Towa ds a P e-diagnose o Su gical Wounds h ough he Analysis o
Visual 3D Recons uc ions
Neus Mun ane Es a ellas3, F ancisco Bonin-Fon 1, Juan J. Segu a-Samped o2,
And es Jim´
enez Ram´
ı ez3, Pep L. Neg e Ca asco1, Miquel Masso Campos1,
F ancesc X. Gonzalez-A gen ´
e2and Gab iel Oli e Codina1
1Depa men o Ma hema ics and Compu e Enginee ing, Uni e si y o he Balea ic Islands,
C a. Valldemossa Km 7.5, 07122 Palma de Mallo ca, Spain
2Depa men o Gene al and Diges i e Su ge y, Uni e si y Hospi al Son Espases,
07122 Palma de Mallo ca, Spain
3G oup o Ingenie ´
ıa Web y Tes ing Temp ano, Depa men o Languages and Compu ing Sys ems, Uni e si y o Se illa,
c/ San Fe nando 4, 41004 Se illa, Spain
Keywo ds: 3D Visual Recons uc ion, S uc u e F om Mo ion, Pos -su gical Wound, Telemedicine.
Abs ac : This pape p esen s a new me hodology o p e-diagnose he s a e o pos -su gical abdominal wounds based
on isual in o ma ion. The p ocess consis o ou majo phases: a) building dense 3D econs uc ion o he
abdominal a ea a ound he wound, b) selec ing an a ea close o he wound o i a plane, c) calcula ing he
dis ance om each poin o he 3D model o he plane, d) analyzing his map o dis ances o in e i he wound
is in lamed o no . This me hod needs o be w apped in an applica ion o be used by pa ien s in o de o sa e
unnecessa y isi s o he medical cen e .
1 INTRODUCTION
The eme gency and he ou pa ien acili ies o he
Spanish public heal h ca e sys em a e usually col-
lapsed by he nume ous o unnecessa y isi s o he
assis ance cen e s ha could be sol ed a home wi h
se e al indica ions gi en by he co esponding spe-
cialis .
In 2017, mo e han 3000 ope a ions we e done
in he Uni e si y Hospi al Son Espases, in Palma de
Mallo ca, om which, only app oxima ely he hal o
hem we e p og ammed. Tha makes a mean o 250
pa ien s pe mon h, 63 pa ien s pe week, and 13 pa-
ien s pe day. I he consul mean ime in Spain is
in ended o ange be ween 6 and 10 minu es, i makes
a mean o 2 hou s a day dedica ed only o ake ca e
o pos su ge y wounds, wi hou aking in o consid-
e a ion he es s o asks, such as, new pa ien s, non
su ge y pa ien s, managemen mee ings o eme gen-
cies.
E e y su ge y pa ien is e alua ed wice a e he
su gical p ocedu e. The i s e iew is a he heal h
cen e a e a week. And i is e alua ed again one
mon h la e as ou pa ien , a he hospi al. One o he
main easons o his e alua ion is o check he su gi-
cal wound and de ec i s in ec ion. Howe e he a e
o in ec ion is low and when i happens i is usually
de ec ed in he eme gency depa men .
This ace- o- ace consul a ions, whe e mos o
hem p esen no anomalies, could be easily managed
emo ely, ha ing a cheape cos and a ec ing less
he pa ien s’ quali y o li e, as hey equi e unneces-
sa y ans e s o he heal h acili y and absences om
wo k. Mo eo e , i he pa ien is unable o suspec
he wound in ec ion on ime, as i usually happens, an
inc ease o he eme gency depa men consul a ions
is p oduced, no mally wi h a delay in he wound in-
ec ion diagnosis which esul s in an inc ease o isi
ime pe pa ien .
Consequen ly, e e y echnological p og ess in he
ield o heal h ca e managemen , in gene al, and in he
pos -su ge y assis ance in pa icula , is e y use ul o
educe he cos s, o imp o e he quali y o assis ance
ime and hus o inc ease he quali y o li e o he
pa ien s. Medical compu e and mobile applica ions
ocused on emo e au oma ic diagnose and pa ien
managemen /moni o ing ha e been ad ancing in he
las yea s ci eneph o low, (Topdoc o s, 2017). La ely,
some s udies in elemedicine suppo he easibili y
and sa e y o emo e ollow-up in su gical wounds
Es a ellas, N., Bonin-Fon , F., Segu a-Samped o, J., Ramí ez, A., Ca asco, P., Campos, M., Gonzalez-A gen é, F. and Codina, G.
Towa ds a P e-diagnose o Su gical Wounds h ough he Analysis o Visual 3D Recons uc ions.
DOI: 10.5220/0006628505890595
In P oceedings o he 13 h In e na ional Join Con e ence on Compu e Vision, Imaging and Compu e G aphics Theo y and Applica ions (VISIGRAPP 2018) - Volume 4: VISAPP, pages
589-595
ISBN: 978-989-758-290-5
Copy igh ©2018 by SCITEPRESS – Science and Technology Publica ions, Lda. All igh s ese ed
589
and he pa ien s sa is ac ion (Segu a-Samped o e al.,
2017). In his la e e e ence as well as in simila
s udies (No dheim e al., 2014), pho os o ideos o
he wound and illed ques ionnai es cons i u e he ex-
changed da a be ween he pa ien and he doc o , bu
is he doc o who always analyzes and e alua es he
ecei ed in o ma ion. Howe e , o ou knowledge,
none o he e ised me hods o Apps is able o es i-
ma e, au oma ically, a eliable p e-diagnose based on
isual da a, and o il e ou hose wounds ha clea ly
p esen a good e olu ion, wi hou he in e en ion o
he physician. Following his line, he Depa men o
Gene al and Diges i e Su ge y o he Uni e si y Hos-
pi al Son Espases is collabo a ing wi h he Sys ems,
Robo ics and Vision g oup o he Uni e si y o he
Balea ic Islands, in o de o go one s ep o wa d in he
design and implemen a ion o a ision-based mobile
App o elemedicine ha can help in he es ima ion
o a p e-diagnose o abdominal pos su ge y wounds.
The objec i e is il e ing, au oma ically, hose wounds
ha po en ially p esen in lamma ion as sign o in ec-
ion and need a ace- o- ace e alua ion in he hospi al
om hose ha cou se a no mal e olu ion and can be
managed a home, sa ing ime and medical esou ces.
The no el y o his wo k is mo e in he me hodology
i sel and he applica ion han in he pipeline o isual
algo i hms designed o ge he objec i e. This me hod
needs o be w apped in o a u u e compac mobile
App, which will con ain addi ional unc ionali ies o
inc ease he communica ion and da a exchange be-
ween doc o s and pa ien s.
The wound analysis p ocess consis s o a pipeline
ha in ol es he nex s eps: a) g ab a ideo sequence,
wi h he mobile, o he abdominal zone a ound he
wound, om side o side, iewing he same a ea bu
om di e en pe spec i es and iewpoin s, b) ex ac
images o he ideo sequence, c) ex ac and ack
common isual ea u es in all he images, d) build a
3D spa se poin cloud using a S uc u e F om Mo ion
(SFM) (Ha ley and Zisse man, 2003) algo i hm, e)
build a dense poin -cloud and a ex u ed meshed su -
ace, ) es ablish a polyline and a plane i ed in his
polyline in a selec ed po ion o he 3D model; his
plane is in ended o be, ei he angen o he abdomen
su ace, o c ossing he abdominal a ea, below he
wound, g) compu e he dis ance be ween each poin
o he 3D model and he plane, and emi a diagnose
unc ion o hese dis ances.
2 METHODOLOGY
Fi s ly, he pa ien mus eco d wi h he mobile ele-
phone a ideo o he wound, om side o side o he
abdominal a ea, in o de o ha e iews om di e en
pe spec i es and iewpoin s. The second s ep is au-
oma ic and consis o ex ac ing all he images om
he ideo sequence. Once he images ha e been ex-
ac ed, he p ocess o 3D econs uc ion s a s au o-
ma ically wi h he ea u e acking p ocess. The SFM
geome ic heo y (Ha ley and Zisse man, 2003) is
based on he acking o a se o wo ld poin s p o-
jec ed in se e al images aken by he same came a
om di e en iewpoin s. These p ojec ed poin s
and hei co espondences in he subsequen images
a e ob ained hanks o a p ocess o a classical isual
ea u e de ec ion and ma ching (Ha ley and Zisse -
man, 2003) using wo epu ed de ec o s in a ian o
o a ion and scale: one de ec o wi h scala desc ip-
o , SIFT (Lowe, 2004), and one de ec o wi h bina y
desc ip o s, ORB (Rublee e al., 2011). Bo h ech-
niques ha e p o ed ex endedly his excellen pe o -
mance in e ms o numbe o ea u es, obus ness and
aceabili y. In a iance o scale and o a ion is impo -
an o his kind o applica ion since he image key
poin s mus be iden i ied in all ames o he ideo se-
quence, which show he a ec ed a ea om di e en
iewpoin s. Figu e 1 shows an image p o ided by he
Uni e si y Hospi al Son Espases o a su gical wound,
wi h he isual ea u es ob ained using he 2 di e en
de ec o s.
(a) (b)
Figu e 1: Fea u e de ec ion wi h: (a) SIFT, (b) ORB.
The ea u e de ec ion wi h he 2 es ed ea u es has
been implemen ed wi h he ea u e de ec o OpenC 2
unc ions. The desc ip o ma ching has been imple-
men ed wi h he FLANN (Muja and Lowe, 2009)
ma che lib a y. Good ma ches (inlie s) a e consid-
e ed o be hose which dis ance be ween co espon-
dences in di e en images is unde a ce ain h eshold
( ypically, ei he 0.02 o 2 imes he minimum dis-
ance be ween all he ma ches). Bad ma ches a e dis-
ca ded.
Gi en he p ojec ion ma ices, he 3D coo dina es
o a wo ld poin can be ob ained om i s co espond-
ing image poin s (in his case isual ea u es) iden i-
ied in se e al iews (ma ching) using iangula ion.
Ideally, he 3D poin should lie in he in e sec ion o
all back-p ojec ed ays. Bu , in gene al, hese ays
will no in e sec in a single poin due o he e o s in-
he en o he ea u e ma ching p ocess. The 3D coo -
dina es o he wo ld poin a e ob ained minimizing he
VISAPP 2018 - In e na ional Con e ence on Compu e Vision Theo y and Applica ions
590
sum o squa ed e o s be ween he measu ed and he
p edic ed image posi ions o he 3D poin p ojec ed
in all in ol ed iews whe e he wo ld poin is isible:
X=a gminx∑ikui−ˆuik2, whe e uiis he p edic ed
image poin and ˆuiis he co esponding measu ed im-
age poin , o all he iimages whe e he 3D poin is
p ojec ed. The p edic ed image poin can be ob ained,
o example, om he ea u e ma ching p ocess, and
he measu ed poin is di ec ly he p ojec ion o he
wo ld poin on he image.
The lib a y OpenMVG (Moulon e al., 2017) im-
plemen s he poin iangula ion and he 3D eco -
e y applying he SFM and he epipola heo ies, e-
co e ing also he came a displacemen , and has been
used o o m spa se 3D poin -clouds om inpu se s
o images. OpenMVS (cDcSeaca e, 2017) p o ided
us wi h a comple e se o algo i hms o eco e a ull,
ine and ex u ed su ace om a se o came a poses
and a spa se poin -cloud.
Figu e 2 illus a es he image p ocessing pipeline
designed and implemen ed o ob ain a dense 3D
model o he eco ded a ea. The ou i s s eps ha e
been p og ammed wi h OpenMVG, and he las ou
s eps wi h OpenMVS. The SFM implemen a ion is
based on (Moulon e al., 2013). This p ocess is inc e-
men al, which means ha , he i s econs uc ion is
done only wi h wo iews, and a e e y i e a ion a new
iew is inco po a ed adding ea u es, some ma ch-
ing wi h he p e ious and needing iangula ion, some
new in he scene. The econs uc ion wi h known
poses e e s o he p ocess o e ining he 3D model
using he came a poses and he 3D poin posi ion el-
a i e o he know came a poses, once he SFM and
he came a mo ion ha e been compu ed. The poin
cloud densi ica ion is based on (Ba nes e al., 2009),
he mesh econs uc ion is based on (Jancosek and
Pajdla, 2014), he mesh e inemen is based on (Vu
e al., 2012), and inally, he mesh ex u ing is based
on (Waech e e al., 2014).
Figu e 2: Image p ocessing pipeline o wound 3D econ-
s uc ion.
A 3D econs uc ion o an abdominal a ea wi h
a pos su gical wound is p esen ed in igu e 3 as
(a) (b)
(c) (d)
Figu e 3: A esul o he 3D econs uc ion pipeline: (a) a
ame o he eco ded ideo, (b) he dense poin cloud, (c)
e ined mesh, (d) e ined and ex u ed mesh.
a sample o he pipeline pe o mance. 130 ames
we e ex ac ed om he ideo. One sample image is
shown in igu e 3-(a) and he ideo can be seen in
h ps://you u.be/XW18WMFZPTw. The appea ance
o he econs uc ed abdomen is highly ealis ic, bu
wi hou a me ic scale i is impossible o in e he 3D
s uc u e dimensions.
Figu e 4: Templa e used o scale he ob ained 3D model.
The ob ained 3D model has no me ical uni s,
since i is compu ed om he images and he i-
sual ea u e coo dina es which a e exp essed in pix-
els. Con e ing hese da a in o me ical da a was nec-
essa y o es ima e he wound s a e om he 3D e-
cons uc ion. To his end, he geome ic empla e o
igu e 4 was designed. This empla e con ains one
colo calib a ion pa e n, no ye used in his wo k,
one geome ical ma ke which side measu es 3.1cm
and a ec angula hole in be ween. The empla e mus
be placed on he abdomen wi h he wound alling
jus inside he ec angula hole. Once he empla e
is co ec ly placed, he ideo can be eco ded. The
scale a io applicable o all econs uc ed 3D poin s
can be calcula ed di iding he ma ke side eal me ic
Towa ds a P e-diagnose o Su gical Wounds h ough he Analysis o Visual 3D Recons uc ions
591
by i s leng h measu ed in he 3D model. A sample
o a ideo eco ded wi h he empla e can be seen a
h ps://you u.be/IN kQlNXbu0.
In o de o pe o m he las s eps o he pipeline
which include he plane i ing and he dis ance calcu-
la ion, he esul ing 3D poin cloud was opened wi h
Cloud Compa e (Gi a deau-Mon au , 2017). Using
his applica ion, he 3D olume was c opped a ound
he wound and scaled acco ding o he measu es p o-
ided by he ma ke o he empla e (i i was a ail-
able). A e wa ds, a polyline was c ea ed inside he
p ocessed 3D olume o i a plane inside i . This
plane was in ended o be, ei he coinciden wi h he
plane o he ma ke , o pa allel o i , angen o he
abdomen su ace, jus a he wound base, o cu ing
he abdomen su ace in ou poin s, below he wound.
Finally, he dis ance o each poin o he cloud o he
i ed plane was calcula ed and expo ed o a cs ile.
The analysis o hese dis ances o each case leads o
an a emp o es ima ed diagnose.
3 EXPERIMENTS
In o de o e alua e he comple e p ocedu e, some
simula ed expe imen s we e ini ially pe o med using
he empla e and a small cable wi h a diame e o 4mm
simula ing an in lamed wound. Figu e 5-(a) shows an
image o one simula ed scene.
(a) (b)
Figu e 5: Simula ed wound: (a) an image o he simula ed
scene, (b) he ex u ed mesh.
The leng h o he ma ke side in he 3D econ-
s uc ion, measu ed in Cloud Compa e was 0.122898
uni s. Knowing ha he eal leng h o he ma ke side
is 3.1cm, he scale ac o se in Cloud Compa e o
he 3 di ec ions (x, y, z) was 0.031(m)/0.122898 =
0.25224169m. In his case, he i ed plane was co-
inciden wi h he plane o he ma ke . Figu e 5-(b)
shows he ex u ed mesh.
Figu e 6 shows he selec ed olume in yellow (a),
he same olume wi h he i ed plane (bo de in whi e
and plane a ea in blue) in (b), and he spa ial map
o dis ances be ween all poin s o he selec ed ol-
ume and he i ed plane, in (c). No ice how he pos-
i i e (yellow-o ange) dis ances ange be ween 2mm
and 4mm, along he ec angle, clea ly di e en ia ing
he cable p o ile om he su ounding a ea (blue). Ly-
ing he i ed plane on he empla e, hese posi i e dis-
ances coincide app oxima ely wi h he cable diame-
e .
Figu e 7-(a) shows an image o ano he in lamed
wound simula ed wi h a small cable. Figu es 7-(b), 7-
(c) and 7-(d) show, espec i ely, he e ined and ex-
u ed mesh, he selec ed olume in yellow wi h he
polyline i ing plane in whi e and he map o dis-
ances om poin s o he plane. The scale ac o e-
sul ed in 0.031m/0.0611679 =0.50680177m, being
0.0611679 he ma ke side leng h measu ed in Cloud
Compa e. The map o dis ances shows clea ly he
linea shape o he simula ed in lamed wound in he
cen e o he ec angle wi h posi i e dis ances a ound
3mm su ounded by poin s ha ma k nega i e dis-
ances below −4mm.
Figu e 8-(a) shows he selec ed olume in yellow
a ound he wound o igu e 3 wi h he polyline i ing
he plane in whi e. In his case, he plane in e sec s he
abdomen below he wound, in one pa , and abo e,
in ano he . This is a case o a wound wi h a good
e olu ion, wi hou in lamma ion. Figu e 8-(b) shows
he co esponding map o dis ances o he i ed plane.
The cen al a ea in yellow indica es whe e he plane
is below he wound and he dis ance is bigge while
he ex emes which end o blue indica e whe e he
dis ance is smalle . In his case, he me ic uni s and
alues ha e no ele ance. Since he e a e no poin s
in he cen e o he map ha ma k he shape o he
wound, one can conclude ha mos likely he e is no
in lamma ion on he explo ed a ea.
Figu e 9-(a) shows an image ex ac ed om a
ideo o ano he example o pos su gical wound wi h
a good e olu ion and no in lamma ion. In his case
he scale is also i ele an . 35 images we e ex ac ed
o build he e ined and ex u ed 3D model, shown
in igu e 9-(b). In his expe imen , he plane was
i ed below he abdomen as shown in whi e in ig-
u e 9-(c). The map o dis ances is show in igu e 9-
(d). The la e shows clea ly he di e ence be ween
he cen al a ea wi h posi i e dis ances co esponding
o he zone wi h maximum cu a u e o he abdomen
(maximum dis ance o he plane), and bo h sides ( op
and bo om) whe e he dis ance be ween he plane and
he abdomen is minimum, wi hou any pa sugges -
ing he p esence o any in lamed a ea in he o m o
a ans e sal line o dis ances clea ly abo e he es .
These ype o esul ing plo would sugges o he pa-
ien and o he doc o , in p inciple, an unnecessa y
ace- o- ace e ision.
Figu e 10-(a) shows an image ex ac ed om
VISAPP 2018 - In e na ional Con e ence on Compu e Vision Theo y and Applica ions
592
(a) (b) (c)
Figu e 6: Simula ed wound: (a) he selec ed olume and he polyline a ound i depic ed in yellow, (b) he polyline in yellow
and he i ed plane in whi e, (c) map o dis ances om each poin o he i ed plane.
(a) (b)
(c) (d)
Figu e 7: Expe imen 2: (a) an image ex ac ed om he
ideo sequence, (b) e ined and ex u ed mesh, (c) he
wound wi h he i ed plane depic ed in whi e, (d) Map o
dis ances om each poin o he i ed plane.
(a) (b)
Figu e 8: Expe imen 3: wound o igu e 3: (a) he wound
wi h he polyline and he i ed plane, (b) Map o dis ances
om each poin o he i ed plane.
ano he expe imen on a eal pos -su gical wound
g abbed wi h he empla e. In his case, he wound
p esen s an e iden in lamma ion o se e al milime-
e s. 17 images we e ex ac ed om he ideo o
build he 3D model. The scaling ac o u ned ou
(a) (b)
(c) (d)
Figu e 9: Expe imen 4: (a) an image ex ac ed om he
ideo sequence, (b) e ined and ex u ed mesh, (c) he
wound wi h he i ed plane, (d) Map o dis ances om each
poin o he i ed plane.
o be 0.359min all di ec ions. Figu es 10-(b), 10-
(c) and 10-(d) show, espec i ely, he e ined ex u ed
mesh, he 3D model wi h he polyline in whi e i -
ing he plane, coinciden wi h he empla e cen al
hole, and he co esponding g aphic o dis ances. The
la e shows how he poin s loca ed in he bo de s o
he ec angle p esen dis ances be ween −5mm and
−10mm, while in he middle, especially in he uppe
pa which coincides wi h he side o he wound ha
has he ma ke a i s le and i is clea ly below he
plane, poin dis ances ange be ween 0mm and 5mm.
Al hough he e is a clea g adien o dis ances be-
ween some pa s o he cen e and he sides o he
e alua ed a ea, suscep ible o being p e-diagnosed as
in lamed, he shape o he wound is no clea ly iden-
i ied, being necessa y a inal and de ini i e diagnose
gi en by he doc o .
Towa ds a P e-diagnose o Su gical Wounds h ough he Analysis o Visual 3D Recons uc ions
593

(a) (b)
(c) (d)
Figu e 10: Expe imen 5: (a) an image ex ac ed om
he ideo sequence, (b) e ined and ex u ed mesh, (c) he
wound wi h he i ed plane, (d) map o dis ances om each
poin o he i ed plane.
Finally, igu e 11-(a) shows one ame o a wound
which was opened due o an in e nal in ec ion. Fig-
u e 11-(b), igu e 11-(c) and igu e 11-(d) show, e-
spec i ely, he e ined and colo ed mesh, he 3D a ea
wi h he polyline i ing he plane, in whi e, and he
map o dis ances. The empla e was pu jus on he
skin, and he plane was i ed a ound he hole. The
dis ance map e idences a blue zone in he middle co -
esponding o he opened wound wi h dis ances be-
low he plane a ound 3mm (−3mm), su ounded by
a o ange a ea wi h dis ances be ween 1mm and 4mm
abo e he plane co esponding o he skin. This g adi-
en o dis ances ma ks clea ly an anomaly in he a ea.
The posi ion in which he empla e is placed on
he wound, i s adjus men o he abdomen, and he
way he plane i i ed in he selec ed olume a ec s
clea ly he ob ained esul s. The p ocedu e needs o
be e ined, bu he ini ial esul s a e clea ly encou ag-
ing.
4 CONCLUSIONS
This pape has p esen ed an inno a i e me hodology
o es ima e a p e ious diagnose o pos -su gical ab-
dominal wounds using isual da a. Al hough expe i-
men s wi h simula ed scenes e eal a clea di e ence
be ween he simula ed in ec ed wound and he back-
g ound, esul s o expe imen s wi h eal wounds a e
s ill on a p elimina y s age, bu poin ing in a clea
good di ec ion. Now he challenge lies, mainly, in
(a) (b)
(c) (d)
Figu e 11: Expe imen 6: (a) an image om he ideo se-
quence, (b) e ined and ex u ed mesh, (c) he wound wi h
he i ed plane, (d) map o dis ances.
wo di e en issues: a) e ine he cu en p ocess o
ge clea e esul s, basically es ing se e al possibili-
ies in he posi ioning o he empla e and in he gen-
e a ion o he i ed plane, and b) in eg a ing his me-
hodology in a so wa e package, which au oma izes
all he s ages o he p ocess un manually wi h Cloud
Compa e o be in eg a ed in a use s applica ion.
ACKNOWLEDGEMENTS
This wo k is pa ially suppo ed by Minis y o
Economy and Compe i i eness unde con ac s
TIN2014-58662-R, DPI2014-57746-C3-2-R and
FEDER unds.
REFERENCES
Ba nes, C., Shech man, E., Finkels ein, A., and Goldman,
D. (2009). Pa chMa ch: A Randomized Co espon-
dence Algo i hm o S uc u al Image Edi ing. ACM
T ansac ions on G aphics, 28(3):24:1–24:11.
cDcSeaca e (2017). OpenMVS. Open Mul i-
View S e eo Recons uc ion Lib a y.
h ps://gi hub.com/cdcseaca e/openMVS.
Gi a deau-Mon au , D. (2017). Cloud Compa e, 3D Poin
Cloud and Mesh P ocessing So wa e Open Sou ce
P ojec . h p://www.danielgm.ne /cc/.
Ha ley, R. and Zisse man, A. (2003). Mul iple View Geom-
e y in Compu e Vision. Camb idge Uni e si y P ess.
Jancosek, M. and Pajdla, T. (2014). Exploi ing Visibil-
i y In o ma ion in Su ace Recons uc ion o P ese e
VISAPP 2018 - In e na ional Con e ence on Compu e Vision Theo y and Applica ions
594
Weakly Suppo ed Su aces. In e na ional Schola ly
Resea ch No ices, 2014.
Lowe, D. G. (2004). Dis inc i e Image Fea u es om Scale-
In a ian Keypoin s. In . Jou nal on Compu e Vision,
60(2):91–110.
Moulon, P., Monasse, P., and Ma le , R. (2013). Global
Fusion o Rela i e Mo ions o Robus , Accu a e and
Scalable S uc u e om Mo ion. In P oceedings o he
In e na ional Con e ence on Compu e Vision.
Moulon, P., Monasse, P., Ma le , R., and O he s (2017).
OpenMVG. An Open Mul iple View Geome y Li-
b a y. h ps://gi hub.com/openMVG/openMVG.
Muja, M. and Lowe, D. (2009). Fas App ox-
ima e Nea es Neighbo s wi h Au oma ic Algo-
i hm Con igu a ion. Sou ce code a ailable a
h ps://gi hub.com/ma iusmuja/ lann/.
No dheim, L., Haa ind, M., and I e sen, M. (2014). E -
ec o Telemedicine Follow-up Ca e o Leg and Foo
Ulce s: a Sys ema ic Re iew. BMC Heal h Se ices
Resea ch, 14(565):58–62.
Rublee, E., Rabaud, V., Konolige, K., and B adski, G.
(2011). ORB: An E icien Al e na i e o SIFT o
SURF. In P oceedings o he 2011 In e na ional Con-
e ence on Compu e Vision, pages 2564–2571.
Segu a-Samped o, J., I.Ri e o-Belench´
on, Pino-D´
ıaz, V.,
S´
anchez, M. R., Pa eja-Ciu ´
o, F., Padillo-Ruiz, J., and
Rod ´
ıguez, R. J. (2017). Feasibili y and Sa e y o Su -
gical Wound Remo e Follow-up by Sma Phone in
Appendec omy: A Pilo S udy. Annals o Medicine
and Su ge y, 21:58–62.
Topdoc o s (2017). Topdoc o s. www. opdoc o s.es/.
Vu, H., Laba u , P., and Pons, J. (2012). High Accu acy and
Visibili y-Consis en Dense Mul i iew S e eo. IEEE
T ansac ions on Pa e n Analysis and Machine In el-
ligence, 34(5):889–901.
Waech e , M., Moeh le, N., and Goesele, M. (2014). Le
The e Be Colo ! La ge-Scale Tex u ing o 3D Recon-
s uc ions. In P oceedings o he Eu opean Con e -
ence on Compu e Vision, pages 836–850.
Towa ds a P e-diagnose o Su gical Wounds h ough he Analysis o Visual 3D Recons uc ions
595