Preliminary Study on the Design of a Low-cost Movement

Analysis System

Reliability Measurement of Timed Up and Go Test

Asma Hassani

, Alexandre Kubicki

, Vincent Brost

and Fan Yang

LE2I CNRS 6306 Laboratory, University of Burgundy, Dijon, France

INSERM U1093 Laboratory, University of Burgundy, Dijon, France

Keywords:

Geriatric Health, TUG Test, Autonomy Assessment, Kinect, Spatio-temporal Movement Parameter Extrac-

tion, 3D Real-time Video Processing.

Abstract:

In this paper, we present experiments on the design of a novel movement analysis system for real-time balance

assessment in the frail elderly. Using the Microsoft Kinect sensors, we capture TUG (Timed Up and Go) tests

and mainly analyze the transfer from sitting-to-standing and back-to-sitting which represent two of the most

commonly executed human movements. Nine spatio-temporal parameters were extracted from recorded joint

positions by 3D skeletal sequence processing. In order to validate and evaluate the developed system, practical

test experiences have been performed on ten healthy young subjects, who were asked to realize the TUG in

three different conditions: normal, cognitive and motor. Obtained results showed good measurement reliability

and reproducibility with important precision. In addition, we observed that even for young healthy subjects,

there is a signiﬁcant difference of movement parameter between normal condition and cognitive condition,

which represents a stimulating result in the dual task paradigm ﬁeld. This preliminary study opens a new

research and development way for geriatric health which implies multiple aspects: user-friendly, hygiene,

low-cost, home-based environment, and automatic autonomy assessment.

1 INTRODUCTION

Geriatric rehabilitation has become a major public

health issue. The balance assessment in the frail el-

derly consists to assess functional capacity and fall

risk, and help determine the underlying reasons for

balance disorders. Also, it can allow to objectify

progress or deterioration and measuring the patient’s

autonomy. Thus, many strategies of balance assess-

ment are developed. Most of these strategies involve

complex operations to perform in clinical environ-

ment and require the presence of a health professional

to determine the score for each person.

Clinical balance assessment can be classiﬁed into

three broad categories: system assessments, quanti-

tative assessments and functional assessments. The

system assessments are helpful to determine the un-

derlying causes of the balance deﬁcit. However, all

balance rating scales are relatively course measures

of complex motor behaviour and all subjective assess-

ments can easily suffer from tester bias (Mancini and

Horak, 2010). Regarding quantitative assessments,

several systems are being exposed to the market, but

they are not scientiﬁcally validated before being of-

fered to therapists, except some systems such as the

static or dynamic posturography platforms. In gen-

eral, functional balance tests assess performance on a

set of motor tasks on a 3 to 5 point scale or use a stop

as watching to time how long the subject can keep up

balance in a speciﬁc posture. Several clinical tests ex-

ist such as Tinetti and TUG. This functional approach

is usually used to detect balance problem existence.

However, clinical balance assessments give subjective

results that are usually not responsive enough to rate

low progress or deterioration in a subject’s ability to

balance (Mancini and Horak, 2010). Indeed, clinical

human evaluation is limited in terms of the parame-

ters evaluated.

In elderly autonomy and independence maintain-

ing context, it is very interesting to realize functional

balance assessments automatically in home-based en-

vironment using artiﬁcial vision technology. If an in-

novative device associated with video processing is

capable to assess the motor abilities of the frail el-

derly, health professionals could be alerted in case

of deterioration. The earlier balance problem detec-

tion and the precocity of rehabilitation could allow,

through a primary or secondary prevention, prolong-

662

Hassani A., Kubicki A., Brost V. and Yang F..

Preliminary Study on the Design of a Low-cost Movement Analysis System - Reliability Measurement of Timed Up and Go Test.

DOI: 10.5220/0004747806620667

In Proceedings of the 9th International Conference on Computer Vision Theory and Applications (VISAPP-2014), pages 662-667

ISBN: 978-989-758-004-8

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

ing autonomy and independence of elderly. In addi-

tion, we can increase the number of studied parame-

ters with fairly accurate measurement, by balance as-

sessment automation.

This paper is organized as follows: in section 2

some related works of human motion analysis for re-

habilitation by computer vision are cited. Section 3

presents the proposed real-time system from data ac-

quisition to feature parameter extraction after a brief

description of the TUG test. Section 4 illustrates ex-

periment tests and obtained analysis results with ten

healthy young subjects. Finally, we discuss and con-

clude this preliminary study in section 5.

2 RELATED WORK

Several methods have been developed for human mo-

tion analysis and applied to rehabilitation research.

For example, marker-based systems (MBS) usually

use infrared cameras to detect markers placed on the

patients’ bodies. These systems are accurate and

yield highly robust measurements of a person’s mo-

tion. However, MBS involve frequently the use of

expensive specialized equipment and are impractical

to move. Furthermore, passive or active markers must

be correctly placed on the body before each capture

session which makes such systems obtrusive and in-

appropriate for home-based rehabilitation.

In (Hagler et al., 2010), the authors propose a sys-

tem for continuous and unobtrusive in-home assess-

ment of gait velocity which is based on estimating

walking speed from noisy time and location data col-

lected by a ”sensor line” of restricted view passive in-

frared motion detectors. Indeed, sensor suites provide

information upon the daily activity levels of moni-

tored subjects, and arrays of such sensors allow ob-

taining velocity measurements on a continuous basis

in home settings. However, such systems do not yield

measurements of the detail necessary for assessment

of fall risk, particularly, spatial and temporal gait pa-

rameters (other than walking speed), TUG duration

and sit-to-stand time.

Recent pervasive/ubiquitous healthcare and wire-

less sensor networks incorporate multimodal sensing

and computer vision. The pervasive and ubiquitous

computing aim to integrate information and comput-

ing into the everyday physical world, so that this tech-

nology is available to everyone in any context. Wire-

less sensor network nodes increasingly include iner-

tial sensors such as accelerometers and gyroscopes

which have been used for activity detection and ges-

tural recognition.

Several studies identify the Kinect’s potential for

use in rehabilitation. Indeed, Kinect is an inexpen-

sive and portable sensor that includes a RGB cam-

era, a depth sensor and a multi-array microphone. It

provides full-body 3D motion capture, facial and ges-

ture recognition. Although the accuracy of Kinect is

limited, it still provides a good tradeoff between cost

and portability and could be used at home for reha-

bilitation exercises (Bonnech

ere et al., 2012). Re-

cently, Kinect has been shown to be able to create

a 3D human model with similar accuracy to expen-

sive and complex 3D body scanning systems (Weiss

et al., 2011). Moreover, it allows to evaluate gait

velocity (Stone and Skubic, 2011), hand and elbow

movements and anatomical landmark displacement

and trunk angle during commonly performed clini-

cal tests of postural control that represent favorable

results when compared to some existing 3D motion

analysis systems. In (Gabel et al., 2012), a method

for full body gait analysis using Kinect is presented. It

consists in applying a supervised learning approach to

automatically and accurately extract a set of gait pa-

rameters, using the 3D virtual skeleton as the input to

the learned model. Likewise, Kinect is able to obtain

some kinematic and anatomical mapping data with

a similar degree of accuracy to more expensive 3D

motion analysis and scanning systems (Clark et al.,

2012).

In this paper, we propose a low-cost, robust and

home-based system for real-time balance assessment

in the frail elderly. Our system can accurately mea-

sure several parameters that have been shown rele-

vant for balance assessment using the Kinect sensor

and Software Development Kit. It automatically and

accurately extracts shoulder kinematics and TUG du-

ration by the use of the 3D skeleton, with robustness

to environmental changes and variations in the place-

ment of the Kinect sensor. We demonstrate how a

rich set of parameters can be extracted. Furthermore,

we suggest that the proposed method is not cumber-

some since in a typical use-case a Kinect sensor can

be placed in a ﬁxed position at home. In addition, our

system requires no human intervention and no mark-

ers or wearable sensors will be attached to the body.

3 FUNCTIONAL BALANCE

ASSESSMENT SYSTEM

DESIGN PRESENTATION

3.1 Timed Up and Go Test Description

The Timed Up and Go test is a clinical measure of

balance and mobility in the elderly and in neurolog-

PreliminaryStudyontheDesignofaLow-costMovementAnalysisSystem-ReliabilityMeasurementofTimedUpandGo

Test

663

ical populations. The TUG is relatively simple, re-

quires no special equipment or training and is easily

included as part of the routine medical examination.

It consists of a sequence of sit-to-stand (STS), walk-

ing a distance of 3 m, turning and back-to-sit (BTS).

The total time taken to complete the TUG test allows

predicting the risk of falling. A score of 12 s is con-

sidered as the upper limit of normal mobility. Abnor-

mal mobility was deﬁned as having a TUG score of

>= 20 s. The TUG has been modiﬁed to add a sec-

ondary task that could be cognitive or motor. In the

TUG-cognitive, subjects were asked to complete the

test while naming some names that begin with a cer-

tain letter and the TUG-motor consists of completing

the TUG while carrying a cup of water. A score of 15

s on the TUG-cognitive and 14,5 s on the TUG-motor

is associated with increased risk of falls (Mancini and

Horak, 2010).

In general, the TUG rests upon one time measure

to evaluate the overall performance of a sequence of

tasks. However, it can provide speciﬁc information

on components of each task that could disclose more

speciﬁc mobility problems. The TUG includes two

actions that are commonly executed throughout the

different stages of the human lifetime: STS and BTS.

For both STS and BTS, the shoulder was the ﬁrst to

move and the last to stop, and it moved in the sagittal

plane with a forward-upward (STS) and downward-

backward (BTS) displacement. These movements al-

low estimating some parameters that were identiﬁed

in the literature as relevant for balance assessment

(Manckoundia et al., 2006). These parameters are the

following: a) movement duration, b) shoulder path

curvature, c) trunk angle, and d) ratio which matches

the vertical phase duration divided by the horizontal

phase duration.

3.2 Experimental Setup and Data

Acquisition

We use image processing technology with Kinect

to detect patients’ TUG movements. The proposed

system automatically produces virtual skeleton cor-

responding to the patient’s joint position, and this

skeleton information allows determining in real-time

spatio−temporal parameters which are relevant for

the balance assessment.

To extract parameters for balance assessment,

we captured Kinect skeleton recordings of the TUG

movement realized by the subject with time synchro-

nized (see Figure 1). Therefore, we used the data ac-

quired to compute the parameters set. The Kinect sen-

sor was placed to capture the image of the subject at

an approximate distance of 2−2,5 m to the chair, at a

height of 50−60 cm above the ﬂoor. The subject di-

rectly faced the Kinect sensor. The Kinect sensor and

its SDK produce a 3D virtual skeleton to establish the

positions of 20 skeleton joints on a human form. For

example, skeleton tracking determines where a user’s

head, hands, knees, and center of mass are. For each

of these skeleton joints, X, Y, and Z values are re-

ported. Kinect provides approximately 30 skeleton

frames per second.

Figure 1: Overview of the proposed system with Kinect sen-

sor.

3.3 Kinematic Parameter Extraction

Process

To begin the calculation of parameters, it is ﬁrst nec-

essary to determine whether the subject is sitting or

not. The posture and position of a subject’s body

joints deﬁne a pose; more speciﬁcally, it is the rela-

tionship of each joint to another. A pose is detectable

by either intersection or position of joints or the angle

between joints, using one or more plane. The distance

between two points for 2D and 3D points is respec-

tively given by:

−x

)

+ (y

−y

)

(1)

and:

−x

)

+ (y

−y

)

+ (z

−z

)

(2)

Regarding the joint angles, we can draw a triangle us-

ing any two joint points. The third point of the trian-

gle is derived from the other two points. The Law of

Cosines formula deﬁned as:

= a

+ b

−2abcosC (3)

where C is the angle opposite side c, gives us the value

of any desired angle. Transforming the formulas to

solve for the unknown angle C yields:

C = arccos(

+ b

−c

2ab

) (4)

Extracted features correspond to the kinematics

of shoulder displacement during STS and BTS and

VISAPP2014-InternationalConferenceonComputerVisionTheoryandApplications

664

the TUG duration. STS is composed of two main

phases: the ﬁrst phase is the forward ﬂexion and the

second, the extension phase, started on lift-off of the

buttocks from the seat and ended when maximum

hip, trunk and knee extension and maximum head

ﬂexion velocity were reached. BTS is characterized

by a movement in the opposite direction than STS:

downward and backward.

Shoulder Path Curvature. Shoulder paths during

forward and backward displacements were similar

and almost straight, therefore the curvature of path

for upward and downward displacements were only

calculeted (Manckoundia et al., 2006). Curvature is

deﬁned as:

cur =

max

(5)

where L corresponds to a straight line passing be-

tween the initial and the ﬁnal position of shoulder

displacement and D

max

means the maximal perpen-

dicular distance measured from the actual path to the

straight line.

Figure 2: Shoulder path during STS transfer. Curvature of

paths is given by the ratio Dmax/L.

Ratio. The ratio is computed using the following re-

lationship:

ratio =

vph

hph

(6)

where D

vph

and D

hph

they represent, respectively, the

vertical and the horizontal movement duration.

Trunk Angle. It refers to the angle θ between the

trunk and the vertical plane passing through the

center of mass of the body (see Figure 3).

Movement Duration. The total movement duration

of shoulder motion, during STS, corresponds to the

time interval between the moment when the shoul-

der depth component exceeded 8, 5% of its initial

position, and the moment when the head vertical

Figure 3: Trunk angle calculation.

component reaches or exceeds 94% of the size of

the person which was calculated. The thresholds are

experimentally determined. In BTS, it is deﬁned

as the time interval between the moment when the

shoulder vertical component dropped its peak value,

and the moment when the vertical components of the

hips reach their minimum values and the trunk angle

reaches its limit.

TUG Duration. TUG duration, measured in seconds,

means the total time taken to perform all TUG tasks.

It corresponded to the time interval between the mo-

ment when the forward phase starts and the moment

when the backward phase ends.

4 EXPERIMENT RESULTS

4.1 Experimental Protocol

Ten healthy young subjects participated in the present

study. Subjects were asked to complete three tri-

als for each of three TUG conditions: the TUG

alone (TUG−normal), the TUG−cognitive and the

TUG−motor. Subjects were given verbal instructions

to rise from a chair, walk 3 m, cross a mark on the

ﬂoor, turn around, walk back, and sit down again.

4.2 Result Analysis

Figure 4 shows STS parameter histograms for

TUG−normal. In this section, the conﬁdence inter-

val (CI) is deﬁned as:

= [x −1, 96

√

;x + 1, 96

√

] (7)

where x, s and n represent, respectively, the mean, the

standard deviation and the size of the sample. The

PreliminaryStudyontheDesignofaLow-costMovementAnalysisSystem-ReliabilityMeasurementofTimedUpandGo

Test

665

(a) Histogram of STS ratio parameter. (b) Histogram of STS curvature parameter.

Figure 4: Histogram of STS ratio, curvature, trunck angle and duration parameters of 10 young healthy subjects (P1-P10).

CI of the ratio, the curvature, the trunck angle and

the duration during STS for the TUG-normal are, re-

spectively, [1,038; 2,798], [0,051; 0,098], [31,162;

41,548], and [0,744; 0,987]. For the ten subjects, 70%

of the mean values of ratio, trunck angle and duration

are contained in their CI. This indicates that the vari-

ability of parameter measurement is reduced between

the different subjects for these three parameters and

we suggest that our system processes good measure-

ment reliability. Regarding the curvature, only 50%

of the mean values are in the CI. Indeed, the calcu-

lation is based on the outbreak of the vertical phase

which takes into account the size differences between

subjects. Therefore, the variability of curvature values

between subjects is not related to a miscalculation of

our system, but rather to inter-individual differences.

In ﬁgure 5, duration of STS was shorter than dura-

tion of BTS which was proved in the work of Mourey

(Manckoundia et al., 2006).

Concerning dual TUG task, in TUG-Normal,

mean values of the TUG duration are between 8,21

Figure 5: Comparison between STS duration and BTS du-

ration.

s and 11,89 s, in TUG-motor, they are between 8,18 s

and 12,65 s and in TUG-cognitive, they are between

8,21 s. and 14,84 s. These values refer to healthy

subjects as the limit score of each test is not exceeded

(see Section 3.1). These results are compatible with

VISAPP2014-InternationalConferenceonComputerVisionTheoryandApplications

666

previous studies. During STS, mean value of trunck

angle is between 18, 30

◦

and 49, 62

◦

, and in BTS, it is

between −3, 53

◦

and 54, 68

◦

The only parameter showing a main effect of the

factor ”condition” is the ratio in BTS transfer. For this

parameter, Kolmogorov-smirnov and Shapiro-Wilk

tests showed that the distribution follows a NORMAL

law (see Figure 6). Results from the one-way analy-

sis of variance (one-way ANOVA), whose factor is

condition, showed that there was a high variability in

the BTS ratio between normal condition and cognitive

condition (F(2, 27)= 4,2954, p= ,02401). A post hoc

within condition analysis was performed and showed

that there is a signiﬁcant difference between normal

condition and cognitive condition, this last result is

very stimulating regarding the literature about dual

task paradigm ﬁeld. In aging, the automatic motric-

ity seems less efﬁcient and some functional activities,

as the TUG, need a cognitive involvement (Teasdale

and Simoneau, 2001). Some authors proposed that

the BTS motion could be an interesting tool to assess

posturo-motor abilities (Manckoundia et al., 2006) in

aged adults. Here we showed that a simple BTS anal-

ysis can reveal an impairment involved by the dual

task condition even in a population of young adults.

Figure 6: Result of one-way ANOVA analysis.

5 CONCLUSIONS

In this work, we have presented a novel movement

analysis system for real-time balance assessment in

the frail elderly. It captured and recorded the TUG

test movement using a Kinect sensor and nine spatio-

temporal parameters were automatically extracted for

sit-to-stand and back-to-sit transfers by 3D real-time

video processing. Obtained experiment results with

ten healthy young subjects showed good measure-

ment reliability and reproducibility with important

precision. In addition, we showed that even in young

healthy subjects, some modiﬁcations of motor pat-

terns can be seen in dual task condition. Moreover,

our system allows detecting some very ﬁne changes

in posturo-motors abilities.

Our future works consist to perform TUG test for

real-time balance assessment in the frail elderly to

validate the proposed system in real world condition.

This study will open a new research and development

way for geriatric health which implies multiple as-

pects: user-friendly, hygiene, low-cost, home-based

environment, and automatic autonomy assessment.

REFERENCES

Bonnech

ere, B., Jansen, B., Salvia, P., Bouzahouene, H.,

Omelina, L., Cornelis, J., Rooze, M., and VAN

SINT JAN, S. (2012). What are the current limits of

the kinect sensor? Proc. 9th Intl Conf. Disability, Vir-

tual Reality & Associated Technologies (ICDVRAT),

pages 287–294.

Clark, R. A., Pua, Y.-H., Fortin, K., Ritchie, C., Webster,

K. E., Denehy, L., and Bryant, A. L. (2012). Valid-

ity of the microsoft kinect for assessment of postural

control. Gait & Posture, 36(3):372–377.

Gabel, M., Gilad-Bachrach, R., Renshaw, E., and Schuster,

A. (2012). Full body gait analysis with kinect. In 2012

Annual International Conference of the IEEE Engi-

neering in Medicine and Biology Society (EMBC),

pages 1964–1967.

Hagler, S., Austin, D., Hayes, T. L., Kaye, J., and Pavel, M.

(2010). Unobtrusive and ubiquitous in-home monitor-

ing: A methodology for continuous assessment of gait

velocity in elders. IEEE transactions on bio-medical

engineering, 57(4):813–820.

Mancini, M. and Horak, F. B. (2010). The relevance of clin-

ical balance assessment tools to differentiate balance

deﬁcits. European journal of physical and rehabilita-

tion medicine, 46(2):239–248.

Manckoundia, P., Mourey, F., Pﬁtzenmeyer, P., and Pa-

paxanthis, C. (2006). Comparison of motor strate-

gies in sit-to-stand and back-to-sit motions between

healthy and alzheimer’s disease elderly subjects. Neu-

roscience, 137(2):385–392.

Stone, E. and Skubic, M. (2011). Evaluation of an inexpen-

sive depth camera for passive in-home fall risk assess-

ment. In 2011 5th International Conference on Perva-

sive Computing Technologies for Healthcare (Perva-

siveHealth), pages 71 –77.

Teasdale, N. and Simoneau, M. (2001). Attentional de-

mands for postural control: the effects of aging and

sensory reintegration. Gait & posture, 14(3):203–210.

Weiss, A., Hirshberg, D., and Black, M. (2011). Home 3D

body scans from noisy image and range data. In 2011

IEEE International Conference on Computer Vision

(ICCV), pages 1951–1958.

PreliminaryStudyontheDesignofaLow-costMovementAnalysisSystem-ReliabilityMeasurementofTimedUpandGo

Test

667