Senso s 2014, 14, 18543-18574; doi:10.3390/s141018543
senso s
ISSN 1424-8220
www.mdpi.com/jou nal/senso s
A icle
Compa ison and Cha ac e iza ion o And oid-Based Fall
De ec ion Sys ems
Ra ael Luque, Edua do Casila i *, Ma ía-José Mo ón and Gema Redondo
Uni e sidad de Málaga, Depa amen o de Tecnología Elec ónica, ETSI Telecomunicación,
29071 Málaga, Spain; E-Mails: [email p o ec ed]s (R.L.); m[email p o ec ed] (M.-J.M.);
[email p o ec ed] (G.R.)
* Au ho o whom co espondence should be add essed; E-Mail: ecasila [email protected];
Tel.: +34-952-132-755 (ex . 123); Fax: +34-952-131-447.
Ex e nal Edi o : Nauman Aslam
Recei ed: 25 June 2014; in e ised o m: 22 Sep embe 2014 / Accep ed: 23 Sep embe 2014 /
Published: 8 Oc obe 2014
Abs ac : Falls a e a o emos sou ce o inju ies and hospi aliza ion o senio s.
The adop ion o au oma ic all de ec ion mechanisms can no iceably educe he esponse
ime o he medical s a o ca egi e s when a all akes place. Sma phones a e being
inc easingly p oposed as wea able, cos -e ec i e and no -in usi e sys ems o all de ec ion.
The exploi a ion o sma phones’ po en ial (and in pa icula , he And oid Ope a ing Sys em)
can bene i om he wide implan a ion, he g owing compu a ional capabili ies and he
di e si y o communica ion in e aces and embedded senso s o hese pe sonal de ices.
A e e ising he s a e-o - he-a on his ma e , his s udy de elops an expe imen al
es bed o assess he pe o mance o di e en all de ec ion algo i hms ha g ound hei
decisions on he analysis o he ine ial da a egis e ed by he accele ome e o he
sma phone. Resul s ob ained in a eal es bed wi h di e se indi iduals indica e ha he
accu acy o he accele ome y-based echniques o iden i y he alls depends s ongly on
he all pa e n. The pe o med es s also show he di icul y o se de ec ion accele a ion
h esholds ha allow achie ing a good ade-o be ween alse nega i es ( alls ha emain
unno iced) and alse posi i es (con en ional mo emen s ha a e e oneously classi ied as
alls). In any case, he s udy o he e olu ion o he ba e y d ain e eals ha he ex a
powe consump ion in oduced by he And oid moni o ing applica ions canno be neglec ed
when e alua ing he au onomy and e en he iabili y o all de ec ion sys ems.
OPEN ACCESS
Senso s 2014, 14 18544
Keywo ds: all de ec ion; sma phone; eHeal h; And oid; accele ome e
1. In oduc ion
Owing o he socio-economic and heal h p og ess expe ienced by de eloped coun ies in he las
20 yea s, he olde popula ion has subs an ially inc eased, especially wi h he aging “baby boome s”
( hose bo n be ween 1946 and 1964). The ema kable g ow h o li e expec ancy has mul iplied he
numbe o senio ci izens ha ace daily he isks o li ing on hei own. Al hough i is well known ha
physical exe cise a oids o delay he onse o diseases, i can also lead o alls, he majo heal h
haza d ha diminishes he quali y o li e. Da a om he Wo ld Heal h O ganiza ion [1,2], suppo ed
by di e en epidemiologic s udies, indica e ha a no iceable pe cen age o senio s aged o e 64
(28%–35%) su e a all each yea . This p opo ion inc eases o 32%–42% o hose o e 70 yea s o
age. In ac , inju ies caused by alls a e one o he main causes o hospi aliza ion o olde pe sons,
equen ly esul ing in a se ious educ ion o hei independen li ing skills and e en dea h. A as
eac ion can ema kably diminish he e ec s o a all on an olde adul , bu an immedia e assis ance is
o en no easible i he inju ed indi idual li es alone and he inju ies p e en he pa ien om seeking
help. Acco ding o [3], he app aised all incidence o independen li ing people o e 75 exceeds 30%
annually, as long as i has been es ima ed ha up o 50% o nu sing home esiden s su e om alls
e e y yea (wi h mo e han 40% alling a leas wice a yea ). In addi ion, up o 12% o all alls cause a
ac u e while 23% o auma ela ed-dea hs in pa ien s olde han 65 (34% in hose olde han 85 yea s)
ollow a all (see [4] o a s a e-o - he-a on his opic). Howe e , physical damages associa ed o alls
a e no he only nega i e e ec ha mus be conside ed. Fea O Falling (FOF) has been ecognized as
a speci ic heal h p oblem, especially o olde people. FOF, which is ypically connec ed o an inc ease
o neu o icism and anxie y, no mally leads pa ien s o s ikingly educe o e ade physical ac i i y.
Thus, he psychological and emo ional consequences o a all con ibu e o deg ade he independence
o he elde ly. Mo eo e , his loss o sel -con idence de e io a es as olde people age, leading hem o a
mo e acu e social isola ion and a lowe quali y o li e.
This pape p esen s he p o o ype o an expe imen al sys em o all moni o ing. The p o o ype
combines an And oid-based sma phone as he pla o m ha dwa e, a mo ion senso (a buil -in i-axial
accele ome e ) and he loca ion se ices suppo ed by he sma phone. The elec ion o a mobile
phone-based sys em has e iden ad an ages. On one hand, mobile phone-based applica ions can
ope a e almos e e ywhe e because o he popula i y, dec easing cos s and po abili y o mobile
de ices and he ubiqui y o mobile echnologies. In ac , he use o sma phones has g own o become
a basic cons i uen o daily ou ine. Besides, mos cu en sma phones seamlessly in eg a e all he
equi ed elemen s (accele ome e s and gy oscopes) o de elop au onomous and sel -su icien all
de ec ion applica ions. An impo an poin in he design o any heal hca e moni o ing applica ion is
e gonomics. Wi eless communica ions clea ly imp o e pa ien s’ mobili y while he eu iliza ion o an
al eady exis ing pe sonal de ice a oids he annoyances o ca ying a sepa a e all de ec ion gadge .
Thus a sma phone o ien ed sys em does no in oduce any speci ic a achable componen in he li e o
Senso s 2014, 14 18545
olde people, who in u n a e becoming less eluc an o admi no el echnologies o imp o e hei
sa e y and independence.
This pape is o ganized as i ollows: a e he in oduc ion o his Sec ion 1, Sec ion 2 e ises he
ela ed wo ks. Sec ion 3 p esen s he gene al s uc u e and objec i es o he de eloped sys em. Sec ion 4
desc ibes he design o he de ec ion algo i hms o be es ed whe eas he global sys em a chi ec u e and
implemen a ion a e p esen ed in Sec ions 5 and 6. In Sec ion 7, he pe o mance o he sys em and he
accu acy o he accele a ion-based de ec ion algo i hms a e e alua ed by means o ex ensi e expe imen s
pe o med on di e en scena ios. Finally, Sec ion 8 d aws he main conclusions o he wo k.
2. Rela ed Wo k
Due o he ad ances in he a ea o elec onic senso s and he widesp ead ex ension and cos
educ ion o pe sonal de ices, he esea ch on sys ems o all de ec ion has ocke ed du ing he las
decade. Recen comp ehensi e wo ks ha e ho oughly add essed he s a e-o - he-a on all de ec ion
sys ems. In his sense, di e en c i e ia ha e been p oposed o ca ego ize he exis ing p oposals.
Fo example, he au ho s in [5] di e en ia e hose s udies ha only ake in o accoun he ecogni ion o
he impac shock om hose ha also conside he “pos - all” phase. Al e na i ely, Pe y classi ies he
de ec ion echniques depending on whe he he use ’s accele a ion is measu ed and u ilized o iden i y
he all [6]. The s a e-o - he-a p esen ed in [7] also dis inguishes be ween wo gene al ypes o all
de ec ion a chi ec u es: con ex -awa e sys ems and hose schemes ha employ wea able de ices wi h
embedded accele ome e s and, in some cases, gy oscopes aimed a sensing he use ’s posi ion. As i
e e s o his la e ype o all de ec ion sys ems wi h body-wo n senso s, he FARSEEING p ojec
( unded by he Eu opean Commission) has suppo ed a sys ema ic bibliog aphic e ision [8] o he
ela ed li e a u e. In he conclusions, he s udy c i icizes he lack o a me hodological consensus o
e alua e he p oposed sys ems.
The ecen s udy in [9] o e s an in e es ing axonomy o he sys ems and algo i hms o all
de ec ion. The pe o med classi ica ion conside s h ee gene al ca ego ies: ambience de ice based,
ision based and wea able de ice based de ec ion sys ems. The ambien based app oaches p opose o
combine audio isual in o ma ion and e en sensing by cap u ing and analyzing loo ib a ional da a.
The alls a e acked by means o p essu e senso s in a ound he use . This may be a cos e ec i e
me hod bu i leads o many alse ala ms due o spu ious alls o o he objec s. On he o he hand
ision based assis i e sys ems u ilize came as (o e en mic ophones) o assess he use ’s beha io and
de ec e en s (as alls) wi hou excessi e in usion in his/he ou ines. The e a e di e en s a egies o
pe o m he all de ec ion om he analysis o he ideo images, such us shape modelling using
spa io empo al ea u es, s udy o he shape changes o he pos u e, 3D head posi ion analysis, e c. Fo
example, he sys ems by Ande son [10], Cucchia a [11], Di aco [12], Fo oughi [13], Fu [14],
Hazelho [15], Jansen [16], Lee [17], Liu [18], Miaou [19], Ni [20], Rougie [21], Sixsmi h [22],
Vishwaka ma [23] o Yu [24] p opose o de ec he alls by p ocessing he cap u ed images o he
moni o ed pa ien (o use ). Howe e his app oach p esen s se e e p ac ical limi a ions. On one hand
he equi ed moni o ing en i onmen (which is no mally limi ed o a closely obse ed oom) is di icul
o implemen and expensi e o main ain. Mo eo e , he quali y o he images (and consequen ly, he
accu acy o all p edic ion) may be s ongly de e mined by he illumina ion condi ions o he oom o
Senso s 2014, 14 18546
by he exis ence o blind spo s whe e he use canno be p ope ly moni o ed [25]. Apa om he
ulne abili y o noises om en i onmen al objec s [26], ano he p oblem o mos exis ing ision based
app oaches is he absence o lexibili y [9], as hey a e usually case speci ic and a e designed and
op imized o a e y pa icula scena io. Besides, he indi idual’s p i acy is comp omised, which
di ec ly ansla es in o a lack o accep ance among use s.
In any case, bo h ambien and ision (o image and audio p ocessing) based schemes apply con ex
awa e echniques ha equi e de ining a supe ision en i onmen whe e he use ac i i y is supposed
o ake place. In a simila way, he p ojec desc ibed in [27] p esen s a all de ec ion a chi ec u e which
is deployed h ough a ne wo k o non-in asi e wea able sensing mo es and an in as uc u e o ixed
mo es ha a e con enien ly dis ibu ed wi hin he supe ised moni o ing scena io. Mo es a e equipped
wi h low-powe MEMS accele ome e s so ha when a all is de ec ed, he ale is o wa ded o a base
s a ion ia he ixed mo e ne wo k. This ne wo king in as uc u e also allows he localiza ion o he
use . The de ec ion algo i hm is based on he angle o change, which is es ima ed om a e aging he
do p oduc o he accele a ion ec o s o e 1 second.
Con e sely, he app oaches ha employ wea able de ices inco po a e speci ic ga men s wi h
embedded senso s (accele ome e s) o es ima e he mo ion (and, in some cases, he loca ion) o he
use ’s body in any unsupe ised en i onmen . I he wea able ga men s a e also enabled wi h a wide
a ea communica ion in e ace (e.g., 3G/4G da a connec ions), he use can be ubiqui ously moni o ed.
Mos sma phone-based all de ec ion sys ems can be clea ly ca ego ized in o his amily o
“ubiqui ous” de ec ion echniques. As a main ad an age, sma phone solu ions do no cons ain he
use mobili y o a pa icula moni o ing zone while hey p o ide a seamless inexpensi e echnology
ha is al eady in eg a ed in he daily li e o mos po en ial use s.
Nowadays, sma phones inco po a e a wide se o embedded senso s, including no only
accele ome e s, bu also came as, mic ophones, digi al compasses, gy oscopes o GPS uni s.
The apidly dec easing cos o he sma phones has os e ed he adop ion o his wea able echnology
and he measu emen o he accele a ion as he basis o all de ec ion. Consequen ly many
sma phone-based a chi ec u es ha e been p oposed o e he las yea s.
Ini ial sma phone-based all de ec ion sys ems [28,29] we e de eloped using he ( oday obsole e
and discon inued) Symbian OS on Nokia phones. In his sense, Google’s And oid is he mos dominan
playe in he sma phone indus y, wi h 78.1 pe cen o he ma ke sha e a he end o 2013 [30].
As a consequence, And oid is selec ed as he Ope a ing Sys em (OS) and p og amming en i onmen
massi ely adop ed by he li e a u e on sma phone-based all de ec ion solu ions. Con e sely, he e is
much less li e a u e de o ed o all de ec ion a chi ec u es deployed on o he ope a ing sys ems o
mobile de ices, such as iOS (as he sys ems p esen ed in [31,32], o [33], whe e an iPhone is in cha ge
o ecei ing and p ocessing he signals om di e se ex e nal mobili y senso s o wa n he use abou
po en ial alls o o es ima e he all isk o pos s oke pa ien s). In [34] (a wo k o 2011) a gene ic all
de ec ion Ja a mul ipla o m so wa e a chi ec u e (using an ex e nal accele ome e ) is implemen ed in
bo h a Symbian phone (Nokia 5800) and And oid sma phones (Samsung Galaxy, HTC He o).
Howe e , he sys em is no es ed and no compa ison be ween he pe o mances o hese wo OS is
o e ed. On he o he hand, au ho s in [35] discuss he capabili ies o di e en mobile ope a ing
sys ems (Windows Phone, Meego Ha ma an, Symbian and And oid) o de elop applica ions o all
Senso s 2014, 14 18547
de ec ion. Au ho s conclude ha he bes op ion is And oid as i p o ides mo e suppo om APIs and
educes he implemen a ion ime.
Pe FallD is he And oid a chi ec u e de eloped by he au ho s o he heo e ical s udy in [36,37].
The sys em, which is ounded on he echnique o he accele a ion h esholds, has se ed as a one o
he e e ences o he implemen a ion o he all de ec ion algo i hms es ed in his a icle. As la e
discussed, he de ec ion decision o Pe FallD is ca ied ou conside ing he alues o he o al
accele a ion o he phone body and he absolu e e ical accele a ion du ing a ce ain obse a ion ime
window. A e y simila p ocedu e is ollowed by he sys ems p esen ed in [38,39]. In he second case,
he pla o m also in o ms abou he di ec ion o he all. In o de o de ec ee alls, au ho s in [40]
sugges measu ing he ne displacemen o he use o he ime du ing which he accele a ion is close
o ze o. Al hough he algo i hm is in ended o And oid-enabled de ices, he de ec ion echnique is no
implemen ed no es ed in an ac ual sma phone.
In o de o minimize he equi emen s o compu a ional powe and eal ime p ocessing in he
sma phone, o he sys ems, such as ha p oposed in [41], jus conside a simple single- h eshold
algo i hm o deploy he all de ec ion. In ano he wo k o he same au ho s [42], an impac is p esumed
i he accele a ion is g ea e han an empi ically adjus ed h eshold o 2.3 g. An impac is only
conside ed o be a all i he inal use o ien a ion is ho izon al.
Ano he And oid App o sma phone-based all de ec ion is iFall (And oid Applica ion o Fall
Moni o ing and Response) [43], which is a ailable a Google Play S o e. The heo e ical basis o he
employed de ec ion algo i hm is simila o hose o Pe FallD, as he u ilized algo i hm is also g ounded
on he same accele a ion h eshold echnique. Unlike Pe FallD, he all de ec ion c i e ion conside ed
by iFall uniquely depends on he global accele a ion.
The wo k in [44] combines a sma senso and came a-enabled sma phone o de ec and check he
occu ence o a all. As soon as he senso de ec s a all e en , li e da a a e s eamed o a emo e
moni o ing poin while he pa ien is pe sonally a emp ed o be con ac ed o ge a ocal o keypad
eedback. In [45] an And oid sma phone wi h 3-axial accele ome e is again conside ed as a eleheal h
de ice. The connec ion o he emo e elemoni o ing uni is accomplished by means o a TCP/IP socke
ia Wi-Fi. The pape in [46] p esen s ano he sma phone-based sys em ha acks he use ’s mo emen s
and au oma ically ansmi s an ala m message o he ca egi e s whene e a all is ecognized.
A mul ilaye pe cep on (i.e., a neu al ne wo k) is u ilized o analyze he use ’s mobili y pa e n.
Vie [47] join ly conside s he in o ma ion om he o ien a ion senso and he accele ome e o a
sma phone o de ec he alls. The au ho s s a e ha he pos u e o he use be o e he all mus be
aken in o accoun o op imize he h eshold ha de e mines a de ec ion. The elec ion and op imiza ion
o he decision h esholds and he obse a ion ime window in accele ome e -based And oid p og ams
a e s ill open issues. The And oid p og am po ayed in [48] pa ame e izes hese alues aking in o
accoun he age, sex and Body Mass Index o he use o be moni o ed.
The au ho s in [49] in oduce a sys em ha makes use o a sma phone wi h an embedded
i-accele ome e moun ed on he wais . By analyzing he da a om he accele ome e , he sma phone
ob ains he in o ma ion abou he use ’s mo ion, which is ca ego ized acco ding o i e di e en
pa e ns. A all is assumed o occu when he Signal Magni ude A ea (calcula ed om he in eg a ion
o he accele ome e signals), he accele a ion magni ude ec o and he il angle simul aneously
Senso s 2014, 14 18548
exceed he co esponding h esholds. In ha case, a Mul imedia Messaging Se ice (MMS) wi h GPS
coo dina es is sen o he emo e moni o ing poin .
The s udy in [50] e alua es he speci ici y and sensi i i y o he sma phones o de ec alls when
hei pe o mance is compa ed o ha achie ed wi h independen accele ome e s. Resul s show ha
sma phones a e a alid op ion o de ec ing alls wi h high accu acy.
Mos p oposed sma phone-based solu ions a e “sma phone-only” a chi ec u es whe e no
supplemen a y componen s (apa om he phone’s buil -in senso s) a e u ilized [51]. The
inco po a ion o addi ional high esolu ion senso s in he sys em may in oduce a highe accu acy in
he measu emen s while a oiding he need o wea ing he sma phone in a ixed posi ion (whe e he
de ec ion p ocess is assumed o be op imal). On he o he hand, ex a elemen s inc ease he cos o he
sys em ha dwa e as well as i may educe he pe cei ed usabili y o he all de ec o (as mo e de ices
mus be wo n by he use ). Fo example, he sys em p esen ed in [52] combines a sma phone and a
digi al wa ch wi h wi eless communica ions ( he EZ430-Ch onos model p o ided by Texas
Ins umen s, which in eg a es a h ee-axis accele ome e ) o de ec he alling and o ge in ouch wi h
he eme gency con ac s.
The pla o m desc ibed in [53] also in eg a es an ex a elec onic de ice (a Senso Tag om Texas
Ins umen s) o moni o he use ’s mo emen s. The accele ome y da a a e p ocessed in he Senso Tag,
so i a all is de ec ed, a message is sen ia Blue oo h o he sma phone (which jus ac s as a
communica ion ga eway be ween he ex e nal sensing de ice and he emo e moni o ing poin ). Pape
in [54] implemen s a all de ec ion sys em wi h an And oid-based wa ch equipped wi h a i-axial
g a i y accele ome e . The wa ch does no inco po a e any Wide A ea Ne wo k communica ion
in e ace. Thus, upon de ec ion o a all, he de ice jus issues a ib a o y ala m.
As i can be deduced om he a o emen ioned s udies, accele ome y is, by a , he mos
ex ensi ely employed me hod o he de ec ion o alls in sma phones. The posi ion whe e he
accele ome e is loca ed (wais , w is , high o ches ) has been used as a c i e ion o classi y he
exis ing solu ions based on i-axial accele ome e s [55]. In some sys ems, he in o ma ion ob ained
om he accele ome e s is combined wi h da a om ine ial senso s o gy oscopes.
The e a e wo gene al s a egies o de ec he all occu ences om he da a ob ained by he
accele ome e s:
• Classi ica ion o he mo emen ounded on Pa e n Recogni ion Me hods (PRM) ha employ
da a bases, aining phases and AI (A i icial In elligence) solu ions. In he sys ems p oposed
by Gan i [56] and Ka an onis [57], in o de o a oid alse ala ms, he ac i i y pa e ns o he
pa ien a e cha ac e ized. Fo ha pu pose, hese au ho s p opose a complex “ aining” phase
whe e he alues measu ed by he mobili y senso s when he use pe o ms daily ac i i ies o
di e se na u e a e s o ed in a da abase. Once his cha ac e iza ion is inished, hese alues
p e iously cap u ed a e u ilized o dis inguish he mo emen s o a no mal si ua ion om an
ale condi ion. Au ho s in [58] employ a wais wo n all de ec ion sys em o compa e
di e en machine lea ning classi ica ion algo i hms o de ec alling pa e ns. The s udy
concludes ha mul ilaye pe cep ons pe o m be e han o he classi ica ion echniques.
This ype o s a egies pe mi s uning he de ec ion algo i hm o he pa icula beha io o he
use . Con e sely, he need o inco po a ing aining phases, da abases and/o AI echniques
Senso s 2014, 14 18549
hinde s hei implemen a ion in a ha dwa e and ba e y limi ed mul i unc ional de ice
like a sma phone.
• De ec ion based on accele a ion h esholds (Th eshold Based De ec ion o TBD): In [59]
Nyan u ilizes an accele a ion h eshold ha is based on he absolu e peak alues o he
accele ome e ’s measu emen s. In he expe imen , he accele ome e is anspo ed in a
ga men nea he shoulde . Kangas [60,61] p oposes ou h esholds o he ollowing
magni udes: he e ical accele a ion, he o al accele a ion ec o , he dynamic accele a ion
ec o and he di e ence be ween he maximum and minimum modules o he accele a ion. A
all is assumed whene e one o hese h esholds is c ossed. The esul s o he pe o med es s
show ha a simple iaxial accele ome e a ached o he wais o head can be accu a e enough
o de ec mos alls, e en wi h qui e simple “ h esholding” algo i hms. Be o e he appa i ion
o he sma phones, he de ec ing de ices ha we e employed by hese algo i hms equi ed a
speci ic and, in some cases, sca cely po able ha dwa e. Howe e , as sma phone na i ely
in eg a es accele ome e s and gy oscopes, me hods based on accele a ion- h esholds p o ide
a good ade-o be ween he esul s and he complexi y o he algo i hm’s implemen a ion.
Table 1 o e s a gene al e iew o he li e a u e on And oid-based sys ems in ended o all de ec ion
o all p edic ion. Fo his pu pose an ex ensi e bibliog aphic e ision was pe o med and 56 a icles
( om 2009 o May 2014) p oposing And oid all de ec o s we e ound. The able classi ies he
p oposals acco ding o di e en c i e ia, summa izing he main cha ac e is ics o he all de ec o s.
The i s classi ica ion dis inguishes he gene al opology o he sys em: body-wo n o con ex -awa e
sys ems. As he able shows, he majo i y o he a chi ec u es can be ca ego ized as body-wo n
sys ems, i.e., body a ea ne wo ks ha include an And oid-enabled de ice (no mally a sma phone).
Ne e heless he e exis also examples whe e he And oid de ice is he cen e o a con ex awa e
sys em (e.g., an embedded compu e ins alled on a wall in [62], which de ec s he alls by means o a
Dopple senso ). Simila ly he e a e also a chi ec u es [63–65] ha combine con ex awa e echniques
and body-wo n de ices. Fo ins ance, in [63] au ho s p opose o combine he da a o ideo came as and
he accele ome e o a wea able And oid pla o m o cha ac e ize he use mobili y and de ec alls.
An impo an aspec in he a chi ec u e o he de ec ion sys em is he ole o he And oid de ice.
Table 1 in o ms i he And oid de ice is employed as a Senso (S), as a Da a Analyze (DA) o decide
i a all has occu ed, as a Communica ion Ga eway (CG) o e ansmi he sensed da a (o he all
de ec ion decision) o a emo e se e , o /and jus as Remo e Moni o ing Uni (RMU) o e ing an
in e ace o wa n abou he alls. As i can be examined in he able, mos solu ions bene i om he
sensing, compu ing and communica ion capabili ies o sma phones, which can implemen and execu e
he all de ec ion algo i hm while simul aneously ac ing as a senso and as a da a ga eway.
In his sense, he able also di e en ia es hose sys ems ha concen a e all he unc ionali ies on a
single sma phone (sma phone-only o SP-only sys ems) om hose which combine a sma phone and
one o se e al ex e nal wi eless senso s. On he o he hand, he e a e a ew examples o sys ems ha
do no conside he use o a sma phone and a e deployed on Speci ic De ices (SD) (a ha dwa e
pla o m pu posely designed o all de ec ion). In his ca ego y we could men ion he wo k in [41]
whe e an And oid-based wa ch is u ilized as he mobili y senso o he sys em.
Senso s 2014, 14 18550
The senso (o senso s) ha he e ised sys ems employ is indica ed in a pa icula column o
Table 1. The measu emen s om he buil -in i-axial accele ome e s o he sma phones a e by a he
mos u ilized magni udes o e alua e he possibili y o a all occu ence. Jus in some cases, embedded
sma phone gy oscopes o ex e nal accele ome e s a e conside ed. The small ange and p ecision o
buil -in accele ome e s ha e been s a ed as inapp op ia e o de ec alls [66]. Howe e , au ho s in [67]
ound ha he use o dedica ed accele ome e s o de ec alls p esen s simila esul s o hose ob ained
wi h sma phones.
Finally, he all decision algo i hm ha is u ilized by he sys em is shown in he las column o he
able. The able e eals ha simple Th eshold-Based De ec ion (TBD) sys ems a e p e e ed o
complex Pa e n Recogni ion Me hods (PRM), which usually impose highe compu ing and memo y
cos s as well as long aining phases o adap he algo i hm o he pa icula cha ac e is ics o he use
o be moni o ed.
Jus a ew o he a o emen ioned p oposed And oid sys ems (such as iFall [43] o ha by
Ke dega i [58]) ha e been eleased o he gene al public. O he a ailable apps (Span ec Fall De ec o ,
Fall Moni o , T3LAB, Fade Fall De ec o , e c.) do no p o ide any de ailed in o ma ion abou he
de ec ion algo i hm employed o any insigh abou he pe o mance achie ed by he so wa e.
As a ma e o ac he e is no consolida ed And oid-based p oduc in his ealm (see Google Play
S o e [68] o mo e de ails abou hese applica ions).
As i e e s o comme cial sys ems o all de ec ion, mos exis ing p o essional solu ions a e
no mally equipped wi h a speci ic a achable ha dwa e which is in cha ge o measu ing he use ’s
mo ion. We can men ion he ollowing p oduc s:
-B ickhouse [69] p o ides a ypical sys em wi h wo unc ional componen s: a po able senso
(which is a ached o he bel and placed on he use ’s wais ) o de ec he mo emen s, and a ixed
ga eway connec ed o he wi ed phone line. The ga eway is in cha ge o ecei ing he signals om he
senso and communica ing any e en ual eme gency si ua ion o he medical s a . Ob iously, his high
cos ly sys em can only ope a e in a home supe ised en i onmen whe e he dis ance be ween he
senso and he ga eway is below a maximum alue in o de o gua an ee he iabili y o he
connec ion. Besides, he employed algo i hm o de ec he alls is no desc ibed.
-Be e buys [70] is an economic sys em which is deployed h ough a se o senso s loca ed in
cushions a hose household loca ions commonly equen ed by he use (on he chai s, beds, loo
ma s, e c.). The de ice is equipped wi h a olume con ol and an in e phone so ha , in case o
eme gency, i emi s a musical one and a lashing LED- ype ligh signal o ale anyone nea by.
Howe e , as in he p e ious example, his sys em has e iden limi a ions o achie e a pe sis en
ubiqui ous all de ec ion.
-ITT EasyLi eS [71] is ano he ype o sys em ha in eg a es he wo componen s (senso and
communica ion uni ) in he same e minal. I consis s o a mobile phone equipped wi h a balance
senso . The manu ac u e s claim ha when he phone is d opped, i au oma ically p oceeds o dial he
co esponding eme gency phone numbe . I s main d awbacks a e wo old: Fi s ly i employs an
uncon en ional de ice, and secondly, he elec ion o he igge ing h esholds is oo simple o ensu e
an accu a e de ec ion.
Senso s 2014, 14 18551
Table 1. Classi ica ion and main cha ac e is ics o he And oid-based solu ions o all de ec ion sys ems.
Pape Yea
GENERAL
TIPOLOGY
ROLE OF THE ANDROID DEVICE
(Possible Role(s) o he And oid-Enabled
De ice:
-Senso (S),
-Da a Analyze o Fall De ec ion (DA),
-Communica ion Ga eway (CG)
-Remo e Moni o ing Uni (RMU)
COMPONENTS:
SP-Only,
SP Combined wi h
Ex e nal Senso s, SD
(Speci ic De ice)
EMPLOYED SENSORS
FALL DECISION
ALGORITHM
-Th eshold-Based De ec ion
(TBD)
- Pa e n Recogni ion Me hods
(PRM)
[43] 2009 Body Wo n S, DA, CG SP-only Buil -in i-axial accele ome e TBD
[36,37] 2010 Body Wo n S, DA, CG Combined (SP and an
ex e nal magne )
Buil -in accele ome e (in [37] also
a magne ic senso ) TBD
[72] 2010 Body Wo n S, DA, CG SP-only Buil -in i-axial accele ome e TBD
[73] 2010 Body Wo n S, DA, CG SP-only buil -in i-axial accele ome e and magne ome e PRM
[74] 2011 Body Wo n S, DA, CG SP-only Buil -in i-axial accele ome e A combina ion o TBD
and PRM (s a e machine-based)
[50] 2011 Body Wo n S, DA, CG SP-only Buil -in i-axial Bosch Senso ec’s 3-axis
BMA150 accele ome e TBD
[75] 2011 Body Wo n CG Combined Speci ic And oid based Pe sonal Ac i i y
Moni o wi h accele ome e TBD
[76] 2011 Body Wo n S, DA, CG SP-only Buil -in accele ome e TBD
[41] 2011 Body Wo n S, DA, CG SP-only Buil in accele ome e and o ien a ion senso TBD
[77] 2011 Body Wo n S, DA, CG SP-only Buil -in i-axial accele ome e TBD
[47,78] 2011
2012 Body Wo n S, DA SP-only Buil -in i-axial accele ome e TBD
[79] 2012 Body Wo n S, DA, CG SP-only Buil -in i-axial accele ome e PRM: ini e s a e machine
[80] 2012 Body Wo n S, DA SP-only Buil -in i-axial accele ome e PRM: sel o ganizing map
[81] 2012 Body Wo n S, DA SP-only Buil -in i-axial accele ome e TBD
[82] 2012 Body Wo n DA, CG
Combined (SP and
ex e nal
accele ome e )
Ex e nal iaxial accele ome e ADXL345 o Analog
De ices connec ed o a BT-enabled wea able uni TBD
Senso s 2014, 14 18558
A high sampling equency o he accele a ion ec o is no mally es ablished o a p ope
compu a ion o FI in he case o sudden alls. Howe e , acco ding o his s a egy, mos alls ha occu
slowly (i.e., wi hou sudden a ia ions o he accele a ion) may go unno iced.
4.3. Pe FallD
Pe FallD [37] algo i hm simul aneously akes in o accoun he alues o he modules o he o al
accele a ion o he phone ( T
A
) and he accele a ion a he absolu e e ical di ec ion ( V
A
), which can
be es ima ed as:
zyzyyzx AAAA
θθθθ
coscossinsin −+= (3)
whe e θy and θz a e he measu ed pi ch and oll alues, which de e mine he mobile phone’s
o ien a ion. These angles a e sensed by he gy oscope in eg a ed in he sma phone.
The algo i hm sepa a ely analyses |AT| and |AV|. Thus, in o de o assess he occu ence o a all, he
algo i hm conside s wo phases o bo h pa ame e s.
I he di e ence o he es ima ed alue o |AT| wi hin an obse a ion ime window (win ) su passes a
ce ain igge ing h eshold (Th ), he pa e n ecogni ion phase is ini ia ed. Du ing his second phase
he di e ence be ween he maximum alue and he minimum alue o |AT| is compu ed wi hin a second
checking ime window (winc ) a e win . I his di e ence does no exceed ano he h eshold (Thc ), a
possible all is conside ed o be de ec ed. A simila p ocess is applied o |A |, wi h he co esponding
ime windows win and winc and he h esholds Th and Thc . A all is only assumed o ha e occu ed
i bo h de ec ion condi ions abou |AT| and |A | a e sa is ied.
4.4. iFall
This algo i hm [43] akes in o conside a ion ha a all ini ially p o okes a sudden and signi ican
dec ease in he accele a ion ampli ude. A e his “ ee- all-pe iod”, he accele a ion expe iences an
ab up spike as soon as he body hi s he loo . Consequen ly, i he accele a ion |AT| c osses a lowe
and an uppe h eshold du ing a ce ain obse a ion ime window, a all is suspec ed. Howe e , he all
is only epo ed i he pa ien eally begins om an up igh posi ion and ends in a ho izon al posi ion.
Fo ha pu pose, i he e ical posi ion is es o ed (o i a d opped sma phone is picked up) wi hin a
“pos - all” obse a ion pe iod, he de ec ion e en is neglec ed. O he wise, i he e ical posi ion is no
eco e ed be o e his ime expi es, he sys em assumes ha he pa ien is lying on he g ound and he
ala m is emi ed.
5. Sys em Design
The gene al wo k low o he de eloped p og am is illus a ed in Figu e 2. As soon as he p og am is
s a ed, a use p o ile is loaded con aining he con igu a ion o he all de ec ion sys em (selec ed
de ec ion algo i hm, sampling equency o he accele ome e , h esholds, e c.) and he pe sonal da a and
p e e ence o he use (e.g., an eme gency con ac lis , ala m one, e c.). These pa ame e s o he
p o ile a e ully con igu able by he use .
Senso s 2014, 14 18559
Figu e 2. Wo king p ocedu e o he sys em.
A e he p og am is pa ame e ized, he moni o ing p ocess is launched. Thus, he eal- ime da a
collec ed by he accele ome e s a e pe manen ly compa ed o he de ec ion h esholds acco ding o he
selec ed algo i hm ( he modules wi hin he smalles dashed box o he igu e a e execu ed). I he
p ese h esholds a e su passed (a all is p esumed because o a ce ain alue o alues o he
accele a ion), a “s a iona y phase” is ini ia ed o con i m ha a all may ha e occu ed. This phase
co esponds o he “pa e n ecogni ion phase” and he “pos - all” obse a ion pe iod o he Pe FallD
and he iFall algo i hms, espec i ely. On he con a y, he du a ion o his phase is se o 0 i hese
algo i hms a e no conside ed.
When he s a iona y phase concludes (and he all is con i med), ano he ime is execu ed. Du ing
his new pe iod, he sma phone emi s an acous ic ala m o in o m he use ha a all has been
D
e ec ion algo i hm design
Yes
Is s a iona y ime
ended?
S a S a iona y
Time
S
TART
END
Load Use P o ile
Adjus Sampling
F equency
Real Time
Moni o ing P ocess
Go o Eme gency Con ac
Lis
Is ale
s opped in
ime?
-S a ale ime
-Loud ‘beeping’
T igge ing
condi ion
sa is ied?
Send
inal SMS
-Ge loca ion
-Ge imes amp
N
o
Yes
Powe Awa eness Design
No
Yes
FALL PATTERN
RECOGNITION
No
END
Yes
No
Make call?
Senso s 2014, 14 18560
de ec ed. I no esponse om he use is ecei ed h ough his ime (a sc een bu on is no ouched), a
all is assumed and he ale no i ica ion is igge ed. In his case, he applica ion ob ains he GPS
coo dina es o he use and a imes amp. Depending on he con igu a ion o he p og am, his
in o ma ion can be di ec ly sen by a ex message o a se o p ede ined con ac s (selec ed by he use
in he eme gency con ac lis ) o , o he wise, he applica ion can make a phone call o a ce ain numbe
also speci ied in he con igu a ion. On he o he hand, i he acous ic ala m is u ned o be o e he
co esponding ime elapses, he applica ion e u ns o he no mal moni o ing p ocess.
6. Sys em Implemen a ion
The p o o ype is ini ially de eloped on a HTC Desi e X sma phone. This de ice ea u es an
ARM-based a chi ec u e, dual-co e CPU wo king a 1 GHz, wi h 768 MB RAM memo y, a 10.16-cm
sc een, GPS senso and 4 GB o in e nal s o age. I is powe ed by a 1650 mAh echa geable li hium
ion ba e y and inco po a es an embedded accele ome e /G-senso . The OS e sion employed by he
phone is And oid 4.0. The sys em is pu in o ope a ion by he wo abo emen ioned so wa e
applica ions: AppPe FallD, which implemen s he de ec ion algo i hms, and AppLoca ionInMaps
(in he emo e moni o ing poin ). Bo h p og ams a e implemen ed in Ja a, wi h Eclipse and And oid
De elopmen Tools (ADT) plugin. The sys em was also ins alled and es ed on an HTC Sensa ion XE
model, p o ided wi h simila senso s, a 10.922-cm sc een and And oid 2.3.4 Ginge b ead OS.
Figu e 3. Snapsho s o he Use In e ace (UI).
The wo main so wa e modules o AppPe FallD a e:
• Moni o ingPe FallD: I includes a UI (Use In e ace), which is designed o olde people by
ollowing he elde ly- iendly design ideas om Ji e bug [107]. Thus, in o de o ease i s use,
he UI inco po a es a educed se o li key bu ons wi h clea op ions and no con using
menus. Th ee sc eensho s o his use in e ace a e illus a ed in Figu e 3.
Senso s 2014, 14 18561
• De ec ion Se ice: I is he moni o ing se ice ha implemen s he all de ec ion algo i hms.
To execu e hese algo i hms, he se ice is in cha ge o collec ing and eco ding he eadings
o he senso s. These eadings a e p ocessed basing on a powe -awa e s a egy.
• O he ou speci ic modules handle he es o he unc ionali ies o he applica ion: he
ansmission o all ale s, he sma phone connec i i y ( ia Wi-Fi o UMTS), he loca ion
p ocessing and he managemen o a SQLi e Da abase o s o e he moni o ed use loca ion.
Addi ionally, AppLoca ionInMaps is he applica ion de eloped o he emo e moni o ing poin .
I s goal is o ecei e, decode and p esen he in o ma ion con ained in he ale messages ha
AppPe FallD ansmi s when a all is de ec ed. Among o he unc ions, he applica ion displays he
pa ien ’s loca ion on a map downloaded om Google Maps web se ice.
7. E alua ion o he Sys em and De ec ion Algo i hms
The analysis ocuses on he pe o mance o he implemen ed all de ec ion algo i hms as well as on
he esou ce consump ion o he applica ion.
The algo i hms we e e alua ed by a se ies o me hodical expe imen s. Thus, a se o di e en
mo emen pa e ns (including alls) a e simula ed by 15 di e en olun ee s (six emales and nine
males, aged be ween 15 and 68 yea s and 150–190 cm all wi h an a e age weigh o 70 kg) in an
indoo en i onmen (a domes ic li ing oom). The subjec s emula ed he alls acco ding o h ee
di ec ions ( o wa d, la e al and backwa d), a di e en speeds and o e a pad o educe he impac . The
es o simula ed mo emen s consis o di e se “Ac i i ies o Daily Li ing” (ADL) such as jogging,
walking, s anding, si ing o answe ing he phone. Expe imen s we e epea ed by changing he posi ion
whe e pa icipan s placed he sma phone: a ached o he wais (by means o a bel ) o nex o he
high (wi hin a ouse pocke ). Each indi idual ca ied ou mo e han 50 mo emen s (comp ising a
leas 25 simula ed alls and 25 simula ed ADLs) o e e y algo i hm and e e y posi ion unde es .
In o de o e alua e he abili y o he algo i hms o disc imina e he all de ec ion pa e ns, we
compu ed he numbe o alse nega i es (i.e., hose alls ha emained unde ec ed) and alse posi i es
(i.e., hose ADL mo emen s ha we e inco ec ly iden i ied as alls and p o oked he ansmission o
an ale ). The es ima ion o he alse posi i es does no ake in o conside a ion he possibili y ha he
use can cancel he ale ing p ocess a e a all is de ec ed and he local acous ic ala m is igge ed in
he sma phone ( ha is o say: use -cancelled ale s a e also compu ed as alse posi i es). The selec ed
h esholds o he algo i hms we e also he same selec ed in he es s in es iga ed in [37].
Fo compa ison pu poses, we se all he h esholds and ime windows o he algo i hms o he same
alues u ilized in he bibliog aphy [37,43,106]. Fo he Pe FallD algo i hm we employed Th = 150,
Th = 6, Thc = 50, Th = 2, win = winc = win = winc = 4 s. Fo iFall, we se he lowe and he uppe
h esholds o 1G and 3.5G espec i ely ( his uppe h esholding limi is also chosen o he basic
algo i hm and o he Fall Index algo i hms). These se ings a e selec ed basing on aining da a and
aiming a minimizing he alse nega i es while educing he alse posi i es o a easonable minimum.
Table 2 p esen s he pe cen ages o alse nega i es ( a io be ween he numbe o alse nega i es and
he numbe o simula ed alls) and alse posi i es ( a io be ween he numbe o alse posi i es and he
numbe o ADL mo emen s) measu ed when he di e en algo i hms a e employed and he
Senso s 2014, 14 18562
sma phone is a ached o he wais . Fo compa ison pu poses, he able also inco po a es he esul s
ob ained wi h a comme cial speci ic de ice o all de ec ion in a e y simila es scena io in [37].
These esul s show ha Pe FallD and iFall algo i hms o e be e esul s han he basic
“ h esholding” me hods (such as he basic moni o ing o he accele a ion and he algo i hm ha is
p esumed o be used in he comme cial p oduc ). We hink ha his is due o he ac ha Pe FallD and
iFall algo i hms assume a mo e complex and ealis ic all pa e n wi h a leas wo phases and a ce ain
“obse a ion window”. This obse a ion window is also de ined by Fall Index (as long as i akes in o
accoun he e olu ion o las 20 samples o he accele a ion componen s). In ac , he Fall Index
algo i hm exhibi s ela i ely good esul s jus basing i s de ec ion decision on he e olu ion o he
changes in he global accele a ion du ing a small ime in e al. The algo i hms ha inco po a e a
longe analysis o he use ac i i y be o e a all is assumed also e eal a mo e homogeneous beha io
when he all pa e n (i.e., he all di ec ion) is modi ied. In his case, esul s (as hose ob ained by he
comme cial de ice) sugges ha he ypology o he es ed alls is a key aspec when assessing he
capabili y o he sys em o de ec he all e en . In any case he bene i s o using ce ain algo i hms a e
no as e iden as hose epo ed in o he s udies, such as [37].
Table 2. De ec ion pe o mance. Compa a i e be ween he di e en algo i hms.
Pe cen age o False Nega i es (%) Pe cen age o False Posi i es (%)
Fo wa d Falls La e al Falls Backwa d Falls O he Ac i i ies (ADL Mo emen s)
Basic moni o ing o he accele a ion [37] 8.0 28.3 5.5 14.6
Fall Index [37] 5.2 13.9 1.8 7.8
Pe FallD Algo i hm 4.5 8.9 14.9 5.9
iFall Algo i hm 8.0 16.0 12.0 10.1
B ickhouse comme cial p oduc [37] 0.8 1.2 29.9 21.9
Fo he case o Pe FallD algo i hm, Table 3 includes he compa ison o he measu emen s when he
posi ion o he sma phone is a ied. Excep o he case o o wa d alls, hese es s indica e ha a
be e pe o mance is achie ed i he sma phone is a ached o he wais . This can be explained by he
ac ha he acking o a poin nex o he wais can be e e lec he mo emen o he cen e o mass
o he body [108]. These esul s a e cohe en wi h he conclusions o he s udy in [61], which
compa ed he pe o mance o accele ome e -based all de ec ion sys ems when he accele ome e
(no in a sma phone) was al e na i ely loca ed on he w is , he wais o he head. In con as wi h
hose s udies ha ecommend a aching he sma phone o he ches [38], e gonomically he placemen
o he de ec ion de ice by he wais also in oduces less es ic ion on body mo emen and educes he
use ’s discom o [50]. Mo eo e , wais bel s a e no mally no conside ed as in asi e by he olde
people [41].
Table 3. Pe o mance o he Pe FallD algo i hm as a unc ion o he sma phone posi ion.
Pe cen age o False Nega i es (%) Pe cen age o False Posi i es (%)
Fo wa d Falls La e al Falls Backwa d Falls O he Ac i i ies (ADL Mo emen s)
Pe FallD
Algo i hm
Wais 4.5 8.9 14.9 5.9
Thigh 3.2 8.7 18.1 20.2
Senso s 2014, 14 18563
An impo an poin in he s udy o accele a ion-based echniques is a p ope selec ion o he
de ec ion h esholds. In mos s udies, he alues o hese h esholds a e heu is ically selec ed. In his
sense, a ade-o o simul aneously a oid alse posi i es (FN) and alse nega i es (FP) mus be
achie ed. To illus a e he impo ance o his ade-o , he sca e plo in Figu e 4 shows he
pe cen ages o alse nega i es and alse posi i es ( o he Pe FallD algo i hm) when one o he
employed h eshold is modi ied. These expe imen al Recei e Ope a ing Cha ac e is ic (ROC)
ype-cu es can be employed o se h eshold alues ha gua an ee he comp omise be ween low FN
and FP pe cen ages.
Figu e 4. Rela ionship be ween he pe cen ages o False Nega i es (FN) and
False Posi i es (FP) o di e en alues o he de ec ion h eshold
Th
when
c
Th
is se o a
ixed alue. Each poin in he g aph co esponds o he u iliza ion o a di e en alue o he
Th
h eshold.
Analysis o he Powe Consump ion
A c ucial aspec when e alua ing a sma phone o ien ed so wa e is powe consump ion. Complex
compu a ion o massi e ope a ion o he senso s may cause hea y ba e y consump ion and make
moni o ing applica ions i ually in easible om a p ac ical poin o iew.
In gene al, powe d ain in sma phones is highly dependen on se e al ea u es such as elec ical
and ne wo k se ing, use loca ion, signal powe , use ac i i y, phone u iliza ion, e c. To isola e he
impac o he all de ec ion app on he sma phone powe consump ion, we pe o m a se ies o es s in
which we compa e he ba e y discha ge o di e en ac i i y condi ions o he all de ec ion sys em.
In pa icula , o each es , he phone ba e y is ully cha ged and, a e ha , he powe s a e is
pe iodically moni o ed. No o he addi ional applica ion was execu ed in he phones du ing he
expe imen s. Th ee di e se scena ios a e conside ed:
• Scena io 1: AppPe FallD uns in passi e mode wi hou execu ing he de ec ion algo i hms.
Consequen ly, he mobili y o he sma phone does no a ec he consump ion.
• Scena io 2: AppPe FallD uns in ac i e mode, i.e., he all de ec ion algo i hms (in his case
Pe FallD) a e ac i a ed. This implies ha he accele a ion alues measu ed by he G-senso a e
con inuously p ocessed. In his case he sma phone is kep in a comple ely s a ic posi ion.
Senso s 2014, 14 18564
• Scena io 3: The applica ion is also ac i e, bu , in his case, he sma phone unde goes a pa e n
o pe iodical simula ed alls. Consequen ly, he co esponding ale ing SMS messages a e
ansmi ed o he emo e moni o ing poin . These SMSs in o m abou he posi ion o he use .
Thus, he GPS coo dina es need o be ob ained.
Figu es 5 and 6 show he e olu ion o he ba e y consump ion o he h ee scena ios when he
moni o ing applica ion is unning du ing 6 h on he HTC Desi e X and HTC sensa ion XE models,
espec i ely. Fo he measu emen s, he ba e y s a e was ob ained by he Diagnosis-Sys em
In o ma ion applica ion p o ided by he And oid Ope a ing Sys em.
Figu e 5. Es ima ion o ene gy consump ion in he HTC Desi e X phone.
Figu e 6. Es ima ion o ene gy consump ion in he HTC Sensa ion XE phone.
Fo bo h sma phone models, esul s indica e ha he moni o ing applica ion has a no negligible
epe cussion on powe consump ion (in bo h cases, he ba e y was exhaus ed be o e 40 h unde he
condi ions o he scena io 2). The g aph o he scena io 3 e idences ha consump ion can se e ely
inc ease i he applica ion u ilizes upda ed in o ma ion om he GPS.
Senso s 2014, 14 18565
8. Conclusions
Sma phone-based a chi ec u es o pe asi e all de ec ion can clea ly bene i om he massi e
social accep a ion and widesp ead ex ension o sma phones. These de ices, which na i ely in eg a e
accele ome e s, gy oscopes and di e se communica ion in e aces (Wi-Fi, Blue oo h, 3G and beyond
da a connec ions), p o ide a cos -e ec i e and e icien solu ion o he deploymen o wea able
sys ems o all de ec ion and ale ing.
This pape has p esen ed a p o o ype o a all de ec ion sys em based on And oid applica ions o
mobile phone pla o ms. The sys em is in cha ge o sending a message o au oma ically es ablishing a
phone call whene e a all is p esumed.
Mos wo ks in he li e a u e abou sma phone-based all de ec ion a chi ec u es base he
iden i ica ion o all pa e ns on he analysis o he da a epo ed by he buil -in sma phone
accele ome e (in some cases, combined wi h he in o ma ion o he phone o ien a ion). Al hough he e
a e solu ions ha employ ained AI sys ems o disc imina e he alls om he con en ional physical
ac i i y o he use s, he ha dwa e limi a ions o he memo y and eal- ime p ocessing capabili ies o
he sma phones ecommend implemen ing less sophis ica ed de ec ion p ocedu es. In his sense, he
majo i y o he p oposals apply simple h eshold-based echniques o p ocess he sequence o
accele a ion da a. In his wo k, he de eloped p o o ype, es ed wi h di e en sma phone models, was
aimed a e alua ing di e en exis ing algo i hms ha u ilize h eshold compa ison me hods o iden i y
he alls. Fo his goal, a wide se o expe imen s execu ed by 15 olun ee s we e conduc ed.
Expe imen s included a mix u e o simula ed alls and con en ional mo emen s. In con as wi h he
conclusions o o he s udies (whe e a new algo i hm is p oposed), he ob ained esul s e lec he
di icul y o de e mining an op imal s a egy o de ec alls. Fo example, no algo i hm achie es an
e iciency highe han 95% o 90% o a oid alse posi i es and alse nega i es, espec i ely. In an
ac ual applica ion en i onmen , his could imply ha many alls could be unno iced while many
mo emen s ela ed o egula ac i i ies could p o oke ala ms ha should be manually deac i a ed by
he use be o e an ale message is sen o a emo e moni o ing poin . The s ong dependence o he
measu ed pe o mance on he ypology (i.e., di ec ion) o he alls indica es ha any all de ec ion
sys em mus be e alua ed h ough an exhaus i e es -plan wi h a high di e si y o mo emen pa e ns.
The s udy o he ade-o be ween “ alse posi i es” and “ alse nega i es” also e eals he impo ance
o he selec ed h esholds, which comple ely go e n he accu acy o he de ec ion p ocess. In addi ion,
he limi a ions in oduced by he ba e y li e ime may become a ema kable elemen o de e mine he
iabili y o his ype o all de ec ion sys ems in a eal applica ion en i onmen whe e a use should be
pe manen ly (24 h a day) elemoni o ed. The cons an use o he accele ome e and (i needed) he
GPS senso by he de ec ion algo i hms undoub edly educes he au onomy and applicabili y o
sma phone-based a chi ec u es (in ou expe imen s, less han 40 h o con inuous moni o ing we e
accomplished). Consequen ly all de ec ion And oid applica ions mus be ca e ully designed o
op imize he access o he employed senso s and o minimize powe consump ion.
E gonomics and usabili y a e o he wo key aspec s o he ac ual adop ion o his ype o
echnology (especially among he olde people, who a e he main a ge o hese sys ems). In his
sense, he need o equen in e ac ion o a no -expe use (ba e y cha ging, cancella ion o alse
Senso s 2014, 14 18566
ala ms, p og amming o de ec ion h esholds, complex aining phases o cha ac e ize he use ’s
ac i i y pa e ns, e c.) may no iceably hinde he accep ance o hese elemoni o ing se ices.
Acknowledgmen s
This wo k has been suppo ed by Eu opean FEDER unds and he Spanish Minis y o Economy
and Compe i i eness (g an TEC2009-13763-C02-01).
Au ho Con ibu ions
R.L., E.C. and M.J.M. concei ed and designed he expe imen s; R.L. and M.J.M. p og ammed he
And oid applica ion, E.C. w o e he pape and elabo a ed he s a e-o - he-a . R.L., M.J.M. and G.R.
pe o med and o ganized he execu ion o he expe imen s;
Con lic s o In e es s
The au ho s decla e no con lic s o in e es .
Re e ences
1. Wo ld Heal h O ganiza ion. Ageing Li e Cou se Uni . In WHO Global Repo on Falls
P e en ion in Olde Age; Wo ld Heal h O ganiza ion: Gene a, Swi ze land, 2008.
2. Yoshida, S. A Global Repo on Falls P e en ion Epidemiology o Falls; Wo ld Heal h
O ganiza ion: Gene a, Swi ze land, 2007.
3. Ho nb ook, M.C.; S e ens, V.J.; Wing ield, D.J.; Hollis, J.F.; G eenlick, M.R.; O y, M.G.
P e en ing Falls among Communi y-Dwelling Olde Pe sons: Resul s om a Randomized T ial.
Ge on ologis 1994, 34, 16–23.
4. Ka lsson, M.K.; Vonschewelo , T.; Ka lsson, C.; Cos e , M.; Rosengen, B.E. P e en ion o Falls
in he Elde ly: A Re iew. Scand. J. Public Heal h 2013, 41, 442–454.
5. Nou y, N.; Rumeau, P.; Bou ke, A.; ÓLaighin, G.; Lundy, J. A P oposal o he Classi ica ion
and E alua ion o Fall De ec o s. I bm 2008, 29, 340–349.
6. Pe y, J.T.; Kellog, S.; Vaidya, S.M.; Youn, J.; Ali, H.; Sha i , H. Su ey and E alua ion o
Real-Time Fall De ec ion App oaches. In P oceedings o he 6 h In e na ional Symposium on
High-Capaci y Op ical Ne wo ks and Enabling Technologies (HONET 2009), Alexand ia, Egyp ,
28–30 Decembe 2009; pp. 158–164.
7. Igual, R.; Med ano, C.; Plaza, I. Challenges, Issues and T ends in Fall De ec ion Sys ems.
Biomed. Eng. Online 2013, 12, doi:10.1186/1475-925X-12-66.
8. Schwicke , L.; Becke , C.; Lindemann, U.; Ma échal, C.; Bou ke, A.; Chia i, L.; Helbos ad, J.;
Zijls a, W.; Aminian, K.; Todd, C. Fall De ec ion wi h Body-Wo n Senso s. Zei sch i Fü
Ge on ologie. Und Ge ia ie. 2013, 46, 706–719.
9. Mubashi , M.; Shao, L.; Seed, L. A Su ey on Fall De ec ion: P inciples and App oaches.
Neu ocompu ing 2013, 100, 144–152.
Senso s 2014, 14 18567
10. Ande son, D.; Luke, R.H.; Kelle , J.M.; Skubic, M.; Ran z, M.; Aud, M. Linguis ic
Summa iza ion o Video o Fall De ec ion Using Voxel Pe son and Fuzzy Logic. Compu . Vis.
Image Unde s . 2009, 113, 80–89.
11. Cucchia a, R.; P a i, A.; Vezzani, R. A Mul i-Came a Vision Sys em o Fall De ec ion and
Ala m Gene a ion. Expe Sys . 2007, 24, 334–345.
12. Di aco, G.; Leone, A.; Siciliano, P. An Ac i e Vision Sys em o Fall De ec ion and Pos u e
Recogni ion in Elde ly Heal hca e. In P oceedings o he Design, Au oma ion & Tes in Eu ope
Con e ence & Exhibi ion (DATE 2010), D esden, Ge many, 8–12 Ma ch 2010; pp. 1536–1541.
13. Fo oughi, H.; Nase i, A.; Sabe i, A.; Yazdi, H.S. An Eigenspace-Based App oach o Human
Fall De ec ion using In eg a ed Time Mo ion Image and Neu al Ne wo k. In P oceedings o he
9 h In e na ional Con e ence on Signal P ocessing (ICSP 2008), Beijing, China, 26–29 Oc obe
2008; pp. 1499–1503.
14. Fu, Z.; Culu ciello, E.; Lich s eine , P.; Delb uck, T. Fall De ec ion using an Add ess-E en
Tempo al Con as Vision Senso . In P oceedings o he IEEE In e na ional Symposium on
Ci cui s and Sys ems (ISCAS 2008), Sea le, WA, USA, 18–21 May 2008; pp. 424–427.
15. Hazelho , L.; Han, J. Video-Based Fall De ec ion in he Home using P incipal Componen
Analysis. In P oceedings o he 10 h In e na ional Con e ence on Ad anced Concep s o
In elligen Vision Sys ems, Juan-les-Pins, F ance, 20–24 Oc obe 2008; pp. 298–309.
16. Jansen, B.; Dekle ck, R. Con ex Awa e Inac i i y Recogni ion o Visual Fall De ec ion.
In P oceedings o he 1s In e nacional Con e ence on Pe asi e Compu ing Technologies o
Heal hca e, Innsb uck, Aus ia, 2006.
17. Lee, T.; Mihailidis, A. An In elligen Eme gency Response Sys em: P elimina y De elopmen
and Tes ing o Au oma ed Fall De ec ion. J. Telemed. Teleca e 2005, 11, 194–198.
18. Liu, C.; Lee, C.; Lin, P. A Fall De ec ion Sys em using k-Nea es Neighbo Classi ie . Expe
Sys . Appl. 2010, 37, 7174–7181.
19. Miaou, S.; Sung, P.; Huang, C. A Cus omized Human Fall De ec ion Sys em using Omni-Came a
Images and Pe sonal In o ma ion. In P oceedings o he 1s T ansdisciplina y Con e ence on
Dis ibu ed Diagnosis and Home Heal hca e (D2H2), A ling on, VA, USA, 2–4 Ap il 2006;
pp. 39–42.
20. Ni, B.; Da , N.C.; Moulin, P. RGBD-Came a Based Ge -Up E en De ec ion o Hospi al Fall
P e en ion. In P oceedings o he IEEE In e na ional Con e ence on Acous ics, Speech and
Signal P ocessing (ICASSP 2012), Kyo o, Japan, 25–30 Ma ch 2012; pp. 1405–1408.
21. Rougie , C.; Meunie , J.; S -A naud, A.; Rousseau, J. Robus Video Su eillance o Fall
De ec ion Based on Human Shape De o ma ion. IEEE T ans. Ci cui s Sys . Video Technol. 2011,
21, 611–622.
22. Sixsmi h, A.; Johnson, N. A Sma Senso o De ec he Falls o he Elde ly. Pe asi e Compu .
IEEE 2004, 3, 42–47.
23. Vishwaka ma, V.; Mandal, C.; Su al, S. Au oma ic de ec ion o human all in ideo. In Pa e n
Recogni ion and Machine In elligence; Ghosh, A., De, R.K., Pal, S.K., Eds.; Sp inge : Be lin,
Ge many, 2007; pp. 616–623.
Senso s 2014, 14 18574
106. Yoshida, T.; Mizuno, F.; Hayasaka, T.; Tsubo a, K.; Wada, S.; Yamaguchi, T. A Wea able
Compu e Sys em o a De ec ion and P e en ion o Elde ly Use s om Falling. In P oceedings
o he 12 h In e na ional Con e ence on Biomedical and Medical Enginee ing (ICBME),
Singapo e, Singapo e, 7–10 Decembe 2005.
107. Ji e bu g Cell Phones o Senso s, 2014. A ailable Online: h p://www.ji e bugdi ec .com/
(accessed on 22 Sep embe 2014).
108. Yang, C.; Hsu, Y. A Re iew o Accele ome y-Based Wea able Mo ion De ec o s o Physical
Ac i i y Moni o ing. Senso s 2010, 10, 7772–7788.
© 2014 by he au ho s; licensee MDPI, Basel, Swi ze land. This a icle is an open access a icle
dis ibu ed unde he e ms and condi ions o he C ea i e Commons A ibu ion license
(h p://c ea i ecommons.o g/licenses/by/4.0/).
