Walk line d awing
BACHELOR’S THESIS
submi ed in pa ial ul illmen o he equi emen s o he deg ee o
Bachelo o Science
in
Compu e Science
by
Pablo Balduz
Regis a ion Numbe 1635296
o he Facul y o In o ma ics
a he TU Wien
Ad iso : I an Viola
Vienna, 25 h Ap il, 2017
Pablo Balduz I an Viola
Technische Uni e si ä Wien
A-1040 Wien Ka lspla z 13 Tel. +43-1-58801-0 www. uwien.ac.a
E klä ung zu Ve assung de
A bei
Pablo Balduz
Medwedweg 3, 1110 Wien
Hie mi e klä e ich, dass ich diese A bei selbs ändig e ass habe, dass ich die e wen-
de en Quellen und Hil smi el olls ändig angegeben habe und dass ich die S ellen de
A bei – einschließlich Tabellen, Ka en und Abbildungen –, die ande en We ken ode
dem In e ne im Wo lau ode dem Sinn nach en nommen sind, au jeden Fall un e
Angabe de Quelle als En lehnung kenn lich gemach habe.
Wien, 25. Ap il 2017
Pablo Balduz
iii
Acknowledgemen s
Fi s o all, I would like o exp ess my since e g a i ude o my ad iso I an Viola o
gi ing me he oppo uni y o wo k on his Bachelo hesis a he Ins i u e o Compu e
G aphics and Algo i hms om Technische Uni e si ä Wien. I eally app ecia e all he
counsel gi en in ou weekly mee ings, as well as his p edisposi ion o help in any aspec
ela ed o he p ojec .
Fu he mo e, I am eally hank ul o my iends and amily o hei suppo du ing all
my s udies and especially du ing my s ay in Vienna wo king on his hesis.

Abs ac
In he ecen yea s, consequence o echnology imp o emen s, a new kind o a has
appea ed. I is called GPS a and consis s in d awing on a digi al map by eco ding he
pa h ollowed using a GPS de ice.
The ac is ha no e e yone is able o make he d awing o a ce ain igu e. Jus he
so-called GPS a is s come up wi h he pa h, so i makes his kind o a no eachable
o some people.
The idea o his hesis is o enable people o c ea e GPS a wi hou elying on imagina ion
o come up wi h a pa h o a ce ain igu e. In o de o achie e ha , a sys em ha inds
ha pa h in a map o a gi en igu e as inpu has been de eloped. In o de o people o
use his sys em, i has been in eg a ed wi hin a mobile applica ion, so use s a e able o
ind he pa h o ollow easily.
ii
Con en s
Ku z assung ii
Abs ac ix
Con en s xi
1 In oduc ion 1
1.1 Mo i a ion .................................. 1
1.2 Me hodological app oach . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 S uc u eo wo k.............................. 3
2 S a e o he a 5
2.1 Spo sapps................................. 5
2.2 Image egis a ion ............................. 6
3 Me hod 11
3.1 Imagecompa ison .............................. 11
3.2 Remapping ................................. 14
4 Implemen a ion 17
4.1 GoogleMaps ................................ 17
4.2 OpenCV................................... 18
4.3 And oid.................................... 21
5 Conclusion and ou look 23
5.1 Summa y and c i ical e iew . . . . . . . . . . . . . . . . . . . . . . . 23
5.2 Fu u ewo k................................. 24
Lis o Figu es 25
Glossa y 27
Ac onyms 29
Bibliog aphy 31
ix
2. S a e o he a
2.2 Image egis a ion
Image egis a ion is he p ocess o es ablishing he spa ial o empo al co espondence
among da a sou ces. This da a may be mul iple images acqui ed a di e en imes and/o
wi h di e en modali y. I is used in compu e ision, medical image, au oma ic a ge
ecogni ion and in he analysis o images and sa elli e da a. The egis a ion is necessa y
o be able o compa e o in eg a e he da a ob ained om hese di e en measu emen s.
[2] [3]
In image egis a ion he e a e no mally ou s eps. Fi s , a su icien numbe o con ol
poin s a e equi ed in o de o es ima e an op imal geome ic ans o ma ion be ween wo
images ( ea u e de ec ion). A e ha , ea u e ma ching is necessa y o ind co esponding
poin s because many ea u e poin s may be ex ac ed om one image bu no om
he o he . Then a ans o ma ion unc ion is used o model he geome ic ela ionship
be ween he wo images and inally, he unknown pa ame e s o he unc ion can be
compu ed acco ding o he con ol poin s. [4]
Among da a images, one o hem is known as he e e ence and ano he image is known
as he a ge . Image egis a ion is a p ocess o inding he spa ial ans o ma ion o
he a ge image o align i wi h he e e ence image and he e a e wo undamen al
app oaches o achie e his ans o ma ion. In ensi y based me hods compa e he in ensi y
pa e ns in he images h ough co ela ion me ics, whe eas ea u e based me hods ind
co espondence be ween image ea u es, such as poin s, lines and con ou s. In ensi y
based me hods egis e comple e images o sub-images. I sub-images a e egis e ed,
cen e s o he co esponding sub-images a e conside ed as co esponding ea u e poin s.
Fea u e-based me hods es ablish he co espondence be ween se e al poin s in images.
Knowing he co espondence be ween se e al poin s in he images, a ans o ma ion is
de e mined o map he a ge image o he e e ence images, es ablishing poin by poin ,
he co espondence be ween he e e ence images and he a ge .
Image egis a ion algo i hms can also be classi ied acco ding o he ans o ma ion
models ha hey use o ela e he a ge image space o he e e ence image space. Rigid
ans o ma ions a e he simples ca ego y; i includes linea ans o ma ions, which a e
ansla ion, o a ion, scaling and o he a ine ans o ma ions. Linea ans o ma ions
a e global in na u e; he e o e, hey canno model local geome ic di e ences be ween
images.
Apa om igid ans o ma ions, he e is ano he ca ego y ha allows elas ic o non- igid
ans o ma ions. These ans o ma ions a e able o locally de o m he a ge image o
align i wi h he e e ence image. Non- igid ans o ma ions include adial basis unc ions,
con inuous physical models and la ge de o ma ion models.
An example o ans o ma ion, which is in ac impo an in his p ojec , is he dis ance
ans o m. A dis ance ans o m is a de i ed ep esen a ion o a digi al image, no mally
only applied o bina y images. Each pixel o he image is labeled wi h he dis ance o he
nea es obs acle pixel. A mos common ype o obs acle pixel is a bounda y pixel, which
6

2.2. Image egis a ion
means a di e en alue pixel, in a bina y image. The esul o he ans o m is a g ay
le el image ha looks simila o he o iginal one, excep ha he g ay le el in ensi ies o
poin s inside o eg ound egions a e changed o show he dis ance o he closes bounda y
om each poin . [5] [6]
Figu e 2.1: Example o how pixel alues a e assigned in a dis ance ans o m. [5]
A ep esen a ion o he example abo e is he ollowing:
(a) O iginal image. (b) Dis ance ans o m image.
Figu e 2.2: Example o he dis ance ans o m applied on a bina y image. [6]
While he p edominan numbe o app oaches is pe o med in he spa ial domain, he e
a e a ew app oaches ha ans o m he da a in o equency domain o sol e a pa icula
egis a ion ask. Spa ial me hods ope a e in he image domain, ma ching in ensi y
pa e ns o images cha ac e is ics. Some o he algo i hms ha ma ch cha ac e is ics
a e de i a ions o he adi ional echnics o manually egis e ing an image, in which an
ope a o decides he co esponding con ol poin s in he images. When he numbe o
con ol poin s exceeds he minimum equi ed o de ine an app op ia e ans o ma ion
model, he i e a i e algo i hms such as RANSAC (Random Sample Consensus) can
be used o obus ly es ima e he pa ame e s o a pa icula ype o ans o ma ion in
image egis a ion. Me hods in equency domain ind he ans o ma ion pa ame e s
o he image egis a ion while wo king in he ans o ma ion domain. These me hods
wo k by simple ans o ma ion, such as ansla ion, o a ion and scaling. Fo example,
7
2. S a e o he a
applying he phase co ela ion me hod o a pai o images p oduces a hi d image, which
has peak in ensi ies a loca ions whe e he wo images ma h he bes . The in ui ion
behind his me hod is ha he esul ing image will ha e he maximum peak alue
when he con en s o he images ma h up exac ly and ha peak loca ion co espond o
he ela i e ansla ion be ween he images [
7
]. Unlike many algo i hms in he spa ial
domain, phase co ela ion me hod is noise, occlusions and o he de ec s ypical o medical
o sa elli e images, insensi i e. Besides, he phase co ela ion uses he Fas Fou ie
T ans o m o calcula e he c oss-co ela ion be ween he wo images, which esul s in
g ea pe o mance imp o emen s. This me hod can be ex ended o de e mine he o a ion
and scale di e ences be ween wo images.
Ano he classi ica ion can be made be ween monomodal and mul imodal me hods.
Monomodal me hods egis e images in he same modali y acqui ed by he only ype o
scanne /senso , while mul imodal me hods end o egis e images ob ained by di e en
ypes o scanne s/senso s. Mul imodal egis a ion me hods a e o en used in medical
imaging since he images o a pa ien a e ob ained by equency om di e en scanne s.
Regis a ion me hods can be classi ied acco ding o he le el o au oma ion hey o e .
The e ha e been de eloped manual, in e ac i e, semiau oma ic and au oma ic me hods.
Manual me hods p o ide ools o manually align images. In e ac i e me hods educe use
bias by pe o ming ce ain key ope a ions au oma ically while s ill elying on he use o
guide he egis a ion. Semiau oma ic me hods make mo e egis a ion s eps au oma i-
cally bu hey depend on he use o e i y he co ec ness o a egis a ion. Au oma ic
me hods do no allow use in e ac ion and make all egis a ion s eps au oma ically.
A measu e o image simila i y quan i ies he deg ee o simila i y be ween he in ensi y
pa e ns o wo images. The choice o a measu e o image simila i y depends on he
modali y o he images o be egis e ed. The mos common examples o image simila i y
measu es include c oss-co ela ion, mu ual in o ma ion, sum o squa es o in ensi y
di e ences, and image uni o mi y a io. C oss-co ela ion, sum o he squa es o in ensi y
di e ences and image uni o mi y a io a e used o egis e ing images in he same
modali y. On he o he hand, mu ual in o ma ion and no malized mu ual in o ma ion
a e he mos popula image simila i y measu es o mul imodal images egis a ion. An
example o simila i y measu es applica ion is shape ma ching. Shape ma ching is highly
ela ed o simila i y measu es because i deals wi h ans o ming a shape, and measu ing
he esemblance wi h ano he one using some simila i y measu e. So, shape simila i y
measu es a e an essen ial ing edien in shape ma ching. [8]
2.2.1 Applica ions
Image egis a ion has applica ions in emo e sensing (ca og aphy upda e) and compu e
isions. Because o he many applica ions o which image egis a ion can be applied, i
is impossible o de elop a gene al me hod ha is op imized o all uses.
Regis a ion o he medical image ( o da a om he same pa ien aken a di e en
imes, such as he de ec ion o changes o con ol o a umo ), o en also in ol es elas ic
8
2.2. Image egis a ion
egis a ion o make ace de o ma ion o he pa ien (due o b ea hing, ana omical changes
and so on). The non- igid medical imaging eco d can also be used o eco d a pa ien ’s
da a o an ana omical a las. I is also used in as opho og aphy o align images aken
om space. Using he con ol poin s, he compu e pe o ms ans o ma ions on an
image so ha he main ea u es a e aligned wi h a second image.
Image egis a ion is an essen ial pa o c ea ing pano amic images. The e a e many
di e en echniques ha can be implemen ed in eal ime and un on in eg a ed de ices
such as came as and came a phones.
9
CHAPTER 3
Me hod
The de eloped sys em inds in a map o pedes ian-accessible a eas a pa h ha makes a
desi ed isual pa e n depic ed as a simple line d awing.
The algo i hm consis s o wo majo s ages: one is he image compa ison made h ough
egis a ion and he second s age is he emapping o he igu e desi ed o ep esen on o
he map.
3.1 Image compa ison
In he i s pa o he algo i hm, he compa ison be ween bo h, he map and he igu e
image is made. The ou come o his s age is o gene a e a new g ay scale image in which
each pixel is labeled wi h a di e en le el o in ensi y. The lowes in ensi y alue a eas
o he new image ( ha means he da kes pa s) encode a highe p obabili y o ind a
ep esen a ion o he igu e in ha a ea. This s age can be unde s ood as a simila i y
e alua ion.
The compa ison is applied on wo bi map images: one is he igu e in ended o be ound
in he map ep esen ed as simple line d awing as showed in Figu e 2.2 and he o he one
is he map. This map image is ob ained om Google Maps bu ge ing an image om
he de aul map ha appea s in he Google Maps applica ion o use in he compa ison
would be useless. Tha map has many di e en colo s, shapes, names o s ee s, oads
and places ha would in oduce many e o s du ing he simila i y e alua ion as we migh
be compa ing pixels o he igu e wi h pixels o he name o a ce ain s ee . To sol e
ha , a new map in which only he s ee laye emains isible and in which he e a e
no di e en shades o colo s mus be gene a ed and i is done ia he Google Maps
Applica ion P og amming In e ace (API). Using he Google Maps API by ollowing he
s eps desc ibed in Chap e 4, we a e able o gene a e a map wi h he desc ibed ea u es,
shown in Figu e 2.2.
11

3. Me hod
(a) Sample igu e
ha could be used
in he compa ison.
(b) Gene a ed map using Google Maps API.
Figu e 3.1: Example o a sample image and map ha could be used in he algo i hm.
The compa ison is no di ec ly made o e wo images like he ones abo e, compa ing
wo bi maps would be e y sensi i e. Unless he igu e pe ec ly ma ches some a ea in
he pa e n o he s ee laye , he di e ences would be e y high. E en i he alignmen
be ween he igu e and he s ee laye is nea ly pe ec , a he momen in which a pixel
o he igu e and a pixel o he map ha does no co espond o any s ee a e compa ed,
i will esul in a high di e ence as well. To sol e his undesi ed high sensi i i y o he
simila i y ans o m gene a ed in his s age, ins ead o a bi map image o he map we use
i s dis ance ans o m image like he one in Figu e 3.2b. This way, du ing he simila i y
e alua ion, e en i he alignmen be ween he igu e and he s ee laye is nea ly pe ec ,
i will esul in a high simila i y sco e due o he much lowe di e ence be ween he
images being compa ed, which is he desi ed beha io o a nea ly pe ec aligned igu e.
The dis ance ans o m allows o be e ob aining a measu e o how close he igu e being
compa ed is o he each s ee in he map g id, as each pixel in he new map indica es
he dis ance o i s nea es s ee .
(a) G ay scaled map. (b) Map’s dis ance ans o m.
Figu e 3.2: Example o he dis ance ans o m applied on a g ay scaled map p e iously
gene a ed.
A e he simila i y e alua ion, a new image bi map is c ea ed. As al eady explained,
12
3.1. Image compa ison
each pixel o his new image indica es he simila i y le el o he igu e a he posi ion
indica ed by ha pixel. A low pixel alue, which co esponds o he da k pa s, means a
high simila i y sco e. This image is ob ained as a esul o he compa ison o he igu e
image all o e he map dis ance ans o m.
Du ing he whole p ocess, he igu e image is cen e ed on each pixel o he map dis ance
ans o m so ha he igu e image i s in he map and he compa ison is ca ied ou ,
ha is o say, he igu e image is no cen e ed on pixels in which pa o he igu e would
emain ou o he map image bounds because he e would be a pa ha could no
be compa ed. I makes no sense o compa e pa o he igu e because he aim o his
compa ison is inding a ep esen a ion o he whole igu e. Du ing each compa ison, he
compu ed alue is assigned o a new image ha will co espond o he esul simila i y
image. I he igu e image is cen e ed o e he map dis ance ans o m and i e a es all
o e i ow a e ow s a ing on he op le co ne , he new image should be illed
ollowing he same di ec ions, ow a e ow in his case. In he end, he simila i y image
dimensions would be he di e ence be ween he igu e and map image dimensions.
The simila i y image pixel alues a e ob ained by compa ing bo h he igu e image
and he co esponding pa o he map dis ance ans o m. In his compa ison he
posi ions o black pixels in he igu e a e aken and used o ob ain he alues o he
pixels co esponding o he igu e in he dis ance ans o m. Fo each o he igu e pixels
we now ha e a alue ha indica es how a hey a e o hei closes s ee in he map
image and by compu ing he sum o hose alues we ha e an indica o o how close he
igu e is o i s ep esen a ion on he map s ee s. The highe his alue is, he wo se he
simila i y as i indica es ha he igu e is a om being ep esen ed in he s ee laye .
A e assigning e e y alue o he new bi map, we may ha e an image like he one in
Figu e 3.3. The da ke pa s indica e high simila i y sco e as al eady said. I he image
gene a ed is oo uni o m ha no hing can be dis inguished, we can a y e e y pixel
in ensi y un il much mo e di e en ia ed a eas appea .
Figu e 3.3: New g ay scale simila i y image ob ained om he compa ison be ween he
map dis ance ans o m and he igu e.
The compa ison made so a only uses he igu e image wi h he de aul o ien a ion and
13
3. Me hod
scale, bu he igu e may be ound in he map wi h a di e en angle o o a ion and size.
In o de o suppo ha , i is necessa y o add i in o he compa ison. In each i e a ion,
he igu e image is o a ed and scaled so ha he sum o he dis ance alues is compu ed
o e e y o ien a ion and scale combina ion. A he end o he i e a ion he lowes alue,
which is he bes esul as p e iously desc ibed in his chap e , is used o he new image
gene a ed and s o ed wi h he co esponding o ien a ion and scale.
As well as he igu e image may ha e di e en scales, so does he map. Nowadays, digi al
maps ha e di e en le els o scale; he mo e we zoom in on a map, he mo e s ee s
appea on he sc een and he mo e we zoom ou on a map, he ewe s ee s we a e able
o see. This is app ecia ed in Figu e 3.4 and means ha he whole compa ison desc ibed
so a needs o be done o di e en map scales in o de o co e a wide ange. In ac ,
his imp o es he chances o inding he igu e in he map.
(a) Zoom le el 1. (b) Zoom le el 2.
(c) Zoom le el 3.
Figu e 3.4: Di e en zoom le els o a map.
3.2 Remapping
The compa ison be ween he images gene a es a new g ay scale map indica ing he pa s
whe e he chances o inding a ep esen a ion o he igu e a e highe . A e ha , he
emapping should be done; his means mo ing he igu e on o he s ee s in he map g id
and his is achie ed using he dis ance ans o m map.
Fo e e y posi ion in he new g ay scale image, besides he alue he e is in o ma ion
abou he igu e o ien a ion and scale. This in o ma ion is enough o do he emap. Fo
a gi en posi ion in he map, he co esponding igu e image wi h i s o ien a ion and scale
in ha posi ion is compu ed in o de o ob ain he posi ions o he igu e black pixels.
Then, he map dis ance ans o m image is used o do he emap.
In he dis ance ans o m image, a pixel ha ing a alue o 12 means ha his pixel is 12
posi ions (pixels) away om he nea es pixel ha has alue 0, which co esponds o a
14
3.2. Remapping
s ee . Only wi h he dis ance ans o m he e is no clue on which di ec ion o ollow
o ge o he pixel wi h alue 0. As he dis ance indica ed by he dis ance ans o m
is he dis ance o he nea es ce o alue pixel, he di ec ion o ollow will be he one
wi h g ea es a e o a ia ion and ha is gi en by he g adien . So in o de o ind
he di ec ion o ollow, he g adien in ha pixel mus be calcula ed. Once ha ing he
di ec ion, he ce o alue pixel is jus 12 posi ions away in ha way (in his example).
Following ha p ocess o e e y o he black pixels posi ions will lead o a bunch o new
posi ions in he map which will co espond o he new emapped igu e as shown in
Figu e 3.5b.
(a) Map and igu e o e layed. (b) Figu e emapped.
Figu e 3.5: Example o he a igu e emapping in o a map o a gi en posi ion.
15

CHAPTER 5
Conclusion and ou look
5.1 Summa y and c i ical e iew
This hesis began wi h a single idea: T y o ep esen a d awn igu e in he s ee s map
laye o a ci y in an au oma ed way. A e doing a p elimina y esea ch, no hing simila
was ound. E e y hing done un il hen was igu es ep esen ed by a is s ha had come
up wi h he pa h in a ce ain egion.
A i s app oach was ca ied ou by compa ing, bo h he image o he co esponding map
and he igu e. This compa ison was made o e he dis ance ans o m o he images.
The esul s ob ained we e no ha good as using he dis ance ans o m in bo h images,
he numbe o da k pixels in he igu e inc ease and he sums o pixel alues di e ences
along he map we e p e y uni o m.
The second app oach imp o ed he esul s. Ins ead o using he dis ance ans o m o
he map and he image, only he map one was used and only he black pixels in he
igu e we e aken o compu e he image wi h he sum o pixel di e ences. This made
his image o ha e much di e en ia ed pa s, which is be e .
Rega ding he implemen a ion pa , he i s echnology chosen o de elop he algo i hm
was Ma labs bu he esul s we e no good a all. In e ms o ime consuming, i was eally
slow so i was decided o change o C++ and use OpenCV lib a ies. Time consuming
imp o ed a lo , al hough i is s ill slow due o he amoun o calcula ions ha a e being
made be ween he images.
As a inal e lec ion on his hesis, he main conclusion would be ha he me hod
pe o med is good as a i s app oach. The esul s demons a e ha he igu e shape is
mo e o less emapped on o he s ee s laye . This wo k has se ed o con i m ha he
ini ial idea was achie able and opens a new pa h in which i can be imp o ed.
23
5. Conclusion and ou look
5.2 Fu u e wo k
The i s app oach o he sys em p esen ed in his hesis opens a new pa h o imp o ing
i and adding new ea u es. The algo i hm de eloped ocuses on he basics, which is
ansla ing a igu e in o he s ee s map laye wi hou any condi ions.
The sys em ea s e e y pa o he map as equal, ha is o say; o example, in a map a
building, a squa e o a pa k may look simila (as an emp y space) bu his ac is no
ele an o he algo i hm. A nex di ec ion o ollow could be somehow o penalize
posi ions o he map wea he hey a e walk able o no , so he emapping does no send
he use o e a building o example.
In e ms o pe o mance, he e is ano he way in which he sys em could be imp o ed.
Cu en ly i akes a long ime o do all he calcula ions, especially i he images used a e
big, so i would be a good idea o op imize i and lowe he ime o execu ion. A good
di ec ion o ollow may be o use GPU compu ings.
24
Lis o Figu es
1.1 Exampleo GPSa .[1]............................ 2
2.1 Example o how pixel alues a e assigned in a dis ance ans o m. [5] . . . 7
2.2 Example o he dis ance ans o m applied on a bina y image. [6] . . . . . 7
3.1 Example o a sample image and map ha could be used in he algo i hm. 12
3.2
Example o he dis ance ans o m applied on a g ay scaled map p e iously
gene a ed..................................... 12
3.3
New g ay scale simila i y image ob ained om he compa ison be ween he
map dis ance ans o m and he igu e. . . . . . . . . . . . . . . . . . . . . 13
3.4 Di e en zoom le els o a map. . . . . . . . . . . . . . . . . . . . . . . . . 14
3.5 Example o he a igu e emapping in o a map o a gi en posi ion. . . . . 15
4.1 S eps o ollow in o de o use OpenCV in an Xcode p ojec . [13] . . . . . 20
25
Glossa y
compu e ision
In e disciplina y ield ha deals wi h how compu e s can be made
o gaining high-le el unde s anding om digi al images o ideos. F om he
pe spec i e o enginee ing, i seeks o au oma e asks ha he human isual sys em
can do. 8, 18
GPS a
o GPS d awing is a me hod o d awing ha uses GPS echnology o c ea e
la ge-scale a wo k. I combines a , mo emen , and echnology. 1, 5
GPU compu ing
Is he use o a G aphics P ocessing Uni (GPU), which ypically
handles compu a ion only o compu e g aphics, o pe o m compu a ion in appli-
ca ions adi ionally handled by he Cen al P ocessing Uni (CPU). 24
Ma lab
(MAT ix LABo a o y) is a mul i-pa adigm nume ical compu ing en i onmen
and ou h-gene a ion p og amming language. 23
open sou ce
Decen alized de elopmen model ha encou ages open collabo a ion. I s
main p inciple is pee p oduc ion, wi h p oduc s such as sou ce code, bluep in s,
and documen a ion eely a ailable o he public. 18
Xcode
In eg a ed de elopmen en i onmen o macOS con aining a sui e o so wa e
de elopmen ools de eloped by Apple o de eloping so wa e o macOS, iOS,
wa chOS and OS. 18, 19
27

Ac onyms
API Applica ion P og amming In e ace. 11
BSD Be keley So wa e Dis ibu ion. 18
GPS Global Posi ioning Sys em. 1, 5
OpenCV Open Sou ce Compu e Vision. 18, 19, 21, 23
OS Ope a ing Sys em. 18
29
Bibliog aphy
[1]
A. G i. A is d aws wo ld’s la ges doodles by iding his bike wi h gps. [Online].
A ailable: h p://www.bo edpanda.com/bike-gps-doodle-s ephen-lund/
[2]
(2017, Ap il). [Online]. A ailable: h ps://en.wikipedia.o g/wiki/Image_ egis a ion
[3] J. F. Ba ba a Zi o á, “Image egis a ion me hods: a su ey,” June 2003.
[4]
Z. X. . Y. Zhang, “A c i ical e iew o image egis a ion me hods,” In e na ional
Jou nal o Image and Da a Fusion, 2010.
[5]
(2016, Ma ch). [Online]. A ailable: h ps://en.wikipedia.o g/wiki/Dis ance_
ans o m
[6]
(2003). [Online]. A ailable: h p://homepages.in .ed.ac.uk/ b /HIPR2/dis ance.h m
[7] [Online]. A ailable: h p://www.cs.u ah.edu/~ssingla/IP/P4/Index.h ml
[8]
R. C. Vel kamp, “Shape ma ching: Simila i y measu es and algo i hms,” Dep .
Compu ing Science, U ech Uni e si y, Tech. Rep.
[9]
[Online]. A ailable: h ps://de elope s.google.com/maps/documen a ion/ja asc ip /
s yling?hl=es-419
[10] [Online]. A ailable: h ps://maps yle.wi hgoogle.com
[11] [Online]. A ailable: h ps://console.de elope s.google.com
[12]
[Online]. A ailable: h ps://de elope s.google.com/maps/documen a ion/ja asc ip /
ge -api-key
[13]
(2015, Decembe ). [Online]. A ailable: h p://s acko e low.com/ques ions/
34340578/ins alling-c-lib a ies-on-os-x
[14] [Online]. A ailable: h ps://b ew.sh
[15] [Online]. A ailable: h ps://de elope .and oid.com/s udio/index.h ml?hl=es-419
[16]
[Online]. A ailable: h ps://www.you ube.com/wa ch? =n 4MEliij14&lis =
PL6 5F68 1ZZzTDq8VI9Jcmb0J99WR Yn4&index=6
31