See discussions, s a s, and au ho p o iles o his publica ion a : h ps://www. esea chga e.ne /publica ion/318100909
Exis ing App oaches o Sma Pa king: An O e iew
Con e ence Pape · May 2017
DOI: 10.1007/978-3-319-59513-9_7
CITATIONS
8
READS
289
5 au ho s, including:
Some o he au ho s o his publica ion a e also wo king on hese ela ed p ojec s:
Au oma ic dia om iden i ica ion o wa e quali y assessmen and moni o ing View p ojec
Face and acial elemen s de ec ion View p ojec
Fe nando En iquez
Uni e sidad de Se illa
43 PUBLICATIONS297 CITATIONS
SEE PROFILE
Luis Miguel So ia Mo illo
Uni e sidad de Se illa
32 PUBLICATIONS382 CITATIONS
SEE PROFILE
Juan An onio Al a ez-Ga cia
Uni e sidad de Se illa
75 PUBLICATIONS630 CITATIONS
SEE PROFILE
F ancisco Velasco-Mo en e
Uni e sidad de Se illa
129 PUBLICATIONS794 CITATIONS
SEE PROFILE
All con en ollowing his page was uploaded by Juan An onio Al a ez-Ga cia on 27 Ap il 2021.
The use has eques ed enhancemen o he downloaded ile.
Exis ing app oaches o sma pa king: An o e iew
Fe nando En ´
ıquez1, Luis Miguel So ia1, Juan An onio ´
Al a ez-Ga c´
ıa1, F ancisco
Velasco2, and Osca D´
eniz3
1Compu e Languages and Sys ems Depa men , Uni e si y o Se ille, 41012 Se ille, Spain
[email p o ec ed],[email p o ec ed],[email p o ec ed]
2Applied Economics I Depa men , Uni e si y o Se ille, 41018 Se ille, Spain
[email p o ec ed]
3VISILAB, E.T.S.I.I, Uni e si y o Cas illa-La Mancha, Ciudad Real, Spain
[email p o ec ed]
Abs ac . A e yea s o echnological ad ances, pa king is s ill a p oblem o
many people, as a ime-consuming ask, hey ha e o ace on a day-by-day basis,
and also o ci ies, ha see how a ic and pollu ion inc eases. The e ha e been
mul iple a emp s o ind a pa ial o global echnological solu ion o his p oblem,
anging om using di e en ypes o senso s o cams o au oma ically de ec ing
ee spaces o collabo a i e apps ha le use s sha e ela ed in o ma ion. In his
pape , we gi e an o e iew o he me hods ha ha e been de eloped so a , show-
ing hei main ea u es, di e ences, p os, and cons, as well as o he ac o s ha
may con ibu e o he success o ailu e o new p oposals ha will come in he
u u e.
Keywo ds: Sma ci y, Pa king, C owdsensing, Compu e ision
1 In oduc ion
Acco ding o he Uni ed Na ions [1], 54% o he popula ion li es in u ban zones and his
is expec ed o g ow o 66% in 2050. This endency makes u ban mobili y mo e di icul
and makes inding a pa king space one o he mos epe i i e p oblems which ci izens
in big ci ies ha e o ace up. Fu he mo e, ehicles c uising o pa king a e esponsible
o a leas 30% o a ic jams [2], wi h he a e age ime o sea ch being mo e han 20
minu es [3]. Besides he pe sonal p oblems ha his can gene a e, his is a p oblem in
e ms o uel consump ion, CO2 emissions and in gene al a was e o esou ces o he
communi y.
Beyond p i a e spaces, he e is no wo ldwide accep ed solu ion o moni o ing e-
hicles ha en e o lea e an a ea, hough se e al e o s ha e been made o sol e he
p oblem in a speci ic con ex . Sma pa king is he e m used o a se o echnologies
and applica ions a ge ing issues ela ed o pa king in Sma Ci ies and his wo k aims
a p o iding an up- o-da e su ey o he mos in e es ing and ele an solu ions. The
es o he pape is o ganized as ollows. Sec ion 2 analyses he s a e o he a in Sma
Pa king solu ions. The main conclusions a e d awn in Sec ion 3.
2
2 Classi ica ion
Pa king solu ions a e o e ed o o -s ee o on-s ee . O -s ee pa king may e e o
mul is o y ca pa ks, while on-s ee e e s o pa king spaces along public oads and
s ee s. This wo k conside s bo h ypes o scena ios. Typically he e is a ocus on in-
as uc u e, whe e se e al senso s a e ins alled o moni o he places, c owdsensing
whe e mobile phones’ d i e s a e he sou ce o he in o ma ion o a hyb id p oposal.
He e, ision solu ions a e analyzed as a speci ic ca ego y due o he inc ease o came as
and he po en ial o his ype o senso .
2.1 In aes uc u e
In o de o ge he occupancy s a us o pa king places, ixed o mobile senso s a e in-
s alled in on/o -s ee pa king o de ec ehicula e en s. Mobile senso s a e no he
mos common solu ion bu Pa kne [4] in San F ancisco is a mos p ominen s udy: axi
ca s collec ed he occupancy s a us o he pa king place when hey passed beside i ,
ga he ing da a om GPS ecei e and ul asonic senso s. Al hough e e y axi can de-
ec mul iple spo s, upda ing he in o ma ion o he same spo can ake 25 minu es wi h
a lee o 300 ehicles. Pa king Spo e is based on he same idea, le e aging sona and
ada senso s o some Fo d ehicles. The e a e also some LiDAR-based solu ions [5]
bu only ocused on su eying pa king spaces wi h one equipped ca . These wo ks a e
ha dly scalable since all he ehicles mus ha e he same kind o senso o else sha e
he in o ma ion h ough he same da abase and he numbe o moun ed senso - ehicles
mus be enough o upda e he in o ma ion equen ly.
Fixed senso s a e he mo e ex ended and popula op ion. Al hough he e a e a wide
a ie y o senso s [6] (ac i e in a ed, ul asound, acous ic, accele ome e , e c.), magne-
ome e is by a he mos common ixed senso . The magne ome e is accu a e al hough
no mally solu ions equi e a leas one senso pe place, inc easing he cos o he de-
ploymen . The senso i sel measu es he cu en magne ic ields and de ec s he a i al
o me al ehicles. Mos municipal deploymen p ojec s o la ge shopping cen e s d ill
magne ome e s on pa king places, sha ing his in o ma ion h ough isual signals o
mobile applica ions. Ins alla ion and main enance p ocesses in ol e access o he p op-
e y and oad su ace so i is only possible o o -s ee o on-s ee wi h go e nmen
pe mission.
Table 1 e e s o he mos p ominen sma ci y pa king solu ions. Sma San ande
and San F ancisco can be seen as pilo s udies. Sma San ande was concei ed as a
Sma Ci y Labo a o y and San F ancisco inished i s pilo s udy a he end o 2013 due
o he cos o senso main enance. Nice wi hd ew mobile applica ion Nice Passpo in
2016 a e some o ganiza ional4and secu i y p oblems5. M´
alaga, London ( he only one
based on RFID), Moscow o Los Angeles a e success cases o Sma Pa king.
4h p://www.20minu es. /nice/1839579-20160504-nice- ois-ans-ap es-ins alla ion- s a ionne-
men -in elligen -dispa ai
5h p://www.le iga o. /poli ique/le-scan/couacs/2014/06/06/25005-20140606 ARTFIG00112-
secu i e-in o ma ique-a-nice-p is-de-cou -es osi-in e omp -une-in e iew.php
3
Fixed senso -based solu ions esea ch is ocused on educing he ins alla ion ime
and cos [7] using su ace-moun ed magne ome e s ha can be glued o he oad and
enla ging ba e y li e o wi eless senso s.
Sma Ci y #places Company URL Yea
San F ancisco 68.2K Fyb ech h p://www. yb - ech.com 2011
San ande [8] 0.4K Libelium h p://bi .ly/2mOd38 2011
Nice 74.5K U bio ica h p://www.u bio ica.com 2012
Los Angeles [9] 6.3K S ee Line h p://la .ms/1BVDxpD 2012
London 83.4K Sma Pa king h p://bi .ly/2mTSTcm 2012
Moscow 950k Wo ldSensing h p://bi .ly/2lCMAbU 2012
Malaga 10 2.2K Pa khelp h p://bi .ly/2mTPNFc 2014
Table 1. In as uc u e pa king p ojec s
2.2 Vision
Due o he inc easing in e es o he scien i ic communi y and use s in gene al by he
echniques o a i icial in elligence based on images, in ecen yea s he e ha e p oli -
e a ed solu ions o he p oblem o pa king con ol h ough sys ems based on ision.
These sys ems, unlike p e ious in as uc u es, a e no ye es ablished, bu a e mos ly
used in con olled and expe imen al en i onmen s. Al hough he e a e companies ha
base hei sys ems on hese echniques, mos o he wo k is s ill unde de elopmen .
In gene al, he echniques o de ec ion o pa king spaces based on ision, su e
om se e al p oblems. The i s one is he quali y o he image. Fo ce ain wo ks based
on objec ecogni ion, he image mus be o su icien quali y o be p ocessed. Du ing
he day, his may no be an obs acle, bu in condi ions o insu icien ligh ing o ad e se
wea he condi ions, i becomes a eal p oblem. Ano he p oblem ela ed o ob aining he
image a e occlusions. Depending on he loca ion o he image cap u e sys em, he e-
hicles hemsel es o su ounding elemen s ( ees, buildings, s ee u ni u e, shadows)
may obs uc he iew o he pa king a ea. This p oblem can be sol ed by changing
he placemen o he cap u e sys em, bu his may no always be possible, o educe he
moni o ed su ace due o he change in pe spec i e. Finally, an inhe en p oblem wi h
his de ec ion echnique is he classi ica ion i sel . Image-based classi ica ion sys ems
ha e p oli e a ed o e he pas decade, bu hey a e s ill a om o e ing he assu ance
o sys ems based on s uc u al elemen s.
6h p://s pa k.o g/ esou ces/pa king-senso - echnology-pe o mance-e alua ion
7h p://www.u bio ica.com/en/inaugu a ion-o -ou -sma -pa king-p ojec -in- he-ci y-o -nice-
/
8h ps://www.wes mins e .go .uk/pa k igh
9h p://pa king.mos. u/en
10 h p://www.eesc.eu opa.eu/?i=po al.en.e en s-and-ac i i ies-sma -ci ies-malaga
4
Howe e , in a o o sys ems o de ec ing pa king spaces using ision algo i hms,
he e is he co e age, cos , and e sa ili y o he sys ems. Since a single cap u e sys-
em can co e dozens (o hund eds) o pa king spaces, he cos s educ ion compa ed o
in as uc u e-based sys ems is signi ican . In addi ion, he main enance cos o hese
sys ems is negligible, excep in hose cases whe e d ones o sa elli es a e equi ed
o acqui e he images. The la e a e usually mul i- unc ional, which leads o ano he
s eng h: e sa ili y. Because he images ob ained can be applied no only o he pa k-
ing con ol, sys ems implemen ing he ision-based solu ion a e o en used o o he
pu poses a he same ime. Examples o his unc ionali y a e su eillance sys ems,
pedes ian con ol, main enance asks, and o he scena ios ha make he in es men
made in he implemen a ion o he image cap u e sys ems o be quickly amo ized.
The pa king lo s de ec ion solu ions based on ision can use di e en sys ems o ge
he images o he zone unde con ol. Among o he , he mos common me hods o he
image acquisi ion a e ex e nal ideo came as, ehicles equipped wi h ision sys ems,
h ee-dimensional cap u e sys ems, and zeni hal o ae ial images ob ained by sa elli es
o d ones.
In he same way, as discussed abo e, de ec ion sys ems a e suppo ed by a se o
algo i hms. These algo i hms can be g ouped by he ea men pe o med o e he im-
ages. Mos wo ks in he s a e o he a base hei de elopmen on some o he ollowing
app oaches:
–Appea ance based app oaches. Based on he compa ison o he cu en appea ance
o a pa king place wi h an o iginal appea ance o he acan s a e. Many echniques
a e ailo ed and ine- uned o speci ic con ex s and scena ios. Howe e , hese ech-
niques can no be easily gene alized, and e en he adap a ion o one solu ion o a
di e en pa king lo is no s aigh o wa d.
–Recogni ion based app oaches. App oaches based on objec ecogni ion aim o de-
ec and classi y he ehicles occupying he pa king space using machine lea ning
algo i hms. Complex app oaches because o he la ge a ie y o he a ge objec s.
–Th ee dimensional image p ocessing.
–Combined echniques applying image p ocessing in o de o imp o e he quali y
and a oid ligh a ia ion e ec s, and machine lea ning algo i hms o classi y image
con en .
Finally, moni o ing echniques can be di ided in o wo ypes depending on he way
in which lo s a e p ocessed:
–Es ima ing occupancy o an en i e pa king lo , o example, by coun ing incoming
ehicles.
–Checking o he p esence o a ehicle in each cell. Mos ision-based app oaches
equi e he p esence o ehicles in indi idual pa king lo s.
Below, a compila ion o some ele an pape s om he ela ed bibliog aphy a e
shown. Among hem, he e a e di e en app oaches as a sample o he he e ogenei y o
he exis ing p ocessing, ecogni ion, and image acquisi ion sys ems.
In [10] is p oposed a dis ibu ed and e icien sys em o sol e he p oblem o pa k-
ing h ough ision sys ems. Fo his pu pose con olu ional neu al ne wo ks speci ically

5
designed o sma came as a e used. Two isual da ase s ha e been used o e i y
he accu acy o he p oposed sys em: PKLo and CNRPa k-EXT. This las da ase was
c ea ed by he own au ho s. Thanks o con olu ional neu al ne wo ks applied in he
classi ica ion p ocess, he p oposed solu ion is obus o images exposed o pa ial oc-
clusion, shadows, and changes in ligh condi ions. In addi ion, i has a good capaci y
o gene aliza ion. A educed e sion o he AlexNe neu al ne wo k we e used in his
pape . The new con olu ional neu al ne wo k is able o ecognize only wo classes: ee
o occupied pa king space.
Au ho s in [11] implemen s a pa king de ec ion sys em based on eal- ime image
p ocessing om ideo came as. I di ides he sys em in o h ee sec ions: image acqui-
si ion module, image p e-p ocessing module and image de ec ion module. In he i s
one, he image is il e ed. The de ec ion module is based on he use o a e e ence image
om an emp y pa king space, wi hou any in e e ence. F om his image, con e ed o
g ayscale, a compa ison sys em is es ablished wi h successi e images. The algo i hm
ob ains he edges o he e e ence image and compa es hem wi h he las cap u ed. Fi-
nally, a unc ion is applied o decide whe he he e e ence image and he compa ed one
ha e simila cha ac e is ics. The accu acy o he sys em is ai (81%).
The solu ion p oposed in [12] uses indi idual images om a single su eillance
came a p e iously ins alled in indoo pa king lo s. This wo k is also based on e e ence
images om he emp y pa king lo . The p ocess is sligh ly di e en om [11]. The
model is gene a ed using P incipal Componen Analysis, suppo ing ligh ing in a iance
o a oid ligh ing changes we e ecognized as po en ial objec s. In addi ion o he sys em
adap a ion o ligh ing changes, ex u es a e also used o de ec objec s and isola e hem
om he backg ound. This makes he sys em much mo e obus unde ex eme ligh ing
changes (la ge ees, s ee lamps o in ense shadows). In a eas wi h high isibili y, he
numbe o occlusions be ween ehicles is minimized and, hence, sys ems use o be
mo e accu a e. This wo k has 90% o accu acy, al hough i can dec ease i occlusions
occu in he image.
In [13], he classi ica ion is done by a 3-laye Bayesian hie a chical de ec ion F ame-
wo k (BHDF). To a oid occlusion p oblems wi h o he ca s and su ounding objec s,
his pape uses he ull image o p ocessing, a he han de e mining he s a e o he
pa king spaces one by one. To analyze he whole pa king space, he scene is di ided in o
3D cubes. Each cube co esponds o a ow o ca lo s. Once he image is ob ained and
he ow o pa king spaces o be p ocessed is de ined, he BHDF amewo k is applied.
This amewo k is composed o h ee p e- ained models, one o he local classi ica ion
(obse a ion laye ), o he wi h he adjacency model (labeling laye ) and, he las , wi h
he seman ic laye . The local classi ica ion model (inpu model) can be pixel-based o
ex u e-based.
The wo k p esen ed in [14] deals wi h he p oblem o he sunligh , he da k shadows
du ing he day, and he low ligh in ensi y a nigh ime. To do his, Pa kLo D uses a clas-
si ie based on he uzzy c-means clus e ing algo i hm (FCM) and a hype -pa ame ic
i wi h pa icle swa m op imiza ion (PSO). The algo i hm has been es ed du ing di -
e en hou s o he day and unde di e en wea he condi ions. Howe e , he nigh ime
use wi h poo ligh ing condi ions is no ad ised. A he beginning o he p ocess, i is
necessa y o de ine he limi s o each pa king space, as well as he o al numbe o lo s.
6
In he nex s age, PCA is applied om he ea u e ec o s ob ained om each image.
Thanks o a use in e ace, he ope a o can co ec he algo i hm ope a ion and show
i any e o occu s. Fu he mo e, he ope a o can e ain he sys em when needed. The
e alua ion p ocess was ca ied ou om 2000 images cap u ed by a came a on a oo .
Each image co e ed 27 pa king spaces.
In [15], pa king s a es a e de e mined by he combina ion o an adap i e back-
g ound sub ac ion algo i hm o mo ing objec s de ec ion (o e coming p oblems o
ligh changes and shadow e ec s), wi h speeded up obus ea u es (SURF) algo i hm
( obus o scale and o a ion changes). Au ho s p opose a solu ion based on he image
appea ance and he compa ison wi h a e e ence image whe e he lo s appea acan .
By using a homogenous ans o ma ion o change he poin o iew, acili a es he ex-
ac ion o he pa king model elimina ing pe spec i e dis o ion. Fo his p ocess, he
use mus p e iously selec a leas ou poin s ha de ine he pa king a ea. Once he
ans o ma ion is comple ed, he pa king lo s a e de ined. Fo he adap i e backg ound
sub ac ion, he Mix u e Gaussian Model algo i hm is used. This equi es a lea ning
p ocess o upda ing he dis ibu ions ob ained by he model h ough a se o images.
Once he backg ound is ex ac ed, he cha ac e is ics o he images o be classi ied a e
ob ained. Fo his p ocess, he algo i hms o Speeded Up Robus Fea u es Speeded (in-
a ian o scale changes) and His og am o O ien ed G adien s (HOG) a e p oposed.
The algo i hms SVM o KNN can be combined du ing he classi ica ion s age. The au-
ho s, h ough hei e alua ion wi h he VI-RAT ideo da abase, conside ha he bes
combina ion o p oposed solu ions is SURF +SVM, ob aining an a e age p ecision
o 93%. Howe e , he main p oblem o his me hod is he lack o obus ness o pa ial
occlusions.
In [16], he aim is no ocused on pa king a eas ecogni ion, bu on de ec ing ca s
and dis ances by means o ehicles equipped wi h a s e eo ision sys em. The au ho s
iden i y h ee me hods o pe o m his de ec ion: using ada , using ac i e senso s (lase ,
ada , lida ) and monocula ision, and inally, using only ision. 2D-based sys ems may
no be enough accu a e in iden i ica ion o he ehicles. To sol e his d awback, his sys-
em p ocess he 3D image iden i ying ce ain po en ially cha ac e is ics belonging o a
ehicle ( e ical edges). The algo i hm is di ided in o wo main s ages: he ex ac ion o
h ee-dimensional cha ac e is ics and he de ec ion o he ehicle om hem. Ex ac ion
o e ical cha ac e is ics allows isola ing mo e p ecisely he obs acles (and ehicles) o
he own highway. Two came as a e used o gene a es 3-D spa se map. As a o emen-
ioned, al hough his kind o algo i hm is no speci ic o pa king lo s ecogni ion, hey
can be applied in his con ex hanks o he capaci ies o ecognize ehicles.
Finally, able 2 shows a schema ic ep esen a ion o some o he mos ele an wo ks
in pa king slo s ecogni ion. In able a e collec ed a ious pa ame e s ep esen a i e o
he di e en solu ions p oposed. These pa ame e s include ha dwa e equi ed, algo-
i hm e o a e, obus ness o ligh ing changes (low , medium W, high ), au oma ion
le el o he ecogni ion and lea ning p ocess (manual iden i ica ion o pa king lo s o
aining p ocess , semiau oma ic sys em allowing manual changes in he aining p o-
cess W, ully-au oma ed sys em ), obus ness o pe spec i e changes (low , medium
W, high ), obus ness o pa ial occlusions (low , medium W, high ), p e e ed en i-
onmen o he sys em use (indoo , ou doo , bo h), deploymen cos s (low , medium
7
i came as a e al eady ins alled W, complex in as uc u e could be equi ed [sa elli es,
d ones, ul asonic senso s] ), compu a ional equi emen s (low [could be high du ing
he aining s age] , mediumW, high ), scalabili y (low , mediumW, high ) and,
inally, he way in which he ca pa ks a e p ocessed (indi idually using masks o col-
lec i ely o e he whole a ea).
Au ho Yea E o a e Ha dwa e equi ed
Ligh ing
Lea ning
Pe spec i e
Occlusions
Co e age
Economic
Compu a ional
Scalabili y
G ouping
G. Ama o e al. [10] 2017 3% Sma -came a  W   Bo h  W  Masks
G. Ama o e al. [17] 2016 9% Sma -came a +RPi  W   Bo h  W  Masks
I. Masmoudi e al. [15] 2016 7% Videocame a W W W  Ou doo W W W Masks
J. Suh e al. [18] 2014 1.90% AVM and ul asound     Bo h    Vehicle
J. Je msu awong e al. [19] 2012 1% Videocame a  W   Ou doo W W  Full a ea
H. Ichihashi e al. [14] 2009 7% Videocame a +se e W W   Bo h W W  Masks
Z. Bin e al. [11] 2009 19% Video came a +CPU     Bo h W  W Masks
C. Huang e al. [13] 2008 6% IP came a +CPU W W   Indoo W W W Rows
G. Toulmine e al. [16] 2006 N/D S e eo- ision sys em +CPU     Ou doo W W  Vehicle
Q. Zhang e al. [20] 2006 20% Sa elli e images     Ou doo    Full a ea
S. Funck e al. [12] 2004 10% CCTV-came as W W   Indoo W  W Masks
K. Yamada e al. [21] 2001 0.60% Videocame a W W   Ou doo W  W Masks
X. Wang e al. [22] 1998 25% Ae ial images  W   Ou doo    Full a ea
Table 2. Vision-based pa king lo de ec ion algo i hms
2.3 Social c owdsensing
Using so wa e applica ions o help d i e s pa k hei ehicles is no new, e en wi h
solu ions ha include he loca ion, ese a ion, access and paymen asks all oge he
as in [23]. Ne e heless, he g owing popula i y o sma phones in he las yea s, ull
o senso s and capable o egis e ing geo-posi ional in o ma ion in sho pe iods o
ime, has c ea ed new ways o sea ching o a pa king spo a o ing he ise o mul iple
mobile and web applica ions. Table 3 shows a compa ison be ween some o he apps
o on-s ee o o -s ee s pa king we can ind on he In e ne . The selec ion c i e ia
used o il e he huge ca alog ha exis s nowadays is based on he popula i y, bu also
on he special ea u es o e ed by some o hem. Mos o he apps a e cen e ed uniquely
on o -s ee pa king, maybe collec ing da a om ins alled senso s o cams. Pa kopedia
(a ailable wo ldwide) o wesma Pa k (wo king in Mad id and Ba celona), a e di e en
examples o hese o -s ee pa king solu ions. The la e o e s he ins alla ion o senso
echnology and a managemen module o owne s o easily moni o hei business while
hey ob ain da a o hei app use s. While he basic ope a ions in hese apps emain
he same, we ind in e es ing p oposals in he li e a u e ha could change he scene
someday, i.e. b inging au oma ic p ice nego ia ion capabili ies [24]. The e a e also apps
ha gi e in o ma ion o on-s ee pa king, ecommending he bes zones o sea ch o
unoccupied pa king spo s, o gi ing he oppo uni y o announce when a use is abou
o mo e a ehicle, lea ing a ee space, and in some cases, allowing ese a ions o his
8
new ee spo s. Each o hese unc ionali ies can use da a ob ained au oma ically by cell
phones (c owdsensing) o in oduced manually by he app use s (c owdsou cing). This
app oach, au oma ic o manual, is pa icula ly use ul o on-s ee pa king, whe e he
scalabili y o o he app oaches is mo e di icul due o he size o he a eas o be co e ed.
While o -s ee pa king in o ma ion can p ima ily help use s educe he p ice o pay o
selec he closes space o hei des ina ion, on-s ee pa king in o ma ion usually has
as i s main goal he mi iga ion o he ”mul iple-ca -chasing-single-space” phenomenon
[25] de i ed om he common ”blind sea ch” s a egy.
On-s ee O -s ee
URL Co e age
Mobile
s a s
pay
ese e
loca ion
p ices
pay
ese e
www.pa kme.com 4200 ci ies,wo ldwide 3 3 3 3 3 3
en.pa kopedia.com 6308 ci ies, wo ldwide 3 3 3 3 3
pa kna .com 70 ci ies, USA & Ge many 3 3
www.waze.com N/A3 3
maps.google.com 25 ci ies, USA 3 3 3
www.wazypa k.com big ci ies, Spain 3 3 3
www.apa candgo.com ai po s/s a ions 2 ci ies, Spain 3 3 3 3
www.pa kapp.com 16 ci ies, Spain 3 3 3 3
www.wesma pa k.com 2 ci ies, Spain 3 3 3 3 3
pa click.es 170 ci ies, Spain & EU 3 3 3 3
www. elpa k.com +60 ci ies, Spain & Po ugal 3 3 3 3 3
www.e-pa k.es 13 ci ies, Spain 3 3
www.pa kwhiz.com +200 ci ies, USA 3 3 3 3 3
www.s ee line.com N/A3 3 3 3 3 3
www.bes pa king.com +100 ci ies & ai po s USA & Canada 3 3 3 3 3
www.spo oops.com wo ldwide 3 3
www.jus pa k.com +1000 ci ies, UK 3 3 3 3 3
pa k ag.mobi 7 ci ies, EU 3 3 *11
www.pee opa k.com na ionwide, Spain 3 3 3 3
www.apa calia.com 3 ci ies, Spain +ai po s & s a ions 3 3 3 3
www.ca pling.com na ionwide, 13 coun ies 3
www.appypa king.com 11 ci ies, UK 3 3 3 3 3 3
Table 3. Sma pa king apps
When he in o ma ion is e ie ed in an au oma ic way, i is usually isualized as a
s a is ical es ima ion, showing he s ee maps wi h di e en colo s indica ing whe he
i is easy, medium o di icul o ind pa king spo s. This coa se g ain classi ica ion ags
can help d i e s heading owa ds zones whe e i is mo e likely o be success ul bu i
ob iously does no gua an ee an a ailable pa king space. This analysis is based on he
p ocessing o da a (i.e. cell phone signals) in a massi e way, gene a ing ma hema ical
models ha ep esen global da a, jus like i is done o show he a ic densi y du ing
di e en hou s o he day. Pa kMe (owned by INRIX), Pa kna , S ee line and Google
Maps, a e examples o pa king solu ions ha employ machine lea ning algo i hms o
p ocess au oma ically e ie ed pa king da a. S ee line is hea ily ocused on da a ana-
ly ics and o e s di e en solu ions, being ”Pa ke ” he one ha shows eal ime pa king
11 Planned