

attached to the shoe, to measure these events.

After detecting both events, some gait

parameters can be calculated. The gait cycle is

divided in two phases – stance and swing. The

majority of the gait cycle is spent in stance phase

(60%) and the rest in swing phase (40%). The stance

phase corresponds to the moment that the foot is in

contact with the ground. The swing phase is the

period during which the leg is out of the ground,

moving to the next strike. Thus, stance phase begins

with HS event and finishes with TO event. Swing

phase begins with TO and finishes with HS. Once

stance and swing phases are detected, stride time,

cadence, average velocity and step length gait

parameters can be calculated. These parameters were

calculated as in (Sabatini et al. 2005; Henriksen et

al. 2004).

2.3.2 Assessment of the Fall Risk

The evaluation of the risk of falling of the subject is

reached by an accelerometer attached to the trunk.

All the processing applied in these signals is adapted

from (Doheny et al. 2012). A band-pass filter of fifth

order between 0.1-10Hz filters the signals, to restrict

the signal. Then, to obtain the displacements of the

subject’s COM the acceleration signals are double

integrated, using a trapezoidal method. The error

associated to the integration (low frequency drift) is

reduced by subtracting the mean of the acceleration

signals before and after each integration, and then

implementing a second-order polynomial fit and a

high-pass filter of fifth order to 0.1Hz.

These signals enable to determine the Root Mean

Square (RMS) for AP and ML directions, sway

range AP and Ml and horizontal displacement of the

COM (Dhor), given by,

 











(10)

where dML and dAP are obtained after both

integrations and correspond to the displacements in

AP and ML directions, respectively. Then, it is

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

790

calculated the horizontal displacement of the COM.

These parameters enable to assess the risk of falling

of the patient.

2.3.3 Statistical Analysis

For each parameter the mean and standard deviation

was calculated. Then, a Student’s t-test was

performed to compare the results obtained with the

crutches and standard walker with the RFS. The

level of significance was set to p<0.05.

3 RESULTS

Figure 4 show portions of the signals of one of the

patients with the crutches, standard walker and RFS,

acquired with the accelerometer attached to the

ankle. As mentioned before, it was used one FSR to

validate the detection of the gait events. The FSR

detects 60ms and 30ms earlier the HS and TO

events, respectively. In the three graphs, it is

indicated, by different markers, both the instants of

HS and TO detected by the accelerometer and by the

FSR. It is also identified some of the gait parameters

determined in the study. Figure 5 a and b are a

portion of the AP and ML accelerations of the trunk,

respectively, with each AD, acquired by the

accelerometer placed at the sacrum. It is

discriminated the AP and ML accelerations. Table 1

presents the mean and standard deviation of each

parameter for 7 patients for the three ADs. As one

can see in Table 1, the values of p-values less than

0.05 were obtained for the same parameters among

the different ADs (stride and stance time, velocity,

cadence and sway range ML). Relatively to the

values, the crutches provide the higher stride, stance,

swing time and step length. The RFS has the lower

values for these parameters, except for the velocity

and cadence. Considering the values acquired by

(Martins et al. 2013) with a similar RFS, but with

laser sensor, for the same diagnosis, one can see that

they have obtained similar values.

In terms of the parameters obtained by the

accelerometer at the trunk, all of them were greater

for the crutches and smaller for the RFS, except for

the RMS AP.

4 DISCUSSION

The goal addressed in this work is to detect

differences between the ADs for the assisted gait.

Hence, the gait parameters are studied in order to

verify which are most affected in patients with

KOA, relatively to gait events and fall risk.

Furthermore, it is intended to verify and validate if

the data extracted from the accelerometer is able to

detect gait events in assisted-gait. As it can be seen

in figure 4 for each gait cycle it is possible to

observe two peaks (HS and TO).

Figure 4: Portion of the signals of the ankle of a patient

walking with a) crutches b) standard walker and c) RFS.

The x-axis and y-axis correspond to the time and

acceleration, respectively. The red triangle and the black

cross correspond to HS and HS with delay, respectively.

The rose star and blue square are relative to the TO and

TO with delay. These graphics are in accordance to (Lee

et al. 2010).

For the assisted gait with crutches, standard walker

and RFS, the moment of HS occur at t=58.5s,

t=59.7s and t=34.0s and TO at t=60.1s, t=61.3s and

t=35.5s, respectively. This means that for each gait

cycle and for each device, it can be detected the time

that the foot contacts the ground and the time that it

leaves the ground, respectively. Hence, with the

identification of these two events, it is possible to

determine the desired gait parameters. The recorded

signals for walking trials are very similar to those

presented in the literature for free walking (Lee et al.

2010). Therefore, it was reasonable to take into

account these signals for further analysis.

Thus, it can be concluded that accelerometers

can effectively be used in the gait analysis of

assisted gait since they verify a relationship between

the measured acceleration signal and gait events.

By observation of Table 1, comparing the stance

DetectionofGaitEventsandAssessmentofFallRiskUsingAccelerometersinAssistedGait

791

Figure 5: Portion of the signals a) AP and b) ML of the

trunk for one patient. The first, second and third for each

graph, correspond to crutches, standard walker and RFS,

respectevely.

Table 1: Devices used and calculated parameters. The

mean and the standard deviation for the parameters for

assisted gait with the three ADs are listed. The values

between brackets correspond to the percentage of the gait

cycle. D:Device, P:Parameters, Cr:Crutches, p:p-value,

SW:Standard Walker, ST:Stride Time, SgT:Swing Time,

StT:Stance Time, C:Cadence, v:Velocity, SL:Step Length,

DC:Displacement COM, R AP:RMS AP, R ML:RMS ML,

SR AP:Sway Range AP, SR ML:Sway.

D P Cr p SW p RFS

Acc.(Ankle)

ST (s)

4.39 1.99

0.03

3.73 1.12

0.01

2.65 0.66

SgT

(s)

1.42 0.53

(32.35%)

0.22

1.21 0.57

(32.44%)

0.65

1.12 0.31

(42.26%)

StT

(s)

2.93 1.56

(67.65%)

0.03

2.51 0.59

(67.56%)

0.005

1.47 0.67

(57.74%)

Camera

(step/

min)

25.76

13.86

0.008

30.59

10.16

0.002

43.35

14.12

(m/s)

0.11

0.05

0.02

0.09

0.04

0.004

0.20 0.03

(m)

0.31

0.10

0.26

0.29

0.18

0.96

0.29 0.09

Acc.(Trunk)

(m)

1.17

0.49

0.13

0.69

0.19

0.57

0.52 0.28

R AP

(m/s

)

0.69

0.24

0.49

0.08

0.56

0.61 0.29

R ML

(m/s

)

0.48

0.34

0.42

0.40

0.10

0.16

0.32 0.06

(m)

1.95

1.25

0.21

1.14

0.30

0.62

0.89 0.56

(m)

1.080.13

0.01

0.770.18

0.007

0.52 0.17

and swing percentages for the crutches, standard

walker and RFS, one can see that the stance phase is

67.65%, 67.56% and 57.74% and the swing phase is

32.35%, 32.44% and 42.26%, respectively.

Considering that normal free gait is characterized by

having 60% of stance phase and 40% of swing phase

(Vaughan et al. 1999), the results obtained with

RFS are the more approximated to these normal

values. The stance (58.06%) and swing (41.92%)

phases percentage are similar to the ones obtained

by Martins et al. (2013) and they justified these

values by the fact that these patients are better

supported by the RFS, and they feel less pain when

loading the affected joint, allowing to perform a

more natural gait. The greater swing phase

percentage with the RFS, relatively to the others

ADs, could be explained by the existence of the

forearm supports, which provide a greater support.

Relatively to the stance phase percentage, it was

verified a decrease with the RFS, comparatively to

the others ADs. Since impaired gait and/or fear of

falling usually results in an increase of stance time

(Kloos et al. 2012), this result shows that the RFS

offered excellent support and stability for the user,

by increasing his sense of security relatively to the

others ADs. The crutches produced the greater stride

time and, consequently, greater stance and swing

time, relatively with the others ADs, which is good,

because it means that the patient spent more time

with the leg, that has the knee injured, in the ground

– stance phase. On the other side, the RFS the lower

values for the stride, stance and swing time, because

of the continuous movement of the subject with this

device. The value of the velocity for assisted gait for

the patients is lower than for the healthy (Martins et

al. 2013). Considering our values, the standard

walker shows the lower value for velocity. This may

be explained because, to walk with this device, the

patient has to stop, lift the AD and move forward,

performing an unnatural gait. On the other side, the

RFS has the higher values for velocity and cadence.

So, this device is the nearest to the healthy and can

be explained by the continuous movement of the

subject with the RFS.

Finally, the step length is almost identical for the

ADs, so devices preserve this feature. The little

increase of the step length for the crutches may be

reached by the fact that these patients have already

walked with crutches before this study.

In terms of the evaluation of the fall risk, in

(Kloos et al. 2012) it is mentioned that the

variability is an indicator of fall risk, which means

that the increase of variability increases the risk of

fall. Thus, the variability of the stride, stance and

swing time (Table 1) was analysed in this study for

further fall risk analysis. As one can see, the

standard deviation is lower for the gait with the RFS

than the others ADs. Therefore, it can be verified

that the RFS provides a greater stability for the

patient. Relatively to the parameters obtained by the

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

792

accelerometer at the trunk, it is known that the

greater they are, the greater is the risk and trend of

fall (Doheny et al. 2012). In Table 1, it is shown that

the crutches present the higher values, so these

devices are of higher risk to the user and the RFS

produced the lower values, except for RMS AP.

Considering the AP signal, it corresponds to the

forward and backward movement of the trunk. One

can see in Figure 5 a) that for the three devices the

signal is positive, meaning that the trunk is leaning

forward when the patient walks. However, the signal

is much higher for the RFS than the others ADs.

This happens because the RFS has to be pushed and

by observation of the authors, users had to lean

forward while pushing this device. This factor can

be due to an incorrect walker height adjustment.

Thus, the AP signal can be an important indicative

for posture correction as well as walker height

adjustment. Therefore, further studies will be

conducted to evaluate this potential clinical

indicator. In terms of ML signal in Figure 5 b), the

RFS presented the lower values, relatively to the

other ADs. The ML movement is a little attenuated

by the RFS and this happens because the user is

supported by the forearm support of RFS, preventing

the trunk oscillation in this direction.

Finally, relatively to the variability of the signals

of the trunk, it is shown in Table 1, that crutches

present a higher standard deviation for all of the

parameters, except for the sway range ML and RMS

AP. The first is higher for the standard walker. The

latter is higher for the RFS as expected, because of

the leaning of the trunk to the front.

To conclude, the authors see the RFS as the best

device for these patients since it provides higher

stability to the users, less risk of fall, a more natural

gait and a continuous movement.

5 CONCLUSION

This work used accelerometers located at the injured

leg’s ankle and trunk to verify and validate the

association between the accelerations signals and the

gait events, detect gait parameters and assess the fall

risk in assisted-gait with crutches, standard walker

and RFs. Further, it was possible to determine

efficiently all the proposed gait parameters in all

devices with patients diagnosed with KOA.

Additionnaly, it can be verified that the RFS

provides a greater stability, reducing the risk of fall

and inducing a more natural gai performance.

ACKNOWLEDGEMENTS

This work has been supported by FCT – Fundação

para a Ciência e Tecnologia in the scope of the

project: PEst-OE/EEI/UI0319/2014.

REFERENCES

Debi, R. et al., 2011. Correlation between single limb

support phase and self-evaluation questionnaires in

knee osteoarthritis populations. Disability and

rehabilitation, 33(13-14), pp.1103–9.

Doheny, E. P. et al., 2012. Displacement of centre of mass

during quiet standing assessed using accelerometry in

older fallers and non-fallers. Annual International

Conference of the IEEE Engineering in Medicine and

Biology Society. 2012, pp.3300–3.

Elbaz, A. et al., 2012. Can single limb support objectively

assess the functional severity of knee osteoarthritis?

The Knee, 19(1), pp.32–5.

Henriksen, M. et al., 2004. Test – retest reliability of trunk

accelerometric gait analysis. Gait & Posture, 19,

pp.288–297.

Kaufman, K. R. et al., 2001. Gait characteristics of

patients with knee osteoarthritis. Journal of

biomechanics, 34(7), pp.907–15.

Kloos, A. et al., 2012. The impact of different types of

assistive devices on gait measures and safety in

Huntington’s disease. PloS one, 7(2), p.e30903.

Lee, J.-A. et al., 2010. Portable activity monitoring system

for temporal parameters of gait cycles. Journal of

medical systems, 34(5), pp.959–66.

Martins, M. et al., 2013. Assessment of walker-assisted

human interaction from LRF and wearable wireless

inertial sensors. In Internation Congress on

Neurotechnology, Electronics and Informatics.

Neurotechnix. pp. 1–8.

Mathie, M. J. et al., 2004. Accelerometry: providing an

integrated, practical method for long-term, ambulatory

monitoring of human movement. Physiological

Measurement, 25(2), pp.R1–R20..

Sabatini, A. M. et al., 2005. Assessment of Walking

Features From Foot Inertial Sensing. IEEE transactions

on bio medical engineering, 52(3), pp.486–494.

Vaughan, C., Davis, B. & Connor, J. C. O., 1999.

Dynamics of Human Gait 2nd ed. C. Vaughan, ed.,

Cape Town, South Africa: Kiboho Publishers.

Watanabe, T. et al., 2011. A preliminary test of

measurement of joint angles and stride length with

wireless inertial sensors for wearable gait evaluation

system. Computational intelligence and neuroscience,

2011, p.975193.

DetectionofGaitEventsandAssessmentofFallRiskUsingAccelerometersinAssistedGait

793