A

mmdimsrv©

1

dE1miprto

s

(e

1h

Disponible en ligne sur

www.sciencedirect.com

IRBM 34 (2013) 92–100

Original article

Homecare monitoring system: A technical proposal for the safety of theelderly experimented in an Alzheimer’s care unit

Le système de surveillance Homecare : une proposition technique pour la sécurité des personnesâgées expérimentées en unité de soins d’Alzheimer

W. Bourennane a,b,∗, Y. Charlon a,b, F. Bettahar a,b, E. Campo a,b, D. Esteve b

a Université de Toulouse, UPS, UTM, LAAS, 31100 Toulouse, Franceb CNRS, LAAS, 7, avenue du Colonel-Roche, 31400 Toulouse, France

Received 11 January 2013; received in revised form 17 January 2013; accepted 13 February 2013Available online 13 March 2013

bstract

Today, the monitoring of dependent people is a real challenge. This is mainly due to the increase in the number of elderly and reduction ofedical staff. However, works done in recent years in France are faced with obstacle of clinical and industrial validation. In this paper, we propose aonitoring system for dependent persons living alone at home or in an institution 24 h/24. This system is based on a presence multisensor network

eployed in the living environment of the monitored person coupled with a wireless identification system. A learning algorithm is implementedn order to define a personalized behavioural model (motions deviation, nocturnal activity, falls, mobility). This model allows the nursing staff

onitoring the behaviour through a web application accessed remotely, and also intervention in case of dangerous situations thanks to an alertystem. In further, we describe the experiments conducted at the Caussade Local Hospital in France with Alzheimer patients and we present theesults of data obtained. Finally, we propose the way for an economic model that would allow to industrialize the system with prospects of clinicalalidation on a larger cohort of patients.

tpi“seatiHd

2013 Elsevier Masson SAS. All rights reserved.

. Introduction

The improvement of care quality and general living con-itions leads to a regular increase in the lifespan average. Inuropean countries, the proportion of people over 65 is up to6% [1]. This is a population that most often involves regularedical monitoring to manage chronic diseases in order to antic-

pate disease signs in frail persons [2,3] or to monitor elderlyeople with aging diseases. The elderly are not the only personsequiring a regular long-term monitoring, persons with disabili-ies, convalescents and children need also a technical monitoring

f the same kind.

The implementation of efficient health policy for risk per-ons is obviously very expensive due to the frequency and

∗ Corresponding author.E-mail addresses: wbourenn@laas.fr (W. Bourennane), ycharlon@laas.fr

Y. Charlon), fbettaha@laas.fr (F. Bettahar), campo@laas.fr (E. Campo),steve@laas.fr (D. Esteve).

tFtoRmwed

959-0318/$ – see front matter © 2013 Elsevier Masson SAS. All rights reserved.ttp://dx.doi.org/10.1016/j.irbm.2013.02.002

he relatively “heavy” therapeutic interventions. Two com-lementary ideas are being explored to advance masteredmplementation: prevention that relies on the concept of peoplefrail” [4,5]. For those, we have proposed an organized regularurveillance to anticipate the onset of problems and treat at thearliest. The second idea is the home care, benefiting from recentdvances in information and telecommunication technologieso alleviate period and hospitalization visits. It is a new clin-cal and technical approach which is complex to implement:omecare is intended to propose a possible technical solution,emonstrate its feasibility and illustrate its working on a pro-ected site. These works are realized under the framework of therench Research National Agency and follow the works done by

he LAAS-CNRS laboratory in monitoring of elderly, dependentr disabled for these last 20 years. They are based on a Frenchesearch National Center prospective study [6] on two main

otivations: the emergence of a dynamic “home automation”hich promised the industrial development of a highly comput-

rized and automated home. And the anticipation that we cano for population aging which involving to innovate in elderly

W. Bourennane et al. / IRBM 34 (2013) 92–100 93

Table 1The system use cases.

Events Area Emergencylevel

Destinations Data Sensor Decisiontime

Platform

Room Hospital/Garden

Physician Nurse Accelerometer ZigBee IR

Fall x x Alert x x x x Real time Real time application/PhoneNocturnal restlessness

(in bed)x Alert x x Real time Real time application/Phone

Nocturnal immobility(in bathroom)

x Alert x x Real time Real time application/Phone

Group activity x Information x x Deferred Web applicationNocturnal restlessness

(longitudinalfollow-up)

x Information x x Deferred Web application

Nocturnal immobility(longitudinalfollow-up)

x Information x x Deferred Web application

Decreased moving level x x Warning x x x Deferred Web applicationIB

mtdIscwsv

msairdibmtHnta

2

dcwstlcf

2

ipkseridfrstbbhF

ial

tetsint

2

ncreased moving level x x Warning x

ehavioral deviation x Warning x

onitoring. In 20 years, the conditions have changed a lot inerms of techniques for a “Smart Home” with the extraordinaryevelopment of telecommunications and computer networks.ndeed the development is multiplying for comfort, energy andecuring material and people. However, the issue of the homeare for elderly, disabled or dependent is still a work in progresshich is characterized by the fact that those actual technical

olutions cannot find a way to the market for two main reasons:alidation and cost.

Homecare leads to an already powerful system for elderlyonitoring. It is discreet and it can identify the monitored per-

on, locates him and warns in danger cases. These performancesre the result of a multisensory approach. Sensors are distributedn the environment. The collected data from these sensors areegularly sent to a computer and processed in real time. Theseata allow building a normality model of habits by using learn-ng techniques. This model is served as reference for comparisonetween the model predictions and real time data in order toake decision automatically for launching an alert or promp-

ing an intervention. Table 1 summarises the use cases of theomecare monitoring system. We present in this paper the tech-ology used and the performances obtained. Finally, we describehe steps which remain to be taken for a final validation before

wider industrial distribution.

. Technological choices

Homecare is a multisensory monitoring system. Sensors areistributed in the environment in “strategic” places. Data areollected, as far as possible, via a wireless communication net-ork and processed in real time in order to diagnose the current

ituation and propose actions. This leads, from a system archi-

ecture point of view, to a set. In Fig. 1, the connection fromocal system to Internet allows organizing exchange between alloncerned stakeholders: physician, hospital service, caregivers,amily. Fig. 1 shows the platform architecture of the system.

ar

x x Deferred Web applicationx Deferred Web application

.1. Sensors

Homecare includes a presence sensors network distributedn living areas of the observed person (room, apartment,roximity. . .). These presence sensors can be in differentinds: light barriers, micro switches, infrared sensors, magneticwitches. The choice of sensors will depend on the consideredxperimentation site and monitoring priorities. The system mustespect person privacy (unobtrusive). For this reason, we use thenfrared detectors. These low-cost sensors are widely used foretecting intrusions. In our case, we adapt this type of productor tracing trajectories of persons while they move in their envi-onment. This is already an old choice that is lead with manytudies for wandering and agitation situations, especially duringhe night observation [7,8]. In order to monitor the nocturnalehaviour of patient, we have installed a pressure detector in theed under the patient’s mattress. This sensor allows us to detectis getting up or sleeping and calculates presence time in bed.ig. 2 shows pictures of these sensors.

The person monitored by Homecare can move in the areasnstrumented by the sensors without constraints. He wears only

miniaturized electronic tag, which allows identification andocalisation to be realized (Fig. 3).

For this device, there is an important requirement of minia-urization because it must be invisible for the user. It can bembedded in clothing or in wristwatch [9] or worn directly onhe body [10]. This tag is a central component of the monitoringystem because it allows the identification of the patient and alsonsures the location in the areas where the presence sensors can-ot be placed. It is equipped with a processor, a radiofrequencyransmitter and an accelerometer for detecting falls [11].

.2. Data collecting

Many solutions are commonly used to collect data throughnalogue and digital channels. Homecare demonstrators are cur-ently equipped with:

94 W. Bourennane et al. / IRBM 34 (2013) 92–100

Fig. 1. Platform architecture.

Fig. 2. Presence sensors used in Homecare system.

system

•

•

•

prts

tsa

t

cotd

Fig. 3. Identification

radio frequency transmitter (868 MHz): for transmitting pres-ence data detection (infrared and pressure detector) to thelocal computer;

digital wireless transmission for identification/location: usedbetween tags and hotspots according to the ZigBee low-powercommunication protocol;

Power Line Interface (PLI): for transmitting identificationdata from the hotspot to the local computer.

When the presence sensors detect movements from the

atients, the real time application running on the local computereceives activation via the radiofrequency transmitter connectedo the computer. At the same time, the tag worn by the patientends his identification frame, each 3 s through the PLI. IR and

Sei

(tag with hot spot).

ag data are then merged and saved into the local database. Theaved format frame is: patient name, number/detection zone/datend time of motion.

In case of near outdoor and collective areas, we use only theag-hotspot to locate and identify the patient.

A database of all sensor data is organized on the localomputer. These data are processed by using learning meth-ds (patient’s habits learning). The habits concerned inhis case are the lifestyle and behaviour dependent onisplacements.

We can observe that people are conditioned by their habits.

chedules and actions are relatively repetitive in their daily;specially elderly who keep daily routines. The monitor-ng concept proposed by Homecare depends on habits for

/ IRB

aiisii

thos

2

ti

•

•

•

2

tcbc““t

r oom r

pcf

uap0twTIts

••

c

x

w(

vbd

R

2

sdc

P

wi

W. Bourennane et al.

utomatically building usual behaviour models by using learn-ng in accordance to time slots. This model is self-adaptive andt anticipates patient displacements. It can take into account thelow changes of behaviour. Real data is compared with the modeln order to measure the discrepancy and so to trigger (or not) thentervention protocol.

The step of experimental normality model construction andhe comparison of predicted with real time data are generic. Weave proposed (in LAAS-CNRS) this concept for drowsinessf car driver [12,13] or for detecting insect infestations in grainilos [14].

.3. Data processing

The availability of a monitoring system can lead to con-inuous measurements and therefore answer to these threessues:

the monitoring in the intimacy periods (when he is alone) suchas nocturnal monitoring: we can access to useful characteri-sations of insomnia, nocturnal restlessness;

the longitudinal monitoring and anticipated detection ofbehaviour changes: the general interest is to follow, forexample, the evolution of convalescence or conversely wors-ening of a chronic illness. More, this organized homemonitoring can be a way for monitoring the “fragility” ascharacterized by the Fried criteria [2,3] and detecting theonset of an aging disease in order to treat it in its earlieststage;

continuous monitoring of specific risks or affections by inte-gration of other sensor types in Homecare system: in CaussadeHospital experiments, the patients are equipped with anaccelerometer for fall detections [11,15,16].

.3.1. Behavioural modelOur concept of behavioural model (Fig. 4) is based on

he time spent learning in each position. Every position isharacterized by its area number Z and the time spent Tsny the patient in this area. In order to get best results, weut up the location data according to time slots (Morning08 a.m.–12 a.m.”, Afternoon “12 a.m.–06 p.m.”, Evening6 p.m.–11 p.m.”, and Night “11 p.m.–08 p.m.”). So, for everyime slot, we have data in the following format:

Area Room

Corrido

Living r

Corrido

Area Time spent

Z1 Ts1

Z2 Ts2

Z3 Ts3

. . Data collected during time slot ex:

Living room

Terrace

. .

Zn Tsn

In second step, we cut up the serial location data to dis-lacements. Every displacement is characterized by beginningoordinates (Z,Ts) and end coordinates (Z,Ts) as shown in theollowing example: p

M 34 (2013) 92–100 95

Tim e spent

3550s

4s

1650s

10s

1436s

3000s

Displacement 4

Displacement 2

Displacement 3

Displacement 1

Area SpentTime

Room 3550s

Corridor 4s

Living room 1650 s

Corridor 10s

Living room 1436s

Terrace 3000s Displacement 5

The distance between each two displacements is calculated bysing LCSS (Longest Common Subsequence) formula [17,18]ccording to Area coordinates. If the LCSS distance = 0, the dis-lacements are classed in a same cluster. If it is different from, a new cluster is created. In result, we obtain a cluster for eachypical displacement. The last step is to process each clusterith k-means clustering method [18] according to time spent Ts.he obtained clusters are characterized by a centroid (Tsb,Tsf).

n aim to define detecting thresholds of outlier displacements,he k-nearest neighbour algorithm (k-NN) [19] is applied as aupervised learning in order to:

define the closest cluster for each displacement;calculate Euclidian distance average x̄ and standard deviation� between centroid of each cluster and his members.

A displacement is considered as normal if the followingondition is achieved:

¯ − σ ≤ d ≤ x̄ + σ (1)

here d is Euclidian distance between the real time displacementTs1,Ts2) and centroid (Tsb,Tsf) of the closest cluster.

In online mode, the abnormal displacements are counted byerifying (1) for each real time displacements. R is the ratioetween the number of abnormal displacement and the totalisplacements during a time slot, as defined in (2).

= Number of abnormal displacements

Tatal displacements∗ 100 (2)

.3.2. Nocturnal restlessnessThe infrared detector placed above patient bed allows mea-

uring the nocturnal restlessness by counting the number ofetections on last ten minutes. The agitation profile P is cal-ulated according to (3):

= N

det ∗ 60 ∗ 10∗ 100 (3)

here N is the number of detections in the last ten minutes, dets the number detection capacity of sensor per second.

Thresholds of restlessness S is defined with the agitationrofile x̄p and standard deviation σ̄p on the last 30 days.

96 W. Bourennane et al. / IRBM 34 (2013) 92–100

Fig. 4. Processin

2

ovR

3

Car

•

•

•

•

tmmo

wcp

otm

4

ipiatSwpta(

5

ch1f

Ttn

caig

fw

system is technically able to insure supervisory functions: Iden-tifying, locating, detecting behavioural deviance and falling.Other features are obviously possible by extending the sensor

Fig. 5. Online processing diagram.

.3.3. Online modeThe behavioural model and restlessness threshold are updated

nce per day through relearning, by including all alarms dataalidated as true. This technique adjusts the threshold values (S,) and avoids the system to trigger false alarms (Fig. 5).

. Deployment of the system in hospital

Homecare was installed in an Alzheimer’s care unit at theaussade hospital in France according to described systemrchitecture in Fig. 6. We have deployed it into one room, cor-idor, common living room and garden:

room of the patient is equipped with infrared presence sensors(8 detectors). These detectors allow us to follow precisely thepatient displacements in his room. The bed is specificallyinstrumented by a pressure detector;

hotspots are deployed in the service in order to identify thepatient;

data are transmitted by a wireless network (infrared data),and a wireless and a power line communication (identificationdata);

patient is equipped with a wireless transmitter tag. This devicesends the identification frame to hotspots continually every3 s.

One resident (woman, 84 years old) was recruited to par-icipate in experimental trials. She was autonomous in her

ovements but needs help in her daily grooming routine. Theedical staff administered informed consent to the individual

r her family before participating in the trials. This hospital

ffi

g diagram.

elcomes patients with Alzheimer disease. With assistance ofaregivers, the tag is packed in order to place it on the back ofatient. This solution is accepted if tag is miniaturized.

We have placed infrared detectors in collective areas with thebjective to monitor the group activity. This parameter allowshe medical staff detecting periods of high risk events and assign

ore caregivers.

. Web interface

A web interface to users has been designed (Fig. 7). Itncludes an access to the central computer. This computer islaced in an accessible space for the nursing staff. This computers connected to Internet network allowing physician or otheruthorized person to consult patient data. The real time applica-ion shows the patient position and can trigger alarms by sendingMS messages to the phones of caregivers. Fig. 7 shows theeb interface. It plans to deliver some features considered byhysicians as essential in his periodical observations. These fea-ures are: distance, motion speed, nocturnal agitation, groupctivity. The physician can visualize these features as curvesFig. 8).

. Results and discussions

Fig. 9 presents the behavioural deviance R of the patient. It isalculated on the last 30 days. These curves show that the patientas had an abnormal behaviour compared to the usual from 27-1 to 29-12, from 08-12 to 11-12 for the morning time slot, androm 30-11 to 03-12 for the night time slot.

Fig. 10 shows an example of nocturnal restlessness profile.his curve allows detecting the period when the patient is agi-

ated. This information is useful for caregivers to monitor theocturnal activity of patient.

By using the pressure detector placed under the mattress, thearegivers can visualise the presence of the patient in his bednd the number of the getting up as shown in Fig. 11. Thisnformation can also be correlated to the restlessness profile toet full activity profile of the patient.

Figs. 12 and 13 present falls detected during 63 days. The trueall alerts are validated by the caregivers. The true falls occurredhen the activity of patient increased.In summary, through these experimentations, Homecare

unctions and the development of specific software for theseunctionalities. However, it is necessary to validate this monitor-ng system. In a complementary step, we need two validations:

W. Bourennane et al. / IRBM 34 (2013) 92–100 97

Fig. 6. Deployment architecture in the care unit.

Fig. 7. Screenshots of the web interface.

Fig. 8. Processed data and displayed on the web interface.

98 W. Bourennane et al. / IRBM 34 (2013) 92–100

0,0%

20,0%

40,0%

60,0%

80,0%

100,0%

21-n

ov.

22-n

ov.

23-n

ov.

24-n

ov.

25-n

ov.

26-n

ov.

27-n

ov.

28-n

ov.

29-n

ov.

30-n

ov.

1-dé

c.2-

déc.

3-dé

c.4-

déc.

5-dé

c.6-

déc.

7-dé

c.8-

déc.

9-dé

c.10

-déc

.11

- déc

.12

-déc

.13

-déc

.14

-déc

.15

-déc

.16

-déc

.17

- déc

.18

-déc

.19

-déc

.20

-déc

.

Morning (08a m-12am) Night (11 pm-08am)

Fig. 9. Behavioural deviance (R).

0,00%

0,50%

1,00%

1,50%

2,00%

2,50%

21:1

0 21

:25

21:4

0 21

:55

22:1

0 22

:25

22:4

0 22

:55

23:1

0 23

:25

23:4

0 23

:55

0:10

0:

25

0:40

0:

55

1:10

1:

25

1:40

1:

55

2:10

2:

25

2:40

2:

55

3:10

3:

25

3:40

3:

55

4:10

4:

25

4:40

4:

55

5:10

5:

25

5:40

5:

55

6:10

6:

25

6:40

6:

55

7:10

7:

25

7:40

7:

55

Moving av erage of P

Fig. 10. Example of restlessness profile (one night).

Fig. 11. Presence of the patient in his bed.

Fig. 12. Fall alerts on 63 days (09/10/2012–09/12/2012).

00,005

0,010,015

0,020,025

0,030,035

Fall alerts

Fig. 13. True fall alerts distribution on 24 h (average of 63 days).

/ IRB

•

•

6

scaoic(thiHd

•

•

sussip

•

•

7

api

8

oipbo(

mtioabTdLmca

wtHctt

noitH

R

W. Bourennane et al.

validation of users and their families: in order to insure theacceptability of this system in large scale;

clinical validation: with evaluation of medical benefits that iscompared with current practices. Given the changes of situa-tions and uses, it is probably a few hundred persons that shouldbe considered on longer periods (over 1 year).

. Economic model

To perform these functions, technologies are available:ensors and hotspots are miniaturized. With wireless communi-ation, these devices can be installed in few hours. The personsre equipped with a miniaturized wearable tag on the clothesr body. A local area network allows collecting data on mon-toring computer that is connected to Internet. The electronicomponents can be chosen with a reasonable cost investmentex.: for domestic use < 1000 Euros). So technically, the objec-ives are reachable. A simple example is validated (Caussadeospital). We could also estimate the cost that can be reasonablef we reach a high diffusion level (a few tens of thousands).owever, the deployment of this technical solution must beeepened according to ambition that we fix:

it can be a simple alert system, which is installed in the liv-ing areas of the dependent person. It secures the environmentby triggering alerts to his family or to monitoring computer24/24 in case of dangerous situations (fall, wandering, andweakness). In this case, Homecare system can be as an exten-sion of property security systems;

it can be the result of a medical prescription that allowsthe home monitoring of some chronic diseases (RESPECTproject is actually under study aiming to monitor frail personsaccording to Fried criteria).

At this stage of study, it is too early to arbitrate these coun-elling. We must first insure that the option is attractive forsers and adapting it with application requirements. This stephould be realized with significant tests in real situations. Withupport of industrial partners, the Homecare project concludedmmediate opportunities to explore two interesting markets asriority:

specialized institutions market: the argument in this case isthe quality and the safety of the support offered to patientsand families. It is also information that can provide the mon-itoring system to caregivers on the behavioural developmentof persons. For the staff, it is helpful and reassuring supportthat allows him to quickly assess habits of the newcomer orto judge the urgency of a situation and to insure coordinatedmonitoring with all the actors;

market of dependent people at home: the argument here is

the security of the frail person and his family. In this case,Homecare system aims to prolong life at home. It can alertfamily during their absence if a dangerous situation occurs oralert the monitoring computer that regroups all subscribers.

M 34 (2013) 92–100 99

The market in France is about 1.2 millions of convalescents,disabled or dependents [20].

. Uses validation

In the current status, we must be able to validate the efficiencynd the useful of the project on real cases. So, it is a rigorousrocess that should lead users to invest in the system (equipment,nstallations, training, monitoring. . .).

. Conclusion

The remote measurement of symptoms is a potentially usefulption for integration in health system. It would allow earlydentification of behaviour changes of frail persons and lead toreventive treatments. This is an ambitious option that shoulde explored deeply by organizing monitoring on large cohortf patients. We are planning this solution in a future projectRESPECT).

Homecare explores the most accessible idea of security homeonitoring. Indeed, in the case of elderly living alone or in pairs,

he family is afraid of accidents, discomfort or weakness. Thissolation risk can be secured through a continuous monitoringf the activities of the person. It is the objective of Homecareccording to specifications: detecting deviation in the patientehavioural observations that can be the marker of problems.he technical proposition of Homecare is a multisensor systemistributed in the living environment of the patient followed.ocation data is collected regularly and used to build behaviouralodel considered as normal. Real time data and predicted are

ompared in order to diagnose critical situations and to triggerlerts.

Homecare was tested in an Alzheimer care unit at the hospitalith the support of caregivers. The obtained results show that

his system is operational and it can be technically deployed.owever, it needs two complementary validations on a wider

ohort of patients. The first is by the patients themselves andheir families and the second is more clinical in order to establishhe medical interest and the connection with the care system.

Finally the question addressed by the project is the eco-omic model. We have two options under study: the equipmentf institutions systematically and the integration of the mon-toring system for maintaining the patients at home. Thesewo approaches are compatibles with the technical advances ofomecare.

eferences

[1] Institut national de la statistique et des études économiques [Online]. Avail-able: http://www.insee.fr/fr/ [Accessed: 05-sept-2012].

[2] Fried LP, Tangen CM, Walston J, Newman AB, Hirsch C, Gottdiener J,et al. Frailty in older adults evidence for a phenotype. J Gerontol A BiolSci Med Sci 2001;56(3):M146–57.

[3] Fried LP, Ferrucci L, Darer J, Williamson JD, Anderson G. Untan-gling the concepts of disability, frailty, and comorbidity: implications forimproved targeting and care. J Gerontol A Biol Sci Med Sci 2004;59(3):M255–63.

1 / IRB

[

[

[

[

[

[

[

[

[

00 W. Bourennane et al.

[4] Boyd CM, Xue QL, Simpson CF, Guralnik JM, Fried LP. Frailty, hospital-ization, and progression of disability in a cohort of disabled older women.Am J Med 2005;118(11):1225–31.

[5] Ahmed N, Mandel R, Fain MJ. Frailty: an emerging geriatric syndrome.Am J Med 2007;120(9):748–53.

[6] Estève D. Conclusions de la commission d’experts Habitat Intelligent,CNRS-SPI, Développement des Soins à Domicile, mars 1997.

[7] Campo E, Chan M, Bourennane W, Estève D. Remote monitoring platformfor prevention and detection of elderly deviant behaviour. 3rd InternationalConference: E-Medical Systems; 2010.

[8] Chan M, Campo E, Esteve D. PROSAFE, a multisensory remote monitoringsystem for the elderly or the handicapped. Proceedings of the ICOST’ 2003,p. 89–95.

[9] Bourennane W, Bettahar F, Campo E. Self-adaptive trajectory learningfor monitoring the elderly wandering. The 10th IASTED InternationalConference on Biomedical Engineering-BioMed 2013 [Article in press].

10] Bourennane W, Charlon Y, Chan M, Estève D, Campo E. Inte-gration of wearable device with actimetry system for monitoringAlzheimer’s patients. International Conference on Wearable Micro andNano Technologies for Personalised Health (pHealth 2011), Lyon (France),2011.

11] Charlon Y, Bourennane W, Campo E. Mise en œuvre d’une plateforme desuivi de l’actimétrie associée à un système d’identification.

12] Santana-Diaz A, Jammes B, Esteve D, Gonzalez-Mendoza M. Driverhypovigilance diagnosis using wavelets and statistical analysis. Intelligent

[

[

M 34 (2013) 92–100

Transportation Systems, 2002. Proceedings. The IEEE 5th InternationalConference; 2002, p. 162–7.

13] Gonzalez-Mendoza M, Santana-Diaz A, Hernandez-Gress N, Titli A.Driver vigilance monitoring, a new approach. Intelligent Vehicle Sympo-sium 2002 IEEE 2002;2:358–63.

14] Agrimip Sud-Ouest Innovation. Ecosilo : une protection raisonnéedes céréales contre les insectes dans les silos de stockage. Agricul-ture/Agroalimentaire. [Online]. Available: http://competitivite.gouv.fr/poles-en-action/fiche-projet-popup-18/ecosilo-178.html?cHash=cf40f15f6edfcabd037adf5fedbe8ec1

15] Yongli G, Yin OS, Han PY. State of the art: a study on fall detection. ManCybern B Cybern 2012;42:729–39.

16] Lombardi A, Ferri M, Rescio G, Grassi M, Malcovati P. Wearable wirelessaccelerometer with embedded fall-detection logic for multi-sensor ambientassisted living applications. Sensors 2009 IEEE 2009, p. 1967–70.

17] Buzan D, Sclaroff S, Kollios G. Extraction and clustering of motion trajec-tories in video. Pattern Recognition, 2004 ICPR 2004 Proceedings of the17th International Conference 2004;2:521–4.

18] Zhang Z, Huang K, Tan T. Comparison of similarity measures for trajectoryclustering in outdoor surveillance scenes. Pattern Recognition, 2006 ICPR2006 18th International Conference 2006;3:1135–8.

19] Hastie T, Tibshirani R. Discriminant adaptive nearest neighbor classifica-tion. IEEE Trans Pattern Anal Mach Intell 1996;18(6):607–16.

20] Drees. Enquête sur l’allocation personnalisée d’autonomie. Statistiques au1er trimestre 2011 no 2; 2011.