Assessment of Walker-assisted Human Interaction from LRF

and Wearable Wireless Inertial Sensors

Maria Martins

, Carlos Cifuentes

, Arlindo Elias

, Valmir Schneider

Anselmo Frizera

and Cristina Santos

Industrial Electronics Department, Minho University, Guimarães, Portugal

Electrical Engineering & Biotechnology Departments, Federal University of Espirito Santo, Vitória-ES, Brazil

Keywords: Man-machine Interaction, Assisted Ambulation, Gait Analysis.

Abstract: This paper describes the assessment of basic walker-assisted human interaction based on a laser range finder

(LRF) sensor and two inertial wearable sensors. Thirteen osteoarthritis patients and thirteen healthy subjects

were selected to be part of this pilot experiment, which intends to acquire and calculate spatiotemporal and

human-interaction parameters from walker-assisted ambulation. A comparison is made between the

spatiotemporal parameters of healthy subjects and the ones of patients with osteoarthritis. Moreover, it is

made an analysis of the effect that change of direction in walker-assisted ambulation can have on

spatiotemporal parameters. Results have shown that 1) velocity, step length and distance to the walker are

significantly affected by the change of direction, and 2) distance to the walker and step length can

distinguish between healthy subjects and patients with osteoarthritis. In terms of human-interaction

parameters, results show that a LRF sensor can correctly describe the trajectory and velocity of the user in

relation to the walker. However, just the wearable sensors can characterize changes in direction. These

results will be further used in the development of a robotic control that intends to detect the user's intention

through LRF and inertial sensors, and respond accordingly.

1 INTRODUCTION

The increase of human average lifespan demands the

need for patient-care technologies. Patient-care

facilities and nursing homes provide a supporting

environment for those elderly and other patients with

motor disability but are labour intensive and hence

expensive and limited.

Currently, canes and wheelchairs are the most

used assistive devices. However, canes do not

provide enough support for the muscles and the use

of the wheelchairs may lead to lower limb muscle

atrophy (Martins et al., 2011). Therefore, research

started to focus on walkers, which are devices that

improve mobility and independent performance of

mobility-related tasks.

Individuals requiring walkers have a reduced

ability to provide the supporting, stabilizing,

propulsion or restraining forces necessary for

forward progression. By decreasing the weight

bearing on one or both lower limbs, walkers may

help these individuals, alleviating pain from injury

or clinical pathology such as osteoarthritis

(Martins et al., 2011).

However, some problems have been reported in

the literature (Bateni and Maki, 2005) regarding

such devices, related to the lack of security and the

cognitive demands. Users must take overly cautious

steps not to push it out too far forward and they are

unsafe to use on uneven/slope terrain.

Thus, researchers on the robotics field started to

investigate how to promote safe mobility, and tried

to standardize and create an effective way to assess

and evaluate human-robot interaction in assisted-

walker gait. In this context, the Smart walkers

emerged (Martins et al., 2011), conventional walkers

adequately instrumented for control purposes, such

as the inference of the walker’s user intent in order

to control its speed, direction and distance

accordingly.

Research often addresses the study of interfaces

that try to recognize the user’s movement and/or

intent without requiring exhaustive manual

operations. Examples include recognition using

cameras (Martins et al., 2011), detection of human

gait using force sensors (Frizera et al., 2010) and

143

Martins M., Cifuentes C., Elias A., Schneider V., Frizera A. and Santos C..

Assessment of Walker-assisted Human Interaction from LRF and Wearable Wireless Inertial Sensors.

DOI: 10.5220/0004624201430151

In Proceedings of the International Congress on Neurotechnology, Electronics and Informatics (SensoryFusion-2013), pages 143-151

ISBN: 978-989-8565-80-8

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

ultrasonic sensors (Kuan et al., 2010).

JaRoW (Lee et al., 2011) was developed to

provide potential users with sufficient ambulatory

capability in all directions and easy-to-use features.

This walker was integrated with laser range finder

(LRF) sensors to detect the location of user’s lower

limbs in real time (Lee et al., 2011). A Kalman and

particle filters were applied to estimate and predict

the locations of the user’s lower limbs and body, in

real time. A PID controller was used that, despite the

good results, it is not certain to be effective when

tested with elderly people, whose behavior has

unpredictable changes, affecting the Jarow

dynamics. In addition, the rotation detection

algorithm, that detects when the user wants to curve,

is based on pre-defined limits that could not be

respected when dealing with elder people, thus

generating false decisions.

In (Ochi et al., 2011), it is proposed a walking

assist system for a body weight support walker

NILTWAMOR to track the walking trajectory of the

user. By using the LRF range sensor, the body center

point (BCP) of the user is estimated and used to

control both the gait velocity and the direction of the

user. The relationships between the facing direction

of the body and the inclination angles of both legs

are taken into account to control the walker’s

direction. However, the manuscript does not discuss

in detail the obtained results. Besides that, tests were

performed with normal healthy subjects.

Despite these studies, no attention has been given

to a quantitative evaluation of human-robot

interaction, i.e. to infer which signals related with

posture orientation and gait pattern can detect user’s

intentions while guiding the walker. Moreover, this

evaluation should be made with target users, like the

elder and other patients with motor disabilities.

So, the challenge to find a more reliable manner

to control the walker remains. As a first step, it is

necessary to access and analyse in detail the signals

of user-walker interaction to infer which ones are

better suited to indicate velocity and orientation

intentions of the user. Afterwards, it is possible to

develop a natural user interface between the walker

and patients and to employ a simple closed-loop

control without requiring any demanding cognitive-

effort from the patient.

In this context, this paper intends to access, study

and analyse basic walker-assisted human interaction

parameters of a walker model with forearm support

with knee osteoarthritis (OA) patients. For this, it

was used a LRF sensor placed on the walker-lower

base and two wearable inertial sensors: one mounted

on the walker and the other placed on the patient’s

body. Specifically, this paper aims to specify and

justify which interaction parameters are better to

interpret user’s velocity and orientation intention, to

then advance, in the next studies, for the

development of a robotic control. The human-walker

interaction measurements consist on the acquisition

of the: Distance between the user body center point

(BCP) and the walker; Angle of BCP orientation

relatively to the walker; Angle between linear

velocity vector and human-walker interaction line;

Angular velocity of the user and Linear velocity of

the user.

It will also be presented a gait evaluation based

on spatiotemporal parameters extracted from the

built-in LRF sensor. This evaluation intends to

detect the effect that a change in direction (making a

curve) has on spatiotemporal parameters. The

calculated spatiotemporal parameters were the gait

cycle, identification of stance and swing phases,

cadence and step length. These parameters were

chosen with base on previous studies (Debi et al.,

2009; Debi et al., 2011; Elbaz et al., 2011) that

compared knee OA subjects with healthy ones

walking without assistance. In those studies it was

suggested that an objective measurement tool such

as spatiotemporal parameters can help in evaluating

knee OA severity, effectiveness of treatment and

might help in disease management. Thus, the

calculation of these parameters with LRF sensor can

be a useful tool in the future to diagnose this type of

patients in assistance gait.

Results were derived from thirteen knee OA

subjects and thirteen normal subjects (control

samples). It is noteworthy that this study was done

with the motors shut down, to enable the evaluation

of the real interaction between the user and the

walker without the interference of any control

strategy.

This paper is organized as follows. Section 2

describes the methodology of this work, where it is

presented the walker and sensors system, the

experimental procedure and the human-walker

interaction parameters. Section 3 presents the

acquired results patients and normal subjects, and

provides for a discussion. Finally, conclusions are

presented in Section 4.

2 METHODS

2.1 Participants

For this study, 40 patients were chosen for inclusion

of individuals of both sexes over 55 years of age,

NEUROTECHNIX2013-InternationalCongressonNeurotechnology,ElectronicsandInformatics

144

able to walk unaided for at least 25 meters, not

having done any rehabilitation treatment for at least

2 months and not be making use of painkillers in the

last 7 days. Diagnosis of Osteoarthritis (OA) was

based on clinical and radiographic criteria of the

American College of Rheumatology, which confer

91% sensitivity and 86% specificity for the

diagnosis (Altman et al., 1986)performed by an

orthopedic surgeon with over 30 years' of experience

in evaluating patients with osteoarthritis and surgery

Total Knee Arthroplasty (TKA), aided by a physical

therapist with 10 years of experience in manual

therapy and functional assessment.

Exclusion criteria of subjects during the selection

process were: recent traumas; history of previous

surgery of the lower limbs, pelvis or lumbar spine;

neuromuscular diseases, other pathological forms of

arthritis, presence of neurologic sequel; and

cardiovascular diseases that contraindicate the

performance of experiments.

At the end of recruitment, 13 participants met all

inclusion criteria (Figure 1).

The control group consisted on 13 healthy

volunteers without any dysfunction on the lower

limbs.

Subjects read and signed an information and

consent form, which was approved by the Federal

University of Espirito Santo’s Health Science Center

Ethics Board.

Figure 1: Diagram of selection and exclusion of study

knee OA patients.

2.2 Protocol

2.2.1 Walker and Data Acquisition Systems

The Smart Walker is presented in Figure 2. This new

robotic walker consists basically of a mechanical

structure with an adaptable height to support the user

in the forearms.

The developed acquisition system consists of a

ZigBee Health Care (ZHC) network that has two

types of devices: three ZigBee End Devices (ZED)

and one ZigBee Coordinator, which is connected to

the PC and receives patient´s signal data from ZEDs

(Cifuentes,2010). The sensors are shown in Figure 2.

One ZED (C) is used to acquire and transmit

signals from the LRF sensor (scanning sensor

Hokuyo URG-04lx) (A) that is connected to a

system microcontroller (B) that performs legs’ path

detection (position and orientation). Specification of

LRF performance and the leg’s path detection

algorithm can be found in (Lee et al., 2011).

The other two ZEDs are integrated with IMU

sensors (ZIMUED) developed in previous research

(Cifuentes et al., 2010). One ZIMUED is located in

the trunk of the patient (D) and the other one is over

the walker (E). These sensors record orientation and

angular velocity of the user and walker.

The IMU signals are obtained every 50 Hz and

the LRF signals every 10 Hz.

Figure 2: Smart Walker hardware architecture.

2.2.2 Experimental Procedure

First, it was established that the walker should have

the motors shut down and the user should walk with

a self-selected speed, during assisted-ambulation.

This was important to obtain the preferred gait speed

of the subject while using the walker without

inducing any artificial motion patterns that could

bias the final results.

Height of the forearm-support is the other

parameter that has been established. It should be

equal to height measured between the elbow of the

user and the ground, trying to force an upright

posture.

All subjects (thirteen osteoarthritis patients and

thirteen normal subjects) were barefoot and asked to

walk three times along a pre-defined 8.9 meters path

(see Figure 3).

Figure 3: Trajectory that user’s performed with the walker.

AssessmentofWalker-assistedHumanInteractionfromLRFandWearableWirelessInertialSensors

145

2.2.3 Human-Walker Interaction

and Temporal Distance Parameters

Calculation

The parameters described in this section are

presented in Figure 4 and surveyed in Table 1.

The human-walker interaction parameters consist

on: Distance between the user body center point

(BCP) and the walker (d), Angle of BCP orientation

relatively to the walker (



), Angle between linear

velocity vector and human-walker interaction line

(



), Angular velocity of the user (wh), Linear

velocity of the user (vh) (Table 1).

Spatiotemporal parameters of gait were also

determined, which reflect the dynamic activity

during human walking: gait cycle (G), identification

of stance (ST) and swing phases (SW), steps length

(SL) and cadence (CAD). The selection of these

parameters was based on previous studies (Debi,

2009;Debi,2011;Elbaz,2011)that compared knee OA

subjects with healthy ones walking without

assistance. In those studies it was suggested that

spatiotemporal parameters are sufficient to evaluate

and manage the knee OA disease. Results of these

studies state that patients with knee OA walk slower;

have a shorter step length; shorter swing phase and

consequent longer stance phase.

Table 1: Human-walker interaction and spatiotemporal

parameters.

Variable Sensor

Human-Walker

Interaction

Parameters

Distance between

the BCP and the

walker

d LRF

Angle of BCP

orientation in

relation to the

walker

θ LRF

Angle between

linear velocity

vector and human-

walker interaction

line

LRF +

IMU

Angular velocity

of the user

wh IMU

Linear velocity of

the user*

Vh LRF

Spatiotemporal

Parameter

Gait cycle G LRF

Stance Phase ST LRF

Swing Phase SW LRF

Step Length SL LRF

Cadence CAD LRF

* Linear velocity of the user will be also considered as

Spatiotemporal parameter.

The detection and calculation methods of these

parameters are described in the next subsections.

i. Distance between the user’s BCP and the

walker (d) and Angle of BCP orientation in relation

to the walker (θ): The applied detection method of

the legs to calculate the position of the BCP, is based

on the work developed in (Lee et al., 2011). The

detection algorithm is divided into four basic tasks:

pre-processing of data, detection of transitions,

pattern’s extraction and estimation of the

coordinates of the legs. In the pre-processing phase

it is performed the delimitation of the region of

interaction. Then, in the detection of transitions

phase it is analyzed the performed laser scanning

and seeks to identify transitions that exceed a certain

threshold. In Figure 5a it is presented a situation

where four transitions are found (indicated by

arrows). These transitions are then stored. Finally,

the coordinates of each leg are estimated and the

algorithm starts to estimate the BCP. This, in turn, is

accomplished by taking the midpoint of the segment

that joins the coordinates of the legs, as illustrated in

In Figure 5b by the cross.

Thus, this algorithm detects the two legs, and the

midpoint of the segment that joins the coordinates of

Figure 4: Scheme of the interaction parameters. Variables

are defined in Table I. Black circle represents user’s BCP

and the yellow box represents the LRF sensor.

a) b)

Figure 5: a) Detection of transitions: 4 transitions; b)

Illustration of the CoM estimation (its location (x,y),

distance between user-walker (d) and orientation (θ)).

the legs is calculated as the BCP position. With this

information, one can know the coordinates (x,y) of

NEUROTECHNIX2013-InternationalCongressonNeurotechnology,ElectronicsandInformatics

146

the BCP, and consequently the distance (d)

orientation (θ) that the user is from the walker.

ii. Angle between linear velocity vector and human-

walker interaction line (ϕ), Angular velocity of the

user (wh): The IMU placed on the user’s CoM

provides for the user’s orientation (ψ). The IMU

placed on the walker provides for the walker’s

orientation (ψ’). From these two angles and θ one

determines (ϕ). Finally, the angular velocity (wh) is

obtained from the gyroscope located in the user’s

IMU.

iii. Linear velocity of the user (vh): This

parameter is the rate of change of the position of a

leg detected by the LRF and is given by:

vh = (ds1+ds2)/dt, (1)

where ds1 and ds2 are the peak-to-peak

amplitudes between two legs of the acquired LRF

sensor signal and dt is gait cycle time. Figure 6

illustrates these variables.

iv. Spatiotemporal parameters: Stance phase

(ST) (swing phase (SW)) begins (ends) when the

foot contacts with the ground and ends (begins)

when the same foot leaves the ground. Gait cycle

consists on the sum of stance and swing phases time.

In order to estimate these two parameters, it is

necessary to detect the foot strike moments during

each cycle. For the LRF signals, these correspond to

the minimum values (dots in Figure 6). In a step

cycle, the stance phase corresponds to the signal

going from the minima to the maxima. The swing

phase is the rest of the cycle.

The step length is the distance (in meters)

between a specific point of one foot and the same

point of the other foot. For the LRF signal, this is

calculated as the difference between the maximum

of one leg and the minimum of the other leg in the

same instant of time, i.e, it corresponds to ‘ds1’, for

example, in Figure 6.

The cadence is defined as the rhythm of a

person’s walk and is expressed in steps per minute

(step/min).

2.2.4 Statistical Methods

The spatiotemporal parameters were considered for

repeated measures ANOVA to test for significance.

A mixed design was used, with within-subjects of

direction (Forward/Curve) and a between-subjects

factor of type of subject (Healthy/Patient). The level

of significance was set at 5%.

3 RESULTS AND DISCUSSION

The values summarized in Table 2 represent the

average value of the calculated Spatiotemporal

parameters of each individual, as well as the distance

(d) human-walker interaction parameter. For the

patients (PTs), the values were calculated based on

the signal of the leg that suffers the most with

osteoarthritis. In the case of healthy individuals

(HIs), the right leg was the one analyzed (no criteria

of choice was used, since they are considered

symmetrical).

In both groups (PTs and HIs) the parameters

were separated by direction. This separation gives

the information of how they can be affected when

the user is changing his direction, by performing a

curve after going forward.

Figure 6: LRF sensor signal of the user’s distance to the

walker. The dots indicate the minimum points that

correspond to foot strike events.

Figure 7 shows the acquired LRF and IMU signals,

while one of the PTs (PT #3) walks with the walker

following the pre-defined path. ‘ψ Angle’ and

‘Angular Velocity’ represent the signals read by

IMUs placed both on the walker and PT. ‘Legs

Distance’ graph illustrates the distance of both PT

legs from the walker and ‘Legs Orientation’ shows

the orientation of each leg relatively to the walker.

‘Human linear velocity (vh)’, ‘Human and

Walker Orientation’ and ‘θ and ϕ’ graphs depict data

calculated from the previous graphs only when the

legs of the PT crossed (this event is represented by

circles in the previous graphs) and they are

represented in strides and steps to better analyze

them. It is noteworthy that these graphs present

discontinuities since they are calculated in the

specific event of crossing legs.

In the following subsections it will be presented

and discussed these results in detail.

3.1 Spatiotemporal Parameters

In the ‘Legs Distance’ graph in Figure 7, the

diamonds and crosses identify the beginning and end

AssessmentofWalker-assistedHumanInteractionfromLRFandWearableWirelessInertialSensors

147

of stance, respectively. These instants allowed

calculating spatiotemporal parameters, as shown in

Figure 6 an

d explained in section 2.2.3.

By analysing Table 2, it is possible to verify that

PTs present longer duration of stance, slower

velocity and shorter step length than HIs. This is in

accordance with the results presented in (Debi et al.,

2009; Debi et al., 2011; Elbaz et al., 2011), where a

complete evaluation and comparison of

spatiotemporal parameters was made between knee

OA patients and healthy subjects without assistance.

In addition, PTs tend to be closer to the walker

than HIs. This happens since PTs tend to be more

supported on the walker, in order to feel more

comfortable, safe and to alleviate knee pain. HIs

tend to be more deviated from the walker since they

do not need an extra support to walk.

Since this study intends to compare the

spatiotemporal parameters between the two types of

subjects, PTs and HIs, and between two types of

direction, forward and curve, it was made a repeated

measures ANOVA to test for significance.

Gait cycle (G) does not present main changes

due to direction (p=0.252) and type of subject

(p=0.655) and no significant interactions between

direction and type of subject (p=0.222). However, it

tends to increase in HIs and decrease in PTs. This

happens because, when HIs perform a curve they

prolong their step; and PTs tend to reduce their time

with the feet on the ground, increasing the number

of steps.

Stance (ST) and Swing (SW) also do not present

main changes due to direction (p=0.644/p=0.640)

and type of subject (p=0.935/p=0.931) and no

significant interactions between direction and type

of subject (p=0.316/p=0.317).

Despite the lack of statistical significance, one

can observe that PTs present lower ST duration and

higher SW duration when compared with HIs. This

was expected since in (Debi et al., 2009; Debi et al.,

2011; Elbaz et al., 2011)the SW was highlighted as

an objective parameter in the comprehensive

evaluation of a PT. They referred that SW may serve

as a simple follow-up measurement in patients with

OA. This importance is given because a knee OA

patient attemptstoavoidpainwhilewalkingby

decreasingloadsfromtheaffectedjoint.However,

SW parameter has no statistical significance in the

current study with assisted gait. This means that as

PT is better supported with the walker, he feels less

pain when loading the affected joint, achieving to

spend more time with the feet on the ground. This

explains the little difference of SW that exists

between PTs and HIs.

One can also observe that PTs tend to have lower

ST and higher SW when performing a curve in

comparison with the forward direction. These events

are opposite to what happens with HIs and can be

related to the confidence and sense of security HIs

have when manoeuvring the walker. So, this

decrease of ST (and consequent increase of SW) in

PTs can be related to the difficulty that PTs have

when performing a curve. Some of them complained

to feel more pain on the knee and some confusion.

Thus, they tend to support less time the foot on the

ground, becoming more suspended on the walker

(this was also observed in the gait cycle parameter).

Cadence (CAD) is a parameter that also presents

no main changes due to direction (p=0.415) and type

of subject (p=0.519) and no significant interactions

between direction and type of subject (p=0.174).

However, this parameter tends to decrease in HIs

cases, and increase, in PTs cases, when performing a

curve. As it was already discussed, when performing

a curve, PTs tend to increase the number of steps

and decrease their length.

The velocity parameter (vh) shows to be affected

by direction (p=0.010), but not to type of subject

(p=0.264). It also does not present significant

interactions between direction and type of subject

(p=0.140). So, this means that all subjects reduced

their vh when performing a curve. Since it is a

change in the path and more difficult to perform than

to walk in straight line, it is understandable that

subjects tend to reduce their velocity.

Step length (SL) is affected by within-subjects

factor (p=0.014) and between-subjects factor

(p=0.000), as well as their interaction (p=0.016).

The distance to the walker (d) is affected by the type

of subject (p=0.017) and by the interaction between

direction and type of subject (p=0.036). However, it

does not present main changes due to direction

(p=0.477).

PTs tend to observe before walk when a curve

appears, i.e. first they turn the walker and then they

follow it. This causes an increased deviation from

the walker (increases d) and a consequent increase

of SL.

So, these spatiotemporal parameters, vh, SL and

d, are important to be analyzed in these two

situations. They can objectively inform about the

level of difficulty and sense of security that PTs with

osteoarthritis sense when maneuvering the walker,

and this will depend on: comfort to guide the walker;

pain on the knee, which influences the type of curve

(close or open curve); and security and confidence

on the device.

It can also be possible to differentiate between

PTs and HIs by analyzing d and SL.

This is not in accordance with previous studies

(Debi et al., 2009; Debi et al., 2011; Elbaz et al.,

2011)where SW was recommended as an objective

parameter to evaluate the degree of knee pain. Since

the current study evaluated OA patients walking

NEUROTECHNIX2013-InternationalCongressonNeurotechnology,ElectronicsandInformatics

148

with assistance, the conditions changed as the

patient has now an extra support that helps in

alleviating pain. By this SW values do not differ

significantly from the healthy subjects.

Table 2: Spatiotemporal parameters of walker-assisted

gait. Average±Standard Deviation values.

Subj.

PT HI

Direction Forward Curve Forward Curve

G (s) 1,52±0,160 1,51±0,256 1,51±0,028 1,63±0,057

ST (%) 58,06±6,535 56,72±7,395 55,33±2,350 58,89±1,922

SW (%) 41,92±6,531 43,24±7,388 44,67±2,354 41,11±1,922

SL (m) 0,22±0,046 0,21±0,032 0,37±0,062 0,31±0,0455

vh (m/s) 0,32±0,158 0,3±0,147 0,44±0,061 0,39±0,047

CAD

(step/min)

79,54±7,895 81,01±13,146 79,13±0,01 73,15±2,547

d (m) 0,44±0,052 0,46±0,053 0,55±0,067 0,53±0,051

3.2 Human-walker Interaction

Parameters

In the ‘ψ Angle’ graph of Figure 7, one can see that

the IMU’s signals provide information about the PT

movement. He is going in straight line and then at

t=1s, he begins to make a curve. Then, at t=3s, he

goes again straight and makes a curve, at t=5s, for

the other side until t=8s. From t=8s to t=10s, he

continues to walk forward and straight.

The ‘Angular Velocity’ graph (Figure 7)

indicates that he increases (in absolute) its angular

velocity (wh) when he starts to curve, by analyzing

the same instants of time as previously.

Therefore, these two parameters can be used to

correctly detect the path that the user is following. In

‘Legs Distance’ graph (Figure 7), one can see that is

hard to distinguish between going forward and

making a curve. However, it can be noticed that

maximum values of right leg are reduced when PT

makes the first curve (t=1s to t=3s). However, this

change is not perceptible or significant in the second

curve.

After observing ‘Legs Distance’ signals from all

the patients, it was concluded that there is a great

variability on this signal. Which means that PTs can

perform a curve in different manners: some hide one

leg; others fend off the legs, or bring them together.

‘Legs Orientation’ (Figure 7) also presents small

changes during the time PT is performing a curve

(t=[3 4]s and t=[5 8] s). Once again, this signal

presents a great variability through PTs.

A possible solution to increase the effects of

making a curve on the LRF signal would be to put

the LRF up to the foot’s height, to detect their

direction. However, this is not possible to detect

with LRF sensor, because the signal becomes

distorted and poor of information. So, the utilization

of a camera, for example, could be a good solution

to detect the feet’s direction.

Thus, LRF sensor is good to detect

spatiotemporal parameters, as it was analyzed

before, but not too good to detect intention of

changing direction.

Moreover, LRF sensor is essential to detect when

legs are crossing with each other (identified by

circles on the graphs). This is an important event to

detect BCP position, since in these instants it is the

midpoint between the legs.

So, Human-Walker Interaction parameters can be

calculated every time the legs cross and are

represented in Figure 7.Distance between the user

and the walker (d) is acquired by the ‘Legs

Distance’ signal and it is marked with circles. Angle

of BCP orientation in relation to the walker (θ) is

acquired by the ‘Legs Orientation’ signal, being the

midpoint between each leg orientation, and is

represented in ‘θ and ϕ’ graph. Angle between linear

velocity vector and human-walker interaction line

(ϕ) is calculated by the sum of ϕ angle of walker and

ψ angle of human, both represented in ‘ψ Angle’

graph and θ. This angle is represented in ‘θ and ϕ’

graph by the designated signal ‘ϕ’. Angular velocity

of the user (wh) are the points marked with a circle

in the ‘Angular Velocity’ graph . Linear velocity of

the user (vh) depends on the time that the user takes

to complete a stride (two steps) and is shown in

‘Human Linear Velocity’ graph.

Looking at ‘Human Linear Velocity’ graph

(Figure 7), one can see that vh decreases when

making a curve, which is in accordance with

previous discussion.

Through ‘Human and Walker Orientation (ψ)’

(Figure 7), one can see that the walker turns first

than the human. This could indicate that the

intention of command is transmitted by the upper

limbs. This needs to be further studied by placing a

rotating handlebar with integrated IMU or force

sensors.

In ‘θ and ϕ’ graph (Figure 7), one can see that ϕ

is better to identify, with significant variability, the

orientation of the subject when compared with θ.

In conclusion, the Human-Walker Interaction

parameters, in the overall are correctly detected and

can describe the interaction between the PT and the

walker.

AssessmentofWalker-assistedHumanInteractionfromLRFandWearableWirelessInertialSensors

149

Figure 7: Human-Walker Interaction measurements with

data acquisition systems.

4 CONCLUSIONS

In the literature there are few studies of walker-

assisted biomechanics, especially regarding walkers

with forearm supports, and there are none describing

human-walker interaction nor gait evaluation

regarding type of direction In relation to

spatiotemporal parameters, the analysis has shown

that 1) velocity, step length and distance to the

walker are significantly affected by the change of

direction, and 2) distance to the walker nd step

length can distinguish between healthy subjects and

patients with osteoarthritis. The Human-Walker

Interaction parameters were correctly detected. LRF

signals can detect the necessary event (when legs are

crossing) to calculate them. However, is necessary,

in further studies, to develop an algorithm, like the

one in (Lee et al., 2011), to track the PT’s legs.

Afterwards, it is intended to advance for the

development of a control strategy with these

parameters and based on the cinematic of the walker

illustrated in Figure 4.

The difficulty in human commands acquisition

for the development of a control strategy is to find a

parameter that can give the information of user’s

orientation, to then detect the orientation commands

(go to the left/right).

Hence, this control should be based on the

minimization of ϕ (should tend to zero), since it was

concluded that this parameter can detect PT’s

orientation. However the authors are not sure if it is

the correct parameter to estimate a change of

direction, since PTs demonstrate to first use the

upper limbs to transmit that command. This problem

will be analysed in further studies.

ACKNOWLEDGEMENTS

This work is financed by FEDER Funds and through

Programa Operacional Fatores de Competitividade

– COMPETE and by National Funds through FCT -

Fundação para a Ciência e Tecnologia under the

Project: FCOMP-01-0124-FEDER-022674. Work

supported by Portuguese Science Foundation (grant

SFRH/BD/76097/2011). The Brazilian National

Council for Scientific and Technological

Development (CNPq) supported this work (Process

# 471804/2012-6).

REFERENCES

Altman R., Asch E, Bloch D., Bole G, Borenstein D,

Brandt K., Christy W., Cooke T. D., Greenwald R.,

Hochberg M, et al. Development of criteria for the

classification and reporting of osteoarthritis.

Classification of osteoarthritis of the knee. Diagnostic

and Therapeutic Criteria Committee of the American

NEUROTECHNIX2013-InternationalCongressonNeurotechnology,ElectronicsandInformatics

150

Rheumatism Association. Arthritis Rheum. 1986

Aug;29(8):1039-49.

Bateni, H., Maki B: Assistive devices for balance and

mobility: benefits, demands and adverse

consequences. Arch Phys Med Rehabil vol 86, jan

2005. Pp 134-144.

Cifuentes, C., Gentiletti, G., Suarez, M. J., Rodríguez, L.

E., “Development of a Zigbee platform for

bioinstrumentation,” in Proceedings of the 2010

Annual International Conference of the IEEE

Engineering in Medicine and Biology Society, 2010,

pp. 390-393.

Debi R., Mor A., Segal O., Segal G., Debbi E., Agar G.,

Halperin N, Haim A, Elbaz A. Differences in gait

patterns, pain, function and quality of life between

males and females with knee osteoarthritis: a clinical

trial. BMC Musculoskelet Disord. 2009 Oct 13;10:127.

doi: 10.1186/1471-2474-10-127.

Debi R., Mor A., Segal G., Segal O., Agar G., Debbi E.,

Halperin N, Haim A, Elbaz A. Correlation between

single limb support phase and self-evaluation

questionnaires in knee osteoarthritis populations.

Disabil Rehabil. 2011;33(13-14):1103-9. doi:

10.3109/09638288.2010.520805. Epub 2011 Jan 5.

Elbaz A., Mor A., Segal O., Agar G., Halperin N., Haim

A, Debbi E, Segal G, Debi R. Can single limb support

objectively assess the functional severity of knee

osteoarthritis?. Knee. 2012 Jan;19(1):32-5. doi:

10.1016/j.knee.2010.12.004. Epub 2011 Jan 26.

Frizera A., Gallego J. A., Rocon E., Pons J. L, Ceres R.:

Extraction of user’s navigation commands from upper

body force interaction in walker assisted gait.

Biomedical Engineering Online 2010, 9(1):37.

Kuan-Ting Yu; Chi-Pang Lam; Ming-Fang Chang; Wei-

Hao Mou.; Shi-Huan Tseng; Li-Chen Fu. An

Interactive Robotic Walker for Assisting Elderly

Mobility in Senior Care Unit, 2010.

Lee, G., Jung, E., Ohnuma, T., Chong ,N., Yi, B. JAIST

Robotic Walker Control Based on a Two-layered

Kalman Filter. IEEE International Conference on

Robotics and Automation, Shangai, China, 2011, pp.

3682–3687.

Martins, M., Frizera, A., Santos, C., Ceres, R..

(2011).Assistive Mobility Devices focusing on Smart

Walkers: Classification and Review. Robotics and

Autonomous Systems. DOI: 10.1016/j.robot.

2011.11.015.

Ochi, Y.; Terashima, K.; Kubo, K.; Miyoshi, T.Tracking

control to human gait of lift-type walking assist robot

using range sensor. Proceedings of the 14th

International Conference on Climbing and Walking

Robots and the Support Technologies for Mobile

Machines, 2011.

AssessmentofWalker-assistedHumanInteractionfromLRFandWearableWirelessInertialSensors

151