Comparison of Recognition Accuracy of ADL with Sensor Wearing

Positions using 3-Axis Accelerometer

D. I. Shin

, S. K. Joo

and S. J. Huh

Department of Biomedical Engineering, Asan Medical Center, Seoul, Korea

Department of Biomedical Engineering, University of Ulsan College of Medicine, Seoul, Korea

Keywords: Accuracy, Activity, Daily Life, Sensor, Position, Accelerometer.

Abstract: The monitoring of single elderly is being more important due to rapid transition to aging society. There are

many bio-signals to monitor the emergent state of elderly. In this paper we propose new criteria to classify

daily life activities using accelerometer and pulse oximeter. We categorized activities with the motility of

real action. The upper most criteria are normal and abnormal activity. The lower criteria are ‘small or large

movement’, ‘periodic or random movement’, ‘no movement or shock’. Then we derive some parameters to

get thresholds to classify these activities according to our new criteria. The main parameters are entropy,

energy and autocorrelation. Some experiments were carried out to determine classifying thresholds. Finally

we got results of classified activities such as ‘no movements’, ‘small movements’, ‘large movements’,

‘periodic movements’ and ‘falls’. We got nearly 100% of classifying result for falls and no movements. In

this case of ‘quasi-emergency state’ our developing device investigates further status of elderly by

measuring of heart rate and oxygen saturation (SpO

) using pulse oximeter. Finally the device decides in

emergency, it sends a short message to server and then connects to the u-Healthcare centre or emergency

centre and one’s family.

1 INTRODUCTION

According to the data from Statistics Korea, the

aging index will increase rapidly from 9.5% (2006)

to 14.3% (2018) and 20.8% (2026). With this trend,

the number of single elderly increases too. Knowing

the emergency status of these single elderly is a

critical issue in the emergency monitoring system.

So we have been developing a monitoring device,

which can be easily worn on an elders’ body. The

wearing position is very important because it must

be very convenient for the elderly. And in the case

of an emergency, the reaction of elderly is also

important for the decision whether he or she is

serious. There were many researches for monitoring

devices(Boo-Ho Yang, Sokwoo Rhee, 2000, P.

Mendoza, P. Gonzalez, B. Villanueva, E.

Haltiwanger, H. Nazeran, 2004, Giuseppe Anastasi,

Marco Conti, Mario Di Francesco, Andrea Passarell,

2009, Francis E.H. Tay, D.G. Guo, L. Xu, M.N.

Nyan, K.L. Yap, 2009, Prajakta Kulkarni, Yusuf

Ozlurk, 2010, Amr Amin Hafez, Mohamed Amin

Dessouky, Hani Fikri Ragai, 2011). In these

researches, there are many considerations about

monitoring devices and systems with respect to u-

Healthcare Monitoring. After all, we conclude that

the ideal wearing position is wrist for now. With the

progress of technology, the device may be the shape

of hearing aid in the future.

In this research, we classified the activity type of

elderly in daily life. Recent researches classified the

activity type with the real action such as walking,

standing, sitting, lying etc.(Arunkumar Pennathur,

Rohini Magham, Luis Rene Contreras, Winifred

Dowling, 2003, A. Mannini, A.M. Sabatini, 2009,

G.M. Lyons, K.M. Culhane, D. Hilton, P.A. Grace,

D. Lyons, 2005, Marcia Finlayson, Trudy Mallinson,

Vanessa M. Barbosa, 2005, Angela L. Jefferson,

Robert H. Paul, Al Ozonoff, Ronald A. Cohen, 2006,

A. Godfrey, A.K. Bourke, G.M. Ólaighin, P. van de

Ven, J. Nelson, 2011). But in fact this kind of

classification is not helpful for the decision of

emergency status of an elderly. So we suggest new

concept of classification criteria. We categorized

activities with the motility of real action. The upper

most criteria will be normal and abnormal activity.

The lower criteria may be ‘small or large movement’,

‘periodic or random movement’, ‘no movement or

180

Shin D., Joo S. and Huh S..

Comparison of Recognition Accuracy of ADL with Sensor Wearing Positions using 3-Axis Accelerometer.

DOI: 10.5220/0005279901800184

In Proceedings of the International Conference on Biomedical Electronics and Devices (BIODEVICES-2015), pages 180-184

ISBN: 978-989-758-071-0

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

shock’.

Once we classify the elderly activity to abnormal

we can further investigate the accurate status with

the reaction button or pulse oximeter, which will be

adopted by our monitoring device. The clinical

importance of oxygen saturation of blood (SpO

) is

mentioned on many articles (Barker SJ, Morgan S.,

2004, Anna Letterstål, Fredrik Larsson, 2007,

Gülendam Hakverdioğlu Yönt, Esra Akin Korhan,

Leyla Khorshid, 2010, Elif Derya Ubeyli, Dean

Cvetkovic, Irena Cosic, 2010)

If we can classify a person’s status by normal or

abnormal, we can make more concrete speculation

in case of abnormal status. As a result, we may

reduce processing resource, power and finally

physical size of the sensor. The more compact size

and reduced processing power will be more

convenient in wearing it.

2 MATERIALS AND METHODS

2.1 System Overview

We extracted acceleration data and oxygen

saturation data from our monitoring device in

developing. Data was moved from the memory of

monitoring device to PC via USB port. Sampling

rate is 10ms/sample and converted by 12bits depth.

Fig. 1 shows the illustration of our monitoring

device. Figure 2 illustrates our processing system.

Personal computer (Pentium V) is used to process

and analyse activities. The LabView

software

from National Instruments is used to acquire and

display the acceleration data from monitoring device.

The Matlab

software is used to process and

analyse the acceleration data.

Figure 1: The illustration of our monitoring device.

Figure 2: The overall processing system.

2.2 Activity Classification

In this research, we classified the activity type of

elderly in daily life. Recent researches classified the

activity type with the real action such as walking,

standing, sitting, lying etc. But in fact this kind of

classification is not helpful for the decision of

emergency status of an elderly. So we suggest new

concept of classification criteria. We categorized

activities with the motility of real action. The upper

most criteria will be normal and abnormal activity.

The lower criteria may be ‘small or large movement’,

‘periodic or random movement’, ‘no movement or

shock’. Figure 3 shows the classification criteria of

our new concept.

Figure 3: Classification criteria of activities.

The 3-axis acceleration data were pre-processed like

below.

= sqrt(a

+ a

) (1)

= LPF(A

) (2)

is root-mean-square of original 3-axis

acceleration data and A

is low pass filtered data

with 5Hz cutoff frequency.

To classify activities, we calculate some

parameters and define threshold of classification.

First, the entropy is measured like below,

Entropy = ∇a/∇t (3)

Entropy is the ratio of acceleration change per

unit time. And the energy is defined like this,

Energy = ∑a (4)

ComparisonofRecognitionAccuracyofADLwithSensorWearingPositionsusing3-AxisAccelerometer

181

It can be interpreted as speed. The autocorrelation is

calculated for the grade of periodicity.

Autocorrelation = Periodicity (a) (5)

2.2.1 Normal Activity Classification

Normal activity is classified with two categories.

The first is the magnitude of movements. This is

judged by the threshold of entropy and energy. The

judge function is described like this.

mov

= a*Entropy + b*Energy (6)

The second category is periodicity and judge

function for this is,

per

= c*Entropy + d*Autocorrelation (7)

2.2.2 Abnormal Activity Classification

Abnormal state is categorized into two classes. One

is ‘no movements’, the other is ‘falls’. When in ‘no

movements’ there might be two situation which are

“in sleep” and “in emergency”. In these situations

our monitoring device will check heart rate and O2

saturation in blood using pulse oximeter.

To determine whether falls or not, we use the

entropy for the threshold function.

fall

= e*Entropy (8)

To determine whether no movements or not, we

use the entropy and the energy for the threshold

function.

nmov

= f*Entropy + g*Energy (9)

2.3 Classifying Algorithm

Figure 4 shows the flowchart of activity classifying

algorithm. Once we start the algorithm, the

accelerometer in the device is powered on. The

acceleration data is acquired with the speed of 100

samples/sec. And then the acceleration data is low-

pass filtered with the 5Hz cut-off frequency. We call

these procedures ‘Pre-processing’ and equation (1)

and (2) show these procedures. Next, we calculate

parameters on equations (3) to (5). Using the entropy

and energy we can calculate the parameter J

nmov

. If

nmov

is less than the threshold T

nmov

, we can judge

there are no movements such as resting or sleeping

state. If J

nmov

is great than the threshold T

nmov

, we

can judge that there are some movements that

include some kind of falls.

The next stage we investigate the parameter J

fall

according to the equation (8). If this is greater than

fall

, a kind of fall must have happened. Once the

state is classified to normal movement, we can

classify to lower categories as shown in figure 3.

In real world, situations are more complex and

ambiguous. So, the classification algorithm is

difficult. But as we refine the algorithm more

accurately, the result will be more realistic.

Figure 4: Flowchart of activity classifying algorithm.

3 RESULT

Figure 5 shows the low-pass filtered acceleration

data. It includes various activities. Small movements

Figure 5: Acceleration data from various activities.

BIODEVICES2015-InternationalConferenceonBiomedicalElectronicsandDevices

182

Figure 6: Classified results for successive various activities.

can be showed in Figure 5(a). These movements

include gripping a pen, writing, moving a paper,

scratching one’s skin, removing glasses etc. In small

movements, it shows small accelerations under 2g

(1g=9.8m/s

). On the other hand, Figure 5(b) shows

large movements such as stretching, doing

gymnastics, standing up suddenly etc. It shows large

acceleration of 5g or more. Sometimes it exceeds

10g but its slope is rather than gradual. Figure 5(c)

shows a typical periodic movement which is walking.

There are two levels of valley, the upper valley

represents backward peak position of hand and the

lower valley represents forward peak position of

hand.

Figure 6 shows classified results for successive

various activities according to our algorithm. The

color bar denotes the class of activity. Data from

monitoring device is transmitted to personal

computer and are processed with Labview

and

Matlab

software to verify our algorithm.

4 CONCLUSIONS

Knowing the emergency status of these single

elderly is a critical issue in the emergency

monitoring system. So we have been developing a

monitoring device, which can be easily worn on an

elders’ body. The wearing position is very important

because it must be very convenient for the elderly.

And in the case of emergency, the reaction of the

elderly is also important for the decision whether he

or she is serious. After all, we conclude that the ideal

wearing position is wrist for now. With the progress

of technology, the device may be the shape of

hearing aid in the future.

In this research, we classified the activity type of

an elderly in daily life. Recent researches classified

the activity type with the real action such as walking,

standing, sitting, lying etc. But actually this kind of

classification is not helpful for the decision of

emergency status of an elderly. So we suggest new

concept of classification criteria. We categorized

activities with the motility of real action. The upper

most criteria will be normal and abnormal activity.

The lower criteria may be ‘small or large movement’,

‘periodic or random movement’, ‘no movement or

shock’. Once we classify the elders’ activity to

abnormal we further investigate the accurate status

with the reaction button or pulse oximeter, which is

already adopted, in our monitoring device. If we can

classify a person’s status to normal or abnormal, we

can make more concrete speculation in case of

abnormal status. As a result, we may reduce

processing resource, power and finally physical size

of the sensor. The more compact size and reduced

processing power will be more convenient in

wearing it.

REFERENCES

Boo-Ho Yang, Sokwoo Rhee, Development of the ring

sensor for healthcare automation, Robotics and

Autonomous Systems 30 (2000) 273–281

P. Mendoza, P. Gonzalez, B. Villanueva, E. Haltiwanger,

H. Nazeran, A Web-based Vital Sign Telemonitor and

Recorder for Telemedicine Applications, in

Proceedings of the 26th Annual International

Conference of the IEEE EMBS San Francisco, CA, USA,

September 1-5, 2004, pp. 2196-2199.

No Movements

Small Movements Large Movements

Small Movements

Periodic Movements

Falls

No Movements

ComparisonofRecognitionAccuracyofADLwithSensorWearingPositionsusing3-AxisAccelerometer

183

Giuseppe Anastasi, Marco Conti, Mario Di Francesco,

Andrea Passarell, Energy conservation in wireless

sensor networks: A survey, Ad Hoc Networks 7 (2009)

537-568

Francis E.H. Tay, D.G. Guo, L. Xu, M.N. Nyan, K.L. Yap,

MEMSWear-biomonitoring system for remote vital

signs monitoring, Journal of the Franklin Institute 346

(2009) 531–542

Prajakta Kulkarni, Yusuf Ozlurk, mPHASiS: Mobile

patient healthcare and sensor information system,

Journal of Network and Computer Applications, 2010

Amr Amin Hafez, Mohamed Amin Dessouky, Hani Fikri

Ragai, Design of a low-power ZigBee receiver front-

end for wireless sensors, Micro Electronics Journal 40

(2011) 1561-1568

Arunkumar Pennathur, Rohini Magham, Luis Rene

Contreras, Winifred Dowling, Daily living activities in

older adults:Part II—effect of age on physical activity

patterns in older Mexican American adults,

International Journal of Industrial Ergonomics 32

(2003) 405–418

G.M. Lyons, K.M. Culhane, D. Hilton, P.A. Grace, D.

Lyons, A description of an accelerometer-based

mobility monitoring technique, Medical Engineering &

Physics 27 (2005) 497–504

Marcia Finlayson, Trudy Mallinson, Vanessa M. Barbosa,

Activities of daily living (ADL) and instrumental

activities of daily living (IADL) items were stable over

time in a longitudinal study on aging, Journal of

Clinical Epidemiology 58 (2005) 338–349

A. Mannini, A.M. Sabatini, Computational methods for

the automatic classification of postures and movements

from acceleration data, Gait & Posture 30S (2009) S26–

S74

A. Godfrey, A.K. Bourke, G.M. Ólaighin, P. van de Ven,

J. Nelson, Activity classification using a single chest

mounted tri-axial accelerometer, Medical Engineering

& Physics, in Press (2011)

Angela L. Jefferson, Robert H. Paul, Al Ozonoff, Ronald

A. Cohen, Evaluating elements of executive functioning

as predictors of instrumental activities of daily living

(IADLs), Archives of Clinical Neuropsychology 21

(2006) 311–320

Thessa I.M. Hilgenkamp, Ruud van Wijck, Heleen M.

Evenhuis, (Instrumental) activities of daily living in

older adults with intellectual disabilities, Research in

Developmental Disabilities 32 (2011) 1977–1987

Barker SJ, Morgan S., A Laboratory Comparison of the

Newest “Motion-Resistant” Pulse Oximeters During

Motion and Hypoxemia, Anesthesia and Analgesia

2004;98(55),S2:A6

Anna Letterstål, Fredrik Larsson, Assessment of vital

signs on admission to short time emergency wards

improves patient safety and cost-effectiveness,

Australasian Emergency Nursing Journal, Volume 10,

Issue 4, November 2007, Page 191

Gülendam Hakverdioğlu Yönt, Esra Akin Korhan, Leyla

Khorshid, Comparison of oxygen saturation values and

measurement times by pulse oximetry in various parts

of the body, Applied Nursing Research, 2010

Elif Derya Ubeyli, Dean Cvetkovic, Irena Cosic, Analysis

of human PPG, ECG and EEG signals by eigenvector

methods, Digital Signal Processing 20 (2010) 956–963

BIODEVICES2015-InternationalConferenceonBiomedicalElectronicsandDevices

184