Da a in B ie 49 (2023) 109448
Con en s lis s a ailable a ScienceDi ec
Da a in B ie
jou nal homepage: www.else ie .com/loca e/dib
Da a A icle
Unde wa e imaging da ase in a e y shallow
wa e en i onmen o P amuka Island, Se ibu
Island Dis ic , Indonesia
Fick ie Muhammad
a , ∗, Poe bandono
a
, Ha ald S e nbe g
b
,
Eka Djuna sjah
a
, Hasanuddin Z. Abidin
c
a
Hyd og aphy Resea ch G oup, Facul y o Ea h Sciences and Technology, Bandung Ins i u e o Technology, Indonesia
b
Depa men o Hyd og aphy and Geodesy, Ha enCi y Uni e si y Hambu g, Ge many
c
Geodesy Resea ch G oup, Facul y o Ea h Sciences and Technology, Bandung Ins i u e o Technology, Indonesia
a i c l e i n o
A icle his o y:
Recei ed 5 June 2023
Re ised 12 July 2023
Accep ed 21 July 2023
A ailable online 26 July 2023
Da ase link: Unde wa e Imaging Da ase
in Nea Sho e Co al Ree En i onmen s a
P amuka Island, Se ibu Island Dis ic ,
Indonesia (O iginal da a)
Da ase link: Unde wa e Imaging Da ase
in Small Ha bo Pond Bay a P amuka
Island, Se ibu Island Dis ic , Indonesia
(O iginal da a)
Keywo ds:
Compu e ision
A -T ack ma ke acking
Unde wa e image y
Na u al seabed
Co al ee en i onmen
a b s a c
In his a icle, we p esen a da ase o unde wa e ideos
cap u ed h ough manual di es in a complex and uns uc-
u ed seabed a ea domina ed by ha bo s uc u es and co al
ee s. The a ea is shallow (0.5 –7.0 m dep h) wi h an en-
closed embaymen o he ha bo a ea, o e ing p o ec ion
om ocean cu en s and wa es. The co al ee a ea is lo-
ca ed in a mo e open ocean sloping gen ly owa d he deepe
seafloo , leading o a mo e p onounced olling shu e e ec
and came a mo ion.
The da ase was collec ed using a GoP o He o 10 came a, em-
ploying a s anda d wide lens wi h a ho izon al field o iew
(FoV) o 109 °and 768 ×432 image esolu ion. The came a is
also equipped wi h an Ine ial Measu emen Uni (IMU) sen-
so , comp ising a 200 Hz equency accele ome e and gy-
oscope. Du ing unde wa e deploymen , he came a is p o-
ec ed wi h a 5 mm hick fla glass panel. This came a se ing
hence c ea es h ee medium laye s o wa e -glass-ai leading
o addi ional e ac ion dis o ion.
To add ess he e ac ion dis o ion, he da ase has been
subjec o p e-calib a ion u ilizing fla e ac i e geome y
ound in he Pinax came a model. The Pinax came a model
∗Co esponding au ho .
E-mail add ess:
fi[email p o ec ed] (F. Muhammad) .
Social media: @fick ie67 (F. Muhammad)
h ps://doi.o g/10.1016/j.dib.2023.109448
2352-3409/© 2023 The Au ho (s). Published by Else ie Inc. This is an open access a icle unde he CC BY license
(
h p://c ea i ecommons.o g/licenses/by/4.0/ )
2 F. Muhammad, Poe bandono and H. S e nbe g e al. / Da a in B ie 49 (2023) 109448
o he unde wa e image y is calcula ed by combining he
aspec s o pinhole calib a ion pa ame e s wi h axial came a
p ojec ion. The main aim o he da ase collec ion is o a-
cili a e he es ing and e alua ion o unde wa e imaging al-
go i hms ha a e used in unde wa e obo ics, such as com-
pu e ision, pho og amme y, and Simul aneous Localiza ion
and Mapping (SLAM).
©2023 The Au ho (s). Published by Else ie Inc.
This is an open access a icle unde he CC BY license
( h p://c ea i ecommons.o g/licenses/by/4.0/ )
Specifica ions Table
Subjec Compu e Vision and Pa e n Recogni ion
Specific subjec a ea Unde wa e imaging: Unde wa e imaging is a c i ical componen o ision-based
na iga ion and mapping in he field o unde wa e obo ics, commonly employed
in compu e ision.
Type o da a 1. Rosbag file; ideo file eco ded as a Robo Ope a ing Sys em (ROS) bag file
o ma .
2. Media in Low Resolu ion Video (LRV) file o ma .
How he da a we e acqui ed The da ase has been acqui ed h ough manual di e in a shallow wa e
en i onmen , employing a GoP o He o 10 came a. The came a is equipped wi h a
s anda d wide lens possessing a field o iew (FoV) o 109 °and 768 ×432 image
esolu ion. The came a is also equipped wi h Ine ial Measu emen Uni (IMU)
senso wi h 200 Hz equency a e o gy oscope and accele ome e . The came a is
p o ec ed wi h fla panel glass housing wi h 5 mm hickness gene a ing h ee
dis inc medium laye s o wa e -glass-ai . The calib a ion da ase eco ds in bo h
unde wa e and in-ai en i onmen s a e also a ailable o acili a e he calib a ion
o he
came a.
Da a o ma 1. Low Resolu ion Video (LRV) file o ma con ains aw ideo oo age and mo ion
me ada a cap u ed om GoP o 10. The me ada a includes IMU eleme y da a
om accele ome e and gy oscope measu emen .
2. Rosbag file o ma eco ded wi h ROS. The osbag file da ase con ains
co ec ed image os opic wi h e ac ion adjus men om Pinax came a
model algo i hm and g ound- u h eco ding om A -T ack Ma ke de ec ion.
Desc ip ion o da a collec ion The aw ideo oo ages we e p e-p ocessed by using ROS ecosys em e sion Noe ic
in Ubun u 20.04. The p e-p ocessing includes g ound u h de ec ion using
A -T ack ma ke algo i hm and co ec ion map applica ion using Pinax came a
model algo i hm. The co ec ion map is used o ini ializing au oma ic came a
calib a ion and e ac ion adjus men . The esul ing p e-p ocessed ideo oo ages
a e hen sa ed as he osbag file.
Da a sou ce loca ion Ins i u Teknologi Bandung, Indonesia
P amuka Island
Se ibu Island Dis ic
Indonesia
−5.7432, 106.61345
Da a accessibili y Reposi o y name: Figsha eDa a iden ifica ion numbe :1. Small ha bo pond bay:
10.6084/m9.figsha e.227699122
. Nea sho e co al ee en i onmen :
10.6084/m9.figsha e.22886957
Di ec URL o da a:
1. Small ha bo pond bay: h ps://figsha e.com/a icles/media/
Unde wa e _ Imaging _ Da ase _ a _ Se ibu _ Island _ Dis ic _ Indonesia/22769912
2. Nea sho e co al ee en i onmen : h ps://figsha e.com/a icles/media/
Unde wa e _ Imaging _ Da ase _ in _ Nea _ Sho e _ Co al _ Ree _ En i onmen s _ a _
P amuka _ Island _ Se ibu _ Island _ Dis ic _ Indonesia/22886957
3. Modified Pinax came a model algo i hm o GoP o 10 came a:
h ps://gi hub.com/fick ie67/pinax-came a-model.gi
The p esen da ase encompasses low- esolu ion ideo (LRV) o ma files, which
a e eadily accessible and a ailable o immedia e u iliza ion. Addi ionally, he
osbag file o ma included in he da ase can be played h ough he u iliza ion o
he ROS (Robo Ope a ing Sys em).
F. Muhammad, Poe bandono and H. S e nbe g e al. / Da a in B ie 49 (2023) 109448 3
Value o he Da a
• The unde wa e image y da ase has been collec ed h ough manual di es in a complex and
uns uc u ed seabed, hus p o iding aluable esea ch ma e ial o esea che s in e es ed in
in es iga ing compu e ision applica ions wi hin eal-wo ld unde wa e en i onmen s.
• The da ase included wi hin hese files has been p e-calib a ed using he combina ion
Pinhole-Axial (Pinax) came a p ojec ion model algo i hm, which can be used o in es iga e
he mul ilaye geome y and e ac ion dis o ion e ec in compu e ision o unde wa e
en i onmen s.
• The da ase is also p o ided by he g ound u h measu emen om A -T ack ma ke acking
o assess he accu acy o any ision-based na iga ion algo i hm obus ness.
• Addi ionally, he aw da ase also s o ed he came a mo ion om IMU measu emen ha can
be used o assess he obus ness o ea u e de ec ion in compu e ision.
1. Objec i e
The main objec i e behind he gene a ion o his da ase is o p o ide an open, accessible,
and comp ehensi e esou ce om au hen ic unde wa e en i onmen s o conduc compu e i-
sion esea ch and s udies. The da ase o e s an oppo uni y o in es iga e he e ec s o he
unde wa e en i onmen cap u ed in he came a scene due o he densi y di e ences in he
medium. The medium di e ences will a ec he ay ligh p opaga ion and cause he e o poin
p ojec ion in he came a axis [1] . The s uc u e o he da ase acili a es in-dep h explo a ion
in he osbag file cu a ed om obo ope a ing sys em (ROS) ools, which con ains aw ideo
oo age, adjus ed ideo oo age, and g ound u h da a ob ained h ough A -T ack ma ke de ec-
ion.
The adjus ed ideo oo age is specifically add essing he e ac ion e ec u ilizing he
Pinhole-Axial (Pinax) came a model algo i hm. The Pinax came a model cons uc s he ay ans-
mission h ough medium laye s wi h fla panel glass geome y om [1] , c ea ing an image ec-
ifica ion co ec ion map. This aspec enables esea che s o examine he impac o lens and
e ac ion dis o ion caused by he unde wa e medium and assess he e ec i eness o he ap-
plied algo i hm. Fu he mo e, he da ase includes came a calib a ion oo age aken in-ai and
unde wa e en i onmen s, enabling he s udy o a ious calib a ion algo i hms and he po en-
ial de elopmen o new came a calib a ion models. The sui abili y o he da ase o 3D seabed
econs uc ion and ben hic habi a mapping is ein o ced by he inclusion o da a eco ded in
a closed-loop ack. Las ly, he da ase se es as a aluable esou ce o analyzing unde wa e
ision-based mapping and na iga ion using compu e ision and ROS.
2. Da a Desc ip ion
The da ase is collec ed h ough manual ( ee) di e se ies in wo di e en a eas a P amuka
Island, Se ibu Island Dis ic , Indonesia as shown in Fig. 1 . These a eas a e loca ed nea he
sho e and ha e shallow wa e dep hs anging om 0.5 o 7.0 m. A ea 1 (see Fig. 1 ) is loca ed in
a small ha bo pond wi h an app oxima e dep h o 0.5 m. I is cha ac e ized by a sandy seabed
and occasionally co e ed by a ha bo pla o m, which is use ul o ea u e acking. A ea 2 (see
Fig. 1 ) is si ua ed in a mo e exposed loca ion and has dep hs o up o app oxima ely 7.0 m. In
his a ea, he dis u bance caused by ocean cu en s and wa es is mo e p e ailing, leading o
oughe came a mo ion. A ea 2 is cha ac e ized by he p esence o co al ee s, con ibu ing o
mo e complex seabed ea u es. The Secchi disk eading in P amuka wa e s is anging be ween
7.0 and 11.0 m, indica ing a clea wa e en i onmen a ound he su ey a eas.
4 F. Muhammad, Poe bandono and H. S e nbe g e al. / Da a in B ie 49 (2023) 109448
Fig. 1. The di e loca ion o da ase acquisi ion wi h (1) he small ha bo pond a ea and (2) he nea -sho e co al ee
en i onmen a ea.
The da a s uc u e encompasses wo dis inc seabed ypes, each o which o e s significan
po en ial o u he u iliza ion. Wi hin he eposi o y, he da a is o ganized in o wo p ima y
ca ego ies: aw media oo age in Low-Resolu ion Video (LRV) file o ma and a osbag file o -
ma , as ou lined in Table 1 . The LRV file con ains addi ional me ada a collec ed om he GoP o
Ine ial Measu emen Uni (IMU), encompassing measu emen s om he accele ome e and gy-
oscope senso s. To ex ac he eleme y da a, a eleme y ex ac ion ool such as he GoP o
Me ada a Fo ma (GPMF) pa se om GoP o [2] can be employed. This da a is aluable o i-
sualizing he came a’s mo ion and o a ion h oughou he di ing ope a ion. Fu he mo e, hese
files can acili a e an in es iga ion o how he came a’s gene al mo ion unde wa e impac s he
obus ness o compu e ision algo i hms.
Table 1
Lis o aw da a (in LRV o ma ) and p ocessed da a (in osbag file o ma ).
Name Type
Video leng h
(seconds) Desc ip ion
GL010221.LRV Video oo age (LRV) 61 In-ai calib a ion pape
de ec ion
GL010221_all.bag Rosbag file
GL010106.LRV Video oo age (LRV) 54 Unde wa e calib a ion pape
de ec ion
GL010106_all.bag Rosbag file
GL010110.LRV Video oo age (LRV) 37 Small-scale ha bo pond
GL010110_all_ma ke .bag Rosbag file
GL010111.LRV Video oo age (LRV) 94 Medium-scale ha bo pond
GL010111_all_ma ke .bag Rosbag file
GL010162.LRV Video oo age (LRV) 125 Co al ee en i onmen 1
GL010162_all.bag Rosbag file
GL010163.LRV Video oo age (LRV) 172 Co al ee en i onmen 2
GL010163_all.bag Rosbag file
The ROS ools acili a e he ideo eco ding and cu a ion in he osbag file o ma . The osbag
file con ains h ee undamen al image opics, as illus a ed in Fig. 2 . Specifically, he aw media
oo age and he p e-p ocessed oo age ha has unde gone pos - e ac ion adjus men wi h a
F. Muhammad, Poe bandono and H. S e nbe g e al. / Da a in B ie 49 (2023) 109448 5
modified Pinax came a model o Ubun u 20.04 wi h OpenCV e sion 4 [3] a e included wi hin
he osbag file. They a e s o ed as came a/image_ aw and ec ified/le /image, espec i ely. This
o ganiza ion allows esea che s o di ec ly compa e and analyze bo h se s o da a. Addi ion-
ally, he osbag file con ains g ound- u h da a ep esen ed in local me ic A -T ack coo dina es.
This coo dina es sys em is ob ained h ough he de ec ion o A -T ack ma ke s by [4] wi h sub-
millime e ep ojec ion e o . This g ound- u h da a se es as a aluable e e ence o p ecise
analysis and e alua ion.
Fig. 2. S uc u e o osbag file: (a) The aw oo age, (b) The pos e ac ion adjus ed wi h Pinax came a model, and (c)
The came a posi ion om A -T ack ma ke acking.
The di e se ies is ca ied ou in he close-loop ack in andom mo ion which is beneficial o
es he close-loop algo i hm in SLAM, as shown in Fig. 3 . The fi s and second da ase in ol es
came a calib a ion oo age ob ained om mul i iew calib a ion pape de ec ion om in-ai and
unde wa e eco dings. The hi d da ase consis s o oo age eco ded in a small ha bo pond
(ci ca 2 ×2 m co e ed in 37 s) wi h ull suppo o closed-loop ope a ions and de ec ion o all
A -T ack ma ke s. In he ou h da ase , he oo age is acqui ed in a medium-sized ha bo pond
(ci ca 5 ×2 m co e ed in 94 s), whe e A -T ack ma ke de ec ion is conduc ed solely a he
beginning and end o he ack. Finally, he fi h and six h da ase in ol es oo age cap u ed in
a co al ee en i onmen wi hou he u iliza ion o A -T ack ma ke de ec ion. The six h da ase
exhibi s g ea e co e age o he co al ansi ion zone, ex ending om he ee fla o he deepe
ee c es . This expanded ange o dep hs (app oxima ely up o 7 m) and inc eased ack leng h
(app oxima ely 8 ×8 m co e ed in 172 s) o e mo e seabed ea u e a ia ions compa ed o he
fi h da ase . The fi h da ase , on he o he hand, co e s a smalle a ea (ci ca 5 ×8 m) in a
sho e ideo du a ion o 125 s. The inc eased co e age and a ia ions in he six h da ase a e
beneficial o es ing he isual ange limi a ions in compu e ision.
Fig. 3. The ajec o y o came a mo emen om he di ing se ies o (a) small ha bo pond (b) medium ha bo pond (c)
co al ee en i onmen 1 (d) co al ee en i onmen 2.
3. Expe imen al Design and Da a Acquisi ion P ocedu e
An illus a ion ep esen ing he p ocess o da a acquisi ion in ol ing he manual di e mo e-
men o a came a subme ged unde wa e in a complex and uns uc u ed seabed can be seen in
Fig. 4 . The ideo eco ding is ini ia ed wi h GoP o 10 came a ha is obus in unde wa e en-
i onmen s and has wide co e age wi h 109 °FoV. The came a is se o ake ideo in 30 F ames

6 F. Muhammad, Poe bandono and H. S e nbe g e al. / Da a in B ie 49 (2023) 109448
pe Second (FPS) which is a s anda d FPS o be used in ea u e de ec ion algo i hms ound in
compu e ision. The came a is also in eg a ed wi h an IMU senso wi h a 200 Hz equency
eco ding a e o ep esen he came a mo ion. Enclosed in a sealed 5 mm hick fla panel glass
housing, he came a is manually maneu e ed (by hand) in he wa e column o cap u e he
desi ed objec s on he seabed. This me hod allows o he collec ion o da a in a con olled
manne , ensu ing he eco ding o specific unde wa e scenes and condi ions o in e es , hus
p o iding aluable esea ch ma e ial o esea che s in e es ed in in es iga ing compu e ision
applica ions wi hin eal-wo ld unde wa e en i onmen s. Addi ionally, esea che s in e es ed in
s udying he unde wa e came a model and i s calib a ion e ec can use p e-p ocessed ideo
files s o ed in osbag file o ma .
Fig. 4. Illus a ion o da a acquisi ion.
The ma ke coo dina e sys em de i ed om A -T ack ma ke de ec ion acili a es he accu a e
me ic localiza ion o came a mo emen s unde wa e [4] . In his da ase , an A -T ack ma ke
wi h dimensions o 4,8 cm is u ilized, and i mus be co ec ly measu ed be o e execu ing he
A -T ack ma ke de ec ion algo i hm. The accu a e specifica ion o he A -T ack ma ke dimen-
sions ensu es p ecise measu emen s du ing he A -T ack ma ke localiza ion p ocess, con ibu -
ing o he o e all high accu acy o he localiza ion esul s.
3.1. Calib a ion oo age
The use o came a senso s shall add ess he lens dis o ion, namely adial and angen ial
dis o ion p io o any u he use in compu e ision [5] . In an unde wa e en i onmen , he
e ac ion e ec shall also be aken in o accoun which may cause la ge e o p opaga ion o e
ime. To mi iga e hese e ec s, he con en ional me hod o s anda d came a calib a ion shall
be applied. Mul i iew came a calib a ion echniques ha e demons a ed sui abili y in he con-
ex o unde wa e e ac ion adjus men . Howe e , i is impo an o no e ha his echnique
equi es he p ocessing o a subs an ial numbe o images o accu a ely compu e he dis o -
ion pa ame e s om a ious angles and posi ions [6] . The calib a ion p ocess en ails using a
known se o con ol poin s, ypically ob ained om a calib a ion pa e n o a ge wi h known
coo dina e o igins and dimensions. One commonly used calib a ion pa e n is a checke boa d
pa e n placed on a igid su ace o ice e sa. To acili a e he came a calib a ion, a ea u e de-
ec ion algo i hm is employed o de ec and loca e he checke boa d pa e n wi hin he came a’s
field o iew [7] . Fig. 5 shows he a ge is cap u ed by he came a om mul iple angles and
F. Muhammad, Poe bandono and H. S e nbe g e al. / Da a in B ie 49 (2023) 109448 7
posi ions, enabling he collec ion o he necessa y ansla ion and o a ion pa ame e s equi ed
o subsequen in insic and ex insic pa ame e calcula ions.
Fig. 5. Came a calib a ion p ocedu e: (a) Illus a ion o came a mo emen abo e checke boa d calib a ion pa e n. (b)
Typical example o he image sequence in he small ha bo pond bay o came a calib a ion unde wa e : (i) e ical, (ii)
oblique.
3.2. Re ac ion adjus men
Fig. 6 illus a es image oo age examples ha demons a e he isual e ec s o he dis o ion
and e ac ion co ec ion achie ed h ough he u iliza ion o he Pinax came a model algo i hm.
The Pinax came a model ep esen s he 3D seabed poin p ojec ion using he Pinhole came a
model combined wi h he fla e ac i e geome y o accumula e he axial poin p ojec ion in
he di e en medium laye s [3] . The Pinax calib a ion model builds upon he s anda d cam-
e a calib a ion p ocess, inco po a ing an addi ional scale adjus men o accoun o e ac ion
e ec s in oduced by he ansi ion be ween di e en mediums, such as wa e , glass, and ai .
These e ac ion-induced scale adjus men s a e conside ed pa o he ex insic pa ame e s in
he Pinax came a model [3] . To apply he Pinax came a model, in insic pa ame e s a e fi s
8 F. Muhammad, Poe bandono and H. S e nbe g e al. / Da a in B ie 49 (2023) 109448
equi ed, which can be ob ained h ough an in-ai calib a ion p ocess. This implies ha he
came a does no need o subme ge unde wa e du ing calib a ion. In he da ase , he in in-
sic pa ame e s a e calcula ed om he s anda d Pinhole came a calib a ion model. The in insic
pa ame e s used in he da ase consis o he ollowing componen s: he ocal leng h, deno ed
as x (458.058) and y (460.956); he came a p incipal poin cx (387.907) and cy (210.587); he
adial lens dis o ion coefficien s K1 ( −0.247) and K2 (0.0869); and he angen ial lens dis o ion
coefficien s P1 ( −0.006) and P2 (0.001).
Fig. 6. Example o image oo age wi h e ac ion adjus men in: (a) unde wa e calib a ion pape de ec ion a a dep h
o 0.5 m, (b) small ha bo pond bay wi h A -T ack ma ke de ec ion a a dep h o 0.5 m (c) co al ee en i onmen a an
app oxima e dep h o 2.0 o 3.0 m.
The ex ac ion o ex insic pa ame e s in ol es 3D seabed poin s p ojec ion unde wa e o-
wa ds he glass panel and hen di ec ly p ojec ing hem on o he came a’s image plane using he
known in insic pa ame e s. The adjus ed oo age inco po a es he s anda d e ac ion indices o
he glass o 1.492, he e ac ion indices o wa e o 1.342 es ima ed wi h a look-up able om
[8] , and in insic pa ame e s om in-ai calib a ion. The glass hickness is se o 5 mm, and
he i ual dis ances o he came a axis a e specified as 0.39 mm. No ably, hese pa ame e s a e
essen ial o he accu a e applica ion o he p e-p ocessed dis o ion co ec ion using he Pinax
came a model algo i hm.
E hics S a emen s
This da a a icle does no in ol e expe imen s on humans o animals.
CRediT Au ho S a emen
Fick ie Muhammad: Concep ualiza ion, me hodology, da a collec ion, da a cu a ion, W i ing
– o iginal d a p epa a ion; Poe bandono: Supe ision, Re iewing, and edi ing; Ha ald S e n-
be g: Supe ision and e iewing, Funding acquisi ion; Eka Djuna sjah: Re iewing; Hasanuddin
Z Abidin: Re iewing.
Decla a ion o Compe ing In e es
The au ho s decla e ha hey ha e no known compe ing financial in e es s o pe sonal ela-
ionships ha could ha e appea ed o influence he wo k epo ed in his pape .
Da a A ailabili y
Unde wa e Imaging Da ase in Nea Sho e Co al Ree En i onmen s a P amuka Island,
Se ibu Island Dis ic , Indonesia (O iginal da a) (Figsha e).
Unde wa e Imaging Da ase in Small Ha bo Pond Bay a P amuka Island,
Se ibu Island Dis ic , Indonesia (O iginal da a) (Figsha e).
F. Muhammad, Poe bandono and H. S e nbe g e al. / Da a in B ie 49 (2023) 109448 9
Acknowledgmen s
The wo k p esen ed he e is unded by Bandung Ins i u e o Technology (ITB) pos g adua e
esea ch scheme as an in eg al pa o he fi s au ho ’s doc o al esea ch. The Ge man Academic
Exchange Se ice (DAAD) is acknowledged o p o iding he schola ship o ca y ou he esea ch
a Ha enCi y Uni e si y (HCU).
Re e ences
[1] A. Ag awal, S. Ramalingam, Y. Taguchi, A heo y o mul i-laye fla e ac i e geome y, in: IEEE Con e ence on Com-
pu e Vision and Pa e n Recogni ion (CVPR), 2012, pp. 3346–3353
.
[2] GoP o IncGoP o Me ada a Fo ma (GPMF) Pa se , 2021 . [Online]. A ailable. h ps://gi hub.com/gop o/gpm -pa se .gi .
[3] T. Łuczy
´
nski, M. Pfings ho n, A. Bi k, The Pinax-model o accu a e and efficien e ac ion co ec ion o unde wa e
came as in fla -pane housings, Ocean Eng. 133 (1) (2017) 9–22, doi: 10.1016/j.oceaneng.2017.01.029
.
[4] H. Ka o, M. Billinghu s , Ma ke acking and HMD calib a ion o a ideo-based augmen ed eali y con e encing
sys em, in: 2nd In . Wo k. Augmen . Real. (IWAR 99), 1999, pp. 85–94, doi: 10.1109/IWAR.1999.803809
.
[5] T. T eibi z, Y.Y. Schechne , C. Kunz, H. Singh, Fla e ac i e geome y, IEEE T ans. Pa e n Anal. Mach. In ell. 34 (1)
(2012) 51–65, doi: 10.1109/TPAMI.2011.105
.
[6] F. De e nay, O. Fauge as, S aigh lines ha e o be s aigh : au oma ic calib a ion and emo al o dis o ion om
scenes o s uc u ed en i omen s, Mach. Vis. Appl. Sp inge Ve lag 13 (1) (2001) 14–24, doi: 10.10 07/PL0 0 013269
.
[7] D. Sca amuzza, F. F aundo e , Visual odome y, Pa I: he fi s 30 yea s and undamen als, IEEE Robo . Au om. Mag.
(Decembe ) (2011), doi: 10.1109/MRA.2011.943233
.
[8] R.W. Aus in, G. Halikas, The Index o Re ac ion o Seawa e - Technical Repo , San Diego Visibili y Labo a o y o The
Sc ipps Ins i u ion o Oceanog aphy, Uni e si y o Cali o nia, 1976
.