3Dobj

ectmodel
ing
Cont
ent
s:
 I
ntr
oduc
tion-
whati
s3Dmodel
ing
?

 ModelSt
ruc
tur
e

 ModelAppear
anc
e

 Model
:Imag
es&Si
mul
ati
on

 Model
ingSof
twar
e

 Howar
eModel
sus
ed?

 Th
eMaj
ori
ssues

 Model
sandRender
ing

 Oper
ati
onsonModel
s

 Repr
esent
ati
onSt
ruc
tur
e

 ModelCl
ass
ifi
cat
ion

 Wh
eredoModel
sComeFr
om?

KeyWor
ds:
 3DRec
ons
truc
tion

 Def
ormabl
eModel

 Mul
ti
gri
dGr
adi
entVec
torFl
ow

 Vi
sualh
ull

 Tex
tur
e
Wh
ati
s3Dmodel
ing
?

Everyc omput er-

render edimag erequi r
est hreees sentialcomponents
:
a3Ds cenedes cri
ption,oneormor es our c
esofl ight,andades cr
iptionoft h
ecamera
orey ev i
ewi ngthes cene.Th escenedes cripti
onistypicall
yc omposedofoneormor e
model s,or3Ds tructures.Typical
lywet hi
nkofamodelasas t
and-alonepar t,
e.g.apenc ilorat ree, andth esceneast heas semblyoft hes epartsintoa
compl ete3Denv ironment .Thisattitudereec tsthemos tcommonpr ocedure
forbui l
dingupa3Ds cene:onebui ldsmanymodel s
,andt h enas semblest hem.
Eac hmodelc ontainstwodes cri
ptions :
amat hematicalrepr esentati
onoft h e
structureoft h
es h
ape,andar ec ipef or_ gur
ingouth ow thats hapewoul dl
ookif
il
lumi nated. Letusex aminet h
esedes cri
pt i
oninturn.

1
.1ModelSt
ruc
tur
e
Thes t ruc t
ur aldes cri
ptioni sbas icallyoneofg eomet ry
.Ittellsuswh eret heobjec tisi
n
spac e, andwh ereitisnot .I
mag ineanempt yc o_ eemugh overinginspac einf r
ontofy ou.
Nowf reez et i
meandl ookatev er ymol eculei nt her oom:
Gener allyeac hmol ec ulewi llbepar toft hemat eri
aloft hemugorpar toft h
eai r
aroundt hemug( notet hatt heai rmol eculesi ns i
det hemugwh eret h eco_eewoul dgo
arenotpar toft hemugi tself;t hey '
repar toft h eai raroundi t).I
fy oupai ntev eryair
mol ec ulewh iteandev erymugmol eculeblac k,t heny ou'l
lhav e
av ery_ nedes cri
ptionoft hemugdownt ot hepr ecisionofth emol eculesi nther oom.
Ther es ul toft hi
sthoug htex per imenth asmanyoft h eproper ti
esof3Dmodel sus edin
comput erg raph i
cs:
I t
'sc onc eptuallystr
ai ght
for war d( ifbulky),
andh asal i
mi tedprecision( mos tc omputerpr og ramsus et hebui l
t-
inc alcul
a-
ti
onh ar dwar ei ntoday'sc omput er s ;t
hishardwar ehasv eryh i
ghbutl imi ted
prec i
s ion) .Somemodel ingmet hodsar ev eryc los et ot hi
sappr oac h;t heycreatepoi nt
s
inspac eorc h
opups pac ever y_ nel yandl abeli tempt yorful l
.
1
.2ModelAppear
anc
e
Theot herpar tofev erymodeli st hes ur facedes cription.Th isboi lsdownt odes cribing
theph y
s i
csofh owt hes urf
ac eoft h emodeli nterac tswi thlight.Happi ly,
ittur nsoutt hat
thes eph ysicsar edr ivenbyaf ewdes criptivet ermst hath avei ntuitivemeani ngforus ,
suchasc ol or, shinines s,andt rans parenc y.
Th usy ouc oul ds ayt haty ourc o_eemugi s
shinyandbl ue, andg oal ongwayt owar dsdes cribingi tsappear anc e.Sinc emodel sof t
en
simul ater ealwor l
ds hapes ,weh av et odec ideh owdet ailed
wewantt obe. Wh endes c
ribingthec o_ eemug 'sg eomet ry
, dowewantt oincl udeev ery
li
ttlebumpi nt hec erami candt hei rreg ulari
t i
esi nt heov er allshape, orisa
mat h ematic allyper fectc y
linderc los eenoug h ?
Wh endes cribingi tsappear anc e,dowewantt oi ncludet helit
t les cratchesi nthe
glaz e,orisi tenoug htos ayt hatiti sjus tunif orml ys h
iny ?
Th eans wer st ot heseques ti
onsdependonh owy oui ntendt ous ethemodel .I
fit's
goingt obes it
t i
ngont hes helfinth ebac kgroundofs omes cene, poorlyl i
tandbar el y
visible,thenabl uec yl
indermi ghtdot h etrick.Buti ft hemugi st hes tarofy ourimag e,
frontandc ent er,larg eandwel l-
lit,th eny ouwi l
lpr obabl ywantmor edet ail
.

2Model
s:I
mag
esandSi
mul
ati
ons
Ther ear et wopr i
nc ipler eas onsf orbui l
ding3Dmodel sont hec omput er:image-maki ng
ands imul ation. Th es etwopr ocedur esar ev erys imilar,butt heyt ypicallyh avedi ffer
ent
goal s.
Imag e- maki ngr equi r esmodel st hatl ookg ood. Th eyneedt obes uff ici
ent lycompl exthat
theyar en'tbor ingt ot heey e,ands uf f
icient lyr ealisti
c( wh ent hat '
st heg oal)t opas s
conv incinglyasag ooddes c r
iptionoft her eals h ape.Theul ti
mat ear biteroft hemodel '
s
qual ityi st heper s onwh ol ooksatt hepi cture.Si mul ationr equir esmodel st hatar e
acc urate. Somes i
mul ationst es tt wo
model smat hemat ic allyt os eei ft hey- tt og eth erwi t
hins omet oler anc e;ther es ul
tis
simpl yay esornoans wer ,rath ert h anani mag e.Cons ideras i
mul ationt hattes tsthe
streng thoft her oofofanews tadi umunderc ondi ti
onsofh eav yrain:
It'
sc rit
ic alt hatt hes i
mul at edr oofh av eex actl
yt h ei ntendeds hapei nor dert o
deter mi neh owmuc hr ainwi llrol lont hes ides .Anai r
planewi ngal s omus tbemodel ed
withh i
g h ac curac yi f we ex pec tt ol ear n any thing meani ng ful about i tsl if
t
char acter i
s ticswh enac tual l
yac tuallybui lt.
Somemodel sar et her es ultofas imul ation: givenac omput er-contr olleddr il
l,abl ockof
wood, andapr og r
amf ort hedri ll
, wh ati sth es hapeoft her es ulti
ngpar t?
Her eweonl ys uppl yth einitialmodel ,andt h ec omput ermodi lest hemodelf orus .We
maybeasi nt erest edi nt hes urfac ev ani shoft he_ nulpiec easi tsg eomet ry.Aholemi gh
t
ber ound, forex ampl e,butwewoul dl iket obes ure
thati tiss moot h.Th es etwoappl icationsh av es pawnedt wor atherdi ff
er entwaysof
des cribing
model s.Onei si nter activeandappr ox imat e, appr opriatef ormaki ngmodel sthatj ust
needtolookgood.Theotherissl
owerandprecise,appropriat
eformodel st
hatmus tbe
accur
ate.Ofcourse,
there'
sah ugeoverl
apbetweent h
es ediff
erentsyst
ems,s i
ncemany
modelsusedforimagesneedt obeveryprec
ise,andsomet imessimulat
ionsjustneed
somethi
ngcloseasas tart
ingpoi
ntorstand-
inforamor ec omplexshape.

3Det
ail
:HowMuc
h?

Detai
lsaddi nteres
t .
Th eamountofdet ai
linamodelc anmaket hedi
ffer
encebetween
aboringandf akeimag e,oronethatisint
erest
ingandr eal
ist
ic.
Forsi
mul at
ion,
detai
ls
canmaket hedi f
ferencebetweeng et
tingananswerthatis
ri
ghtorwrong .

3.
1Vi
sualDet
ail

Earlyc omput erg raph icsimag esus edv er ys impl emodel s :

At abl ewasal ongbox ,
madeupofs ixper f
ec tlyat,per fectlys moot h ,perfectly
coloreds i
des . Realt abl esar emuc hmor ei nter esti
ngbec aus et heyh av edet ails.They
haves t ruct
ur aldet ail,andappear ancedet ai l(alsocalledt ex ture).Lookatanyman-
madeornat ur al3Dobj ect,andy ou'l
ls eeaweal thofs h
apedet ailandappear anc e
detail:not h
ingi sper fec tl
yat ,smoot h, orf eat ureless.Tomakei nteres ti
ngand
believabl eimag es ,weneedmodel sthath aveenoug hr i
ch nes st hattheybeg int omat ch
thec ompl ex ityoft h
er ealwor l
d( i
deal izedf or msar eus ef ulf orar tandeduc ation,but
event h erec ompl exityc anh elpy oug ety ourpoi ntac r
oss ).Th ec ompl ex ityofwh ats eem
tobes implef ormsc anbes tag gering:lookc los elyatthec oiledl iner unningf roma
teleph oneh ands ett ot hebas e.Th ewi resf ol l
owav eryc ompl icatedpat h,theg aps
betweenc oilsi nc reas eanddec reas e, thec oilst hemselv esl oopar oundandt hrough
eac hot her.Th ev ani shont hewi rei sn'tcons ist enteither:th erear es cr atchesand
discolor ati
onsal lal ongi tsleng th,somet i
mesi nrandompl ac esands omet i
mesi n
organi zedc hunks( l
ikewh er eth ec ordg otc aug htands crapedi nadoor wayoneday ).I
t'
s
thi
ski ndofdet ailt hatmakesamodeli nter es ti
ng,andc reat ingt h i
sdet ai li
soneoft he
principlec hal leng est oany onewh omakesmodel s.
Ing ener al
,mor edet ailint heg eomet r yandt h eappear anc ei sal way sbet ter.

Butther
ear et woc aveat st
ot hi
sprinci
ple:moredetailtakesyou(thedes i
gner)more
ti
met ocreat e,andittakest hecomputer(therenderer)mor eti
met opr oc
ess.Int
h e
previ
ousex ampl e,aver ydetai
ledmugs itt
inginadarkc ornerwil
ljustslowdowny our
pict
ure-
maki ngandaddnot hi
ng.Thesit
uationsthat3caus ethemostt r
oubleoccurin
animati
ons ,whenas hapes tartsoutfarawayandbar elyvisi
ble,
andl atercomesto
occupymuc hoft heimag e(suchasa
basebal
lh urtli
ngtowar dsthec amera).
3.
2Si
mul
ati
onDet
ail

Ifyou'rebui l
dingamodelofas taplerandy ouwanttomakes urethatt
hes tapleswi
ll
bedel iveredc orr
ec tl
y,i
twillbei mportanttobeveryaccurat
ei ny
ourmodeloft hel
it
tle
hollowdepr es si
oni nthebasewh erethes tapl
efolds.
Buti fyouonl ywantt omakes ur
et hatapileofpaper30pag esthi
ckwillit
intheopening
betweent h eheadandt hebas e,t
h eexactshapeofthehollowwon'
tmat ter
.Findingt
he
rig
h tlevelofdet ailisamatt erofmat chi
ngpurposetoef f
ort.

4Model
ingSof
twar
e
Therear es everalt ypesofpr ogramsav ailablet oh elpyouc reate3Dmodel s.
These
programsar eusual lyreferredt oasmodel ers.Th epersonus ingth eprog r
ami salso
somet imesr eferredt oast h emodel er,butt hatc anbec onf using(anddeh umanizi
ng )
.It
seemsmor epersonalt othinkoft h eper sonast hedes i
g ner(th i
sisal somor e
meani ngfulthant heg ener i
candunc ompl iment ar yuser
).Onei deat hatisc ommont o
mostmodel ersist hatt hedes i
gnerc anc reat eac opyofs omebas icform( suchasa
blockorpat ch
),andt henmodi fyittoc reat e
partoft hemodel .Thebas icf ormsar ecalledt h epr i
mitivesfort hatmodel er,andeac h
copyofi sani nstanc eoft hatpr imi t
ive.Soabl oc kyrobotmadeof20r ectangularblocks
contains20i nstanc esoft hepr imor dialbloc k.Thi sapproachi sremi niscentofPlato's
noti
onofi dealfor ms ;thepr ototypes hapei sthei deal,
andins tancesar epr acticalr ealizationsoft h
ati deal.

4.
1Int
erac
tiv
eModel
ers

Onec lassofmodel erisint eractive.

Typic allyth
edes i
gnersitsi nf r
ontofat erminal
withakey boardandmous e, andmani pul atess hapesont hescr een. Becaus ethei mage
isonlyt wo-dimens ional ,differents y
stemsh avedi f
ferentconvent i
onst oal l
owy outo
mani pulat eobjec tsin3D. Interac t
ivemodel erstypicall
ycontainav arietyofdes ignaids
tohel py ouac hieves omepr ecis
ioniny ourmodel :
Ac ommonex ampl eisto\ snap"point s
ont hemodelt oot h ermodelpoi ntsortoani nvi
s i
bleg r
id.Model sarebui ltby
int
er ac t
ivelysel ecting,creat i
ng, modifyi
ng ,andas sembl i
nginst ancesofpr i
mi t
ives.The
shapest ypic
allyh av e\handl es"thatallowy out ocont r
olthesh ape. Handl esmayc omein
thefor mofpoi nts,c urves,ori nteract
ivet ools.Oftenonedet er mi nesthes urface
charac terist
icswh ent hesh apei sfir
stc r
eat ed,butt hentext
ur esmaybel aidont he
surfacet omodi fyit.
4.
2Sc
ript
edModel
ers
Anot herclassofmodel erisreliesonani nputscripttodefinethemodel .Thisi
s
typicall
yat extthathasbeenc r eatedbyt hedes ig
nerus i
nganyc onventional
textedi t
or.Thescriptspeciesth esh apesinthemodelonebyone, i
dent i
fyi
ng
eac honebyi tsprimiti
ve,andth epar amet ersthatspeci
fyth ati
nstance.
Bec auses cri
ptedmodel ersallowy out ospec i
fyoperati
onsnumer icall
yr ather
thani nt
erac t
ivel
y,theyareidealf orv eryprecisemodeling.Forexampl e,it
mi g
h tbedi f
fic
ulttointeracti
velypl aceonebl ockata37: 5_angletoanot her,
buti t
'stri
vialtospec i
fywheny ouc ant ypethatang l
ein.

4.
3Ot
herModel
ers
Thereareotherwaystobuil
dmodels
,forex
amplebydi
git
izi
ngexi
sti
ng3D
str
uctures
,orbyanalyzi
ngphot
ogr
aphs.Wewil
lret
urnt
othesetec
hniques
neartheendofthispaper.

5Howar
emodel
sus
ed?
Therear emanyus esf or3Dmodel s .Wh i
let h erei sanobv i
ousdes iretoex perienc ean
objectvisual ly{h enc ec omput erg raph ics{t her ear eot heri mpor tantmot ivati
onst hat
formal ar g erandbr oaderc ont extf ors tudy ingobj ec tmodel s.Someoft hes eus es( with
exampl es )f or3Dmodel sar el istedbel ow. Not et h ateac honet endst oemph as ize
eit
h erthev isualors i
mul ationas pec toft hemodelasdi scus sedear li
er.
 Tov isual izedes igns :Howdoesapr oduc tl ookbef or ei tismanuf ac tured?
 Wh ati st hebes tc ombi nationofs hape, color ,lay out, etc .
?
 Toas sessappear anc e: Howdoess omeex istingenv ironmentl ooki fc hang esar e
madet oi t?Doesi tneedmor eorl es sl i
gh ting?Ar et hev i
s tasappr opr i
ate?
 Toobs erv epar tr elat ions hips: Howdot hing s_ ttog eth er?Wh atdoesanex pl oded
viewl ookl ike?Dopar tst ouc horc ollidet h ats h ouldbes epar ated?
 Toc hec kf eas ibility(ofmanuf actur e): Doest h eobj ec tmeett hedes i
gn
spec icat i
ons ?Cani ts atisfyc ons traint sons tr
es s ,l
oadi ng ,heatt rans fer,etc.?Can
itbeel ec ti
v elymac hined, cas t,ors cul pted?
 Todet er minec ost ,volume, ar ea, mac hiningt i
me, etc .
:Howmuc hmat erialis
needed?Howmuc hmat erialmus tber emov edbymac h ining ?Wh ati sthec r
os s-
sec tionals hapeofv ariousi nternals truct uresi nac ompl icatedas sembl y?
 Todet ermi nef ait hfulnes st oph ysicalph enomena: Howdoest hes urfaceoft h e
obj ec ti nterac twi thl i
g ht?Howdot h es urfac epr oper t i
es( smoot hnes s,roug hness,
wav es ,etc .
)dependont heg eomet ryoft heobj ect?Howc l osei nappear anc et o
\
r eal ity"c anweg et ?
 Toex ercisedi s playal gorith ms :Wh atl imit smi ghtweh aveont hec omput ations
ov ert h eobj ec tmodel s?Wh atdet ermi nesg raph icalal gor ithmc ompl ex i
ty?Wh at
 di
splayalgor
ith
msar eneededtoc reateimagesofapart
icul
arki
ndofmodel ?
 Toex pr
essartis
ticgoals
:Whatjuxtaposi
tionoft
herealandimag
inaryobjec
ts
wi
llprotr
ayanar ti
sts'
vis
ualmes sageandmood?Howmuc hcantheimaginat
ion
bestret
chedtovisuali
zeart
ifi
cialworlds
?

6Th
emaj
ori
ssues

Themaj ori ss uesi nvolv edi nobj ec tmodel ingi nc ludet hec omput at ionalc os tof
themodel ,itsel ec tivenes si nmodel ingt h edes iredph enomena, itsi mpl ement at i
on
compl ex i
t y,andt hemet h odsus edt oac qui re( orc reat e)dat at hatdes cribe
theobj ec tg eomet ry.Th ec omput at i
onalc os tofamodelmaybeg aug edi nt er msof
comput ers tor ag es pac e, obj ec tc ons tr uc tiont ime, displ ayt ime, ori nappl icat ionus e.
Forex ampl e, th es torag es pac eofamodelmaybebas edont hes pac er equi redt o
storeeac hpr imit iveandt h enumberofpr imi tivesneededt or epr es enti t. Th us
as pher ei sar elat ivelys i mpl epr imitiv e; col lec tionsofs ph er esg rowl inear ly
witht henumberofs pher esi nt hemodel . Pol yg onmodel smayr equi rev ar iabl e
lengths tor ag ef oran( ar bit r
ar ys ided)pol y gon, plusaddi tionall inks truc t ures
toor gani z eas etofpol yg onsi ntoac los eds urf ac e. I
nobj ec tc ons tr uc tiont ime, the
costsmaybas edonwh er et hedat ai sandh owt hedat ai sc apt ur ed. Cl ear lymor e
automat eddat ac apt ur ec anf av oroneobj ec tmodelov eranot hert hatr equi res
consider abl emanualef f or t.Onc et hedat adef ini ngt hemodeli sav ailabl e, var iat
ions
fromt heor ig i
nali nputs h ouldbepos s iblet oc leanupnoi sei ntroduc edbyaut omat ed
met hods ,var ys h apepar amet er sobt ai nings imi larbuti ndiv idualmember sofanobj ect
class, ors i
mpl y
correc tdes ig neri nputer ror s. Comput erg r aph ic sh asf requent l
yf oc us edont h e
algorit hmi cs i
deoft hedi s playt as k.Th er ear ec os tsas soc iat edwi tht hedi s play
algorit hmsus edt ov isual i zeeac ht ypeofobj ec tmodel .Vis ual izat ionc os tc anbeasl ow
ast hatofs impl el ineart rav er salandwi ref ramedi splayofpol yhedr alobj ec ts,oras
costlyandc ompl exasr ay -trac eds hadedr ender ing sofal gebr aics urf ac es .Th e
fundament alobs er vationi st h atdi spl ayc os tisdependentont heobj ec tmodel ;
comput erg raph icsh istor yi s, i
nal ar gepar t, thes tudyofpr oc edur est odi s playa
particul arc las sofobj ec tmodel s .Asi nmanyot h erc omput ers cienc eal g or ithms ,
onepay si nt imef ori nc reas ingg ener al i
tyoff unc t ion.Th ust h ec os tofr ay
tracingal soent ailst hepowert odi spl ayawi dev ar ietyofobj ec tmodelt ypes ;
attheot herendoft hes pec t
r um, as impl edept hbul eral gorit hmi sh ig hly
electiv ef orr eal-t i
medi spl aypr ov idedonei sr es t rictedt opol yhedr almodel s .
Finally ,thedi spl ayc os ti sal s odependentoni mag equal i
tyi s sues ; forex ampl e,
highr es olut ionorani mat edi mag eswi llr equi rebet t
ermodel s( andus ual lymor e
detail edmodel s ),visual iz ations t udiesmi ghtr equi repar ticul arobj ec ts ur fac e
proper ti
esf oradv anc edl ig htingmodel s,et c. Model sar eus edf orot h erpur pos es
besidesdi spl ay ,howev er ,andt her elat ivec os tsandc ompl ex ityoft hes eus edmus talso
bec ons i
der edwh ens el ec tingamodelt ype. Twoc ommonus esar et het rans for mationof
anobj ec tinpos ition, size, ors hape, andt h emeas urementofobj ec tg eomet r i
c
proper ti
ess uchasv olume. Th es eus esh av ec os t sas s ociat edwi tht h eg eomet r i
c
algori
th msbutt hes t
andar dc ompl ex itymeas ur esmus tbewei gh edag ai nstth eex pec ted
use.Forex ampl e, themor eef ficientc omput at i
onalal gorit hmi nt heg ener alc as emay
havec ons t ant swh i
chmakei tl esss at i
s fac toryi nt h eex pec t
edmaj or ityof
cases:amodelc ons i
s ti
ngofar bi trar ypol y gonswi llh av el esss canc onv ersion
effi
cienc yt h anonec ompos edent i
r elyoft r i
ang les .Al so, af requent lyr epeat ed
algori
th ms houl dbemademor eef f icientt h anoneus edonc e. Thusas impl er
buttheor et i
c allyc ostlieral g orithmmaybemor eef ficientwh enmul t ipliedby
expectedus e. Wh enpr og rammert imei sf ac tor edi n, thec ost sofag raph i
cal
algori
th mbec omeac ompl exmi xofs pec i
alc as ec ons ider at ions ,ov er allef ficienc y,
andpr og rammeref fi
c ienc y
. Cer tainl yonei mpor t anti s sueCer tainlyonei mpor t anti s s
ue
inobjec tmodel ingi sef fectivenes s: wh eth ert hemodelac t uallyr epr es ent sth edes ired
phenomenaandpr ovidesc omput at ionalanal og soft heappr opr iatev is ualorph y
s ical
propert ies .Poorv isualqual it ymaybe( thoug hisnoti dent icalt o)poormodel ing .Th us
thevisualef fec tiv enes sofas t
ic kg ur eofah umani sc ompr omi s edbyt h ez erot hic knes s
modelofbodys eg ment s .Ont heot herh and, af rac talmodelofamount ai n
maynott ellusany t
hingaboutt hef ormat ionpr oc es sofr ealmount ains ,but
thevisualel ec tandt hes t
at isticalpr oper tiesoft her es ultings ur fac emaybe
quiteconv inc ing .Th ec ompl ex ityofamodelmaybej udg edbet terasat rade- o_bet ween
thenumberofpr imitivesandt hei nh er entc ompl ex ity( wi t
h outbei ngc irc ular
inreas oni ng )oft h eindi v i
dualpr imi tives .Th usatoneendoft hes cal ear eobj ec t
swh i
c h
canbec h ar ac ter izedasamul ti
pl i
c ityofs impl ef or ms ;att heot herendar eobj ec ts
whichs eemi nh er ent lyc ompl ex .Forex ampl e, mac hinepar ts, building s ,andot h erman-
madeobj ec tss eemt obemos tlyamul ti
pl icityofs impl ef orms ,wh i
lebi ol og i
c aland
othernat ur als t ruc t
ur es( mount ains ,tex tur es )ar ei nh erent lyc ompl ex . Of tenobj ec t
datawi llbeav ai lablenat ural lyi nonef or my etneededi nanot h er.Th isr equi res
convers ionf romonef ormt oanot h er .Thei nputmaybef ormat tedbyanav ailabl ei nput
device( di gitizer , lasers canner ,tomog raph i
ci mag es ,manualdr awi ngi nput ,
et c .)
, but
therequi r ement soft heappl icationmaydi c t
at ec onv er siont oanot h erf ormf ormor e
conveni entdi s playorc omput ation. Forex ampl e, sur fac epoi ntdat amaybe
approx imat edbypol yg ons
orcurveds ur fac est oper mitv i
suals urfac er ender ingi nwor kstat i
onh ar dwar e;
construc tives ol idg eomet rymodel smi gh tbec onv er tedt ov ox elsf orv ol ume
comput at ion
.
7Model
sandRender
ing
Itisimpor tanttounder standthedi fferenc ebet weenmodel sandr enderi
ng .
Model sdes c ri
betheobj ectandi t
sat tributess uchass hapeorg eometry,color,
reecti
v i
ty,transmittance,surfacesmoot hnes s
, andt ext
ur e.Shadetreesmaybe
usedt oas cribedif
fer entvis
ualizational gorit
h mst odifferentpartsofamodel .
Butitist h
er enderi
ngal gori
thmi tselfwh ichtrans f
ormst hemodelt oasc r
een-
basedv iewf romag ivenc amer apos iti
on, projectsthe3Ddat aintodisplay
coordinates ,deter
mi nesthev is
iblepor tionsoft hes cene, andconvertstheobject
propertiesi ntopix
elv aluesint h
ec ont extofl ightsources ,at
mos pheri
ceffec t
s,
anti-
aliasing ,i
magec ompos iti
ng,andc olormodel s.
8Oper
ati
onsonmodel
s
Wewi lli gnor emos toft h er ender ingi s suesast heyar edi sc us sedel sewh er ei nt h i
s
Tut or ial .Wepr oc eedt odi sc us st heoper at ionsappr opr iat et oobj ec tmodel s .
Thes ei nc ludet rans format ions , chang eofdet ail,meas ur ement ,combi nat ion,
def or mat ion, anddi spl ay .Tr ans f
or mat ionsf or mt hebas isofmos tmodel ings ys t
ems .
Thes ei nc ludet h ewel l-knowng eomet rict r ans for mat ionsoft rans l
at ion, rot ation,
dilat ion( sc aling ), andr eec tion. Ch ang eofdet ailist h epr oc es sofaddi ngorr emov ing
infor mat iont opr oduc eanearg rainedr epr es ent at ion( forbet terv isualef fec t)or
simpl ifyot heroper at i
ons( forc omput at ionalef f cienc y).Us ual lys omei nt er pol at i
on
met h od
isus edt oaug mentdet ai l:forex ampl e, set tingc ur veds ur fac epat chest opoi ntdat at o
produc es moot hc ont i
nuousappr ox i
mat ionst ot heex pec t edor i
g inal( par tly -sampl ed)
dat a. Ont heot h erh and, av er ag ingors ubs ampl ingoft hedat amaybeus edt or educ e
det ai l
. Of tenmodel sar eg roupedi ntoh ier ar ch i
cs t ruc tur est os impl ifyt hei r
org ani z ation: amodelc ons is tsofs ub- model s ,eac hofwh ic hc ons istsofs ub- model s, etc.,
downt ot hel ev elofoneormor epr imi ti
v et ypes .Anal ter nat iv eh ier arc h i
cs truc turemay
bef or medbydef iningt h es ub- model st obet hes ameobj ec tast hepar entobj ec t,onl yat
al es serdeg reeofdet ai l.Dur ingmodelr ender i
ng ,th eappr opr iat elydet ailedmodeli s
selec tedf ordi s pl aybas edont hepr ojec t eds izeoft h eent iremodelont h edi s play
surf ac e. Dur ingani mat ion, ast h es izec hang es ,th emodel sf romt wodet ai llev elsmay
bev isual lybl endedwi theac hot her .Anumberofmeas ur ement smaybemadeonobj ec t
model s. Th et opol og ymaybec omput edt odet er mi newh et hert hemodeli st otal ly
connec tedorc ons i
stsofs epar at ec omponent s.Thepr es enc eoft opol og i
c alh ol esc an
bedet er mi ned. Th eov er allc ons istenc yoft heobj ec tmaybet est ed; forex ampl e, in
surf ac e- bas edmodel si ti sdes irabl et out il
iz et opol og i
c al lywel l-def inedEul er
oper at or st oi ns ur et h
ev alidi tyoft heobj ectmodel sbei ngi nt er ac tivel ydes igned. Ot her
usef ulorc ommonoper at ionsi nc ludepoi nt -
t o-poi ntdi stanc emeas urement ,
comput at ionofs urfac ear eaorv olume, anddet ermi nat ionofs ur fac et ang ent sor
nor mal sf ort her ender i
ngpr oc es s. Obj ec tmodel smayof t enbec ombi nedt of or mnew
andmor ec ompl exobj ec ts .Th ec ons truc tives olidg eomet rys ys t
emi si nf ac tbas edon
Bool ean
combi nat ionast h ef undament aloper at i
on. Th ec ombi nat ionoper at ionsar e
union( ev er yth ingt hati si nei th erobj ec t), i
nt er sect ion( ev er yth i
ngt h ati si n
bot hobj ec ts ), di _er enc e( ev ery thingt h ati si noneobj ec tbutnott heot h er) ,
ands ymmet ricdi _er enc e( ev er ythingt hati si nei therobj ec tbutnoti nbot h ).
Ot herc ombi nat ionoper at ionsar et hec utors li
cewh ichex pos est h ei nt eriort o
view, andt hec ov eroper ationwh ichde_ nest h er elat ions hipofoneobj ec tt o
anot h eri nt he_ nis hedi mag e. Th ec ov er ingi sani mag epr oper tywh ichdoes
notnec es sar ilys temf romar eal3Dmodelmot ivat i
on. Th er ear eanumberofway sin
wh i
c hanobj ec tmaybedef or med. Def or mat ionenr ic hest h es elec tionofobj ectmodel s
with outr es or tingt og ener at ionoft otal lynewdat as et s. Def or mat i
onoper ations
includes kew, s tret ch ,fol d, andper tur b( e. g.r andoml y, stoc h as tical ly,orf rac tal l
y )
.
Thedi s pl ayoper at ioni st hemos tv isibl eoper at iononanobj ec tmodel .
Therearemanyrenderi
ngal
gorit
hms,manyofwhi
charedis
cus
sedelsewher
e
inthi
sTutor
ial
:wir
e-fr
ame,vi
sibl
eli
ne,vi
sibl
esurf
ace,r
aycast
ing
,andray
trac
ing.

9Repr
esent
ati
ons
truc
tur
es
Aswer emar kedear lier,ther epresentations tr
uc t
uresus edf oranobj ec tmodel
maybeei therdec larat i
veorpr ocedural.
I nadec larati
ver epres entation,the
modeli sex plicit
lyembeddedi nas t
andar dcomput ationaldat as truc
t ure.Ina
procedur als ch
eme, themodeli sembeddedi ntoanyc onv enientc omput ational
procedur e,suchasaf ormula, implici
tequat i
on, orarbitr
ar ycode. Inthef ormer,
datai sretrievedbys earch,index i
ng,orpoi nterchasing;inthel atter,datai s
obtainedbyi nvokingapr ocedur ewithpas sedpar ametersors endingames sag
e
toanobj ec twh ic
ht henex ecut esar espons e.

1
0Modelc
las
si_
cat
ion

Themaj orc onc ernoft heres tofthisdiscus si

onisat ax onomyofmodel s.Wec l
as si
fy
model sintot wobr oadcateg or
ies:boundar ys c
hemesandv olumetri
cs chemes .
Ina
boundar yrepr esent ati
ont hesurfaceoft heobj ectisappr oximatedbyorpar t
itioned
int
o( non-over lappi ng
)0-,1-,or2-dimensionalpr imiti
ves .Wewillexami neinturn
representationsc onsist
ingofpr i
mi ti
vesf ormedbypoi nts,impli
citsurfacesoral gebrai
c
equations,pol ygons ,fr
actalsandg raftals,andc urveds urfacepatches.Inav olumet r
ic
representationt h e3Dv olumeoft h
eobj ec tisdecompos edinto(poss i
blyoverl
appi ng)
primiti
vev olumes .Underv olumetricschemeswedi sc
us sv oxel
s,oct
r ees,constructi
ve
solidgeomet r y,spec i
aliz
ed( sing
lepr i
mi tive)systems, potenti
alfunc t
ions,andpar ti
cle
systems.

1
0.1Sur
fac
eandboundar
ymodel
s
Thes i
mpl es
ts urf
acemodeli sj
us tacol
lect
ionof3Dpoi
nts.Theymaybe
organizedinas impl
elist
, ormaybemor ehigh
lyst
ruc
tur
edasc ontour
s, sl
ices
,
orsec t
ions.Surf
acesrepresentedbypoint
srequi
reafai
rlydens edist
ributi
onof
poi
nt sforac c
uratemodeling.

1
0.2I
mpl
ici
tsur
fac
es/Al
gebr
aicequat
ions
Theses ur
facesarede_nedast hesol
uti
onstoalgebr
aicformulas.
Onefamili
ar
classofsuchs ur
facesarethequadri
cs.
Ingeneral
,th
es urf
aceisthesol
uti
on
toanequat i
ons uchasF(x;y
;z)=0andnumer i
caltec
hniquesorsearc
hare
requir
edifFi snotanaly
tic
.

1
0.3Pol
ygons
Pol ygonal( polyhedr al)model sareoneoft hemos tcommonl yenc ountered
repr esentat i
onsi ncomput erg r
aph i
c s.
Themodel sar edefinedasnet wor ksofpol yg
ons
formi ng3Dpol yh edra. Eac hpolygon( pri
mitive)c ons i
stsofs omec onnectedv er-
tex ,
edg e,
andf ac es tructure.Therear eav arietyofdat as tr
uc turespos s ible.
Gr oupsofpol ygonsar es t
oredi nlis
t s,
tables ,
orl inkeds truc t
ur est ofacilitate
traver salofc onnec t
edf aces,theedg enet work, etc.Thepol ygonsar es i
zed, sh
aped,and
pos i
t i
oneds ot hatt heyc omplet elytil
et h
er equir eds urf
ac eats omer esol uti
on.Poly
gon
model sarer elat ivel
ys i
mpl etodef ine,mani pulat e,anddi splay. Theyaret he
commones tmodelpr oc essedbywor kstati
onh ardwar eandc ommer c i
algr aphics
sof t
war e.Ing ener alpol ygonsar ebes tatmodel ingobj ectsmeantt oh aveats urf
aces,
thoug hwi t
hal arg eenoug hnumberofpol ygonsqui teintricateandc omplexobj ec
tscan
ber epr esented. \Larg eenoug h\maymeanh undr edsoft housandsofpol y
g ons !

1
0.4Fr
act
alsandGr
aft
als
Alimi t
at ionoft hes ur facepoi ntandpol yhedralmodel sisthenec ess i
ty(us ually)of
explici
t lystoringal lth erequisitedat a. I
ftheobj ectst obemodel edar eh ighlyirreg ular
orc ompl ex ,s
uc hasmount ainsorc louds ,theirstatist
icals h
apepr oper tiesmaybeus ed
tobet t
eradv antag e. Fractalsandg raf talscreat esurfacesv i
aani mplicitmodelt h at
produc esdat awh enr eques ted.Thes ur facesaref requent l
ybas edonpoi ntorpol ygonal
model st hatar edec ompos edi nto_nerand_ nerdet ailasther equirement soft h
e
displaydi ctate.Frac talsus espec i
al\ random"pr ocessest oachieveanat ur all
y
mot iv
at eds tatisti
cali r
regularitysuchasdemons t
ratedi nrocks,mount ains ,c
louds ,
coas t
lines ,etc.Graf talsus edet er
mi nis ti
cproc essest omodelmor er epet iti
vepat t erns
suchast rees,leav es ,snowakes ,etc.Tr uef r
ac t
alsh aveas elf-si
milar i
typr oper t
y, wh ere
thef r
equenc ydis t
ribut i
onofs hapei ss upposedt obet hesamenomat t
erwh at
scaleisc hosenf orex aminingt heobjec t.Inprac t
ice,therear esever alvar iants
ont het ruef r
ac taldef i
nit
ion.

1
0.5Cur
veds
urf
aces
Sincepol ygonsar eg oodatr epr esent i
ngats urf
ac es ,cons ider ableef forth as
beenex pendeddet er mini ngmat hemat ic alformul ationsf ortr uec urveds ur f
ac es.
Althoug hi mplicits urfac esmayex hibittr ueandc ont i
nuousc ur vature,theneed
tos olveequat i
onsi sc os tlyandus uallyundes i
rabl e.Mos tc urv eds urfaceobj ec tmodels
aref or medbyoneormor epar amet r
icf unc t
ionsoft wov ariabl es( biv
ar i
atef unc t
ions)
.
Eac hc urveds urfac ei sc alledapat ch;pat chesmaybej oinedal ongt heirboundar yedges
intomor ec ompl exs urfac es.Us uallypat c hesaredef inedbyl owor derpoly nomi als
(typicallyc ubi
c s)g ivingt hepat c heas il
yc omput edmat h emat icalpr opertiess uchas
wel l
-def i
neds ur fac enor mal sandt angent s,andc omput ablec ontinuitycondi ti
ons
betweenedg e-adj ac entpat ches.Th esh apeofapat chisder ivedf romc ontrolpoi ntsor
tang entv ectors;th er ear ebot happr oximat i
ngandi nter polat ingt ypes.Thef ormert ake
theappr oximat es h apeoft h
ec ont rolver tices;thel attermus tpas st hrought hem. Ther
e
arenumer ousf ormul ationsofc urveds ur faces,inc luding :Bez ier ,
Her mi t
e, bicubic,B-
s
pli
ne,
Beta-spl
ine,
pol
ynomi
al,
rat
ionalpol
ynomi
al,
car
dinals
pli
nes
,compos
ite
s
pli
nes
,et
c .

1
0.6Vol
umeandCSGmodel
s
Thef i
rstv olumet ri
cmodelweex amineist hev ox elmodel .Her espacei scompl etely
_ll
edbyat esselationofcubesorpar al
lelopi pedsc al
ledv oxels(volumeel ement s).
Usuallyt herei sadens i
tyorothernumer i
c alv alueas s oci
atedwi theac hvoxel.Storinga
hig
hr esolut i
ont esselat
ionisexpensiveins pac ebuts implei ndatas tr
ucture( justa
large3Dar rayofv alues)
.Usuallysomes tor ageopt i
mi zati
ons chemesar er equiredf or
detailedwor k(1Kx1 Kx1 Kspaces).Specialt ec hni
quesar eneededt oc omput es urface
normal sands hadingtosuppressthebox i
nes soft her awv oxelprimitiv
e.Vox eldat ais
commonl yobt ainedinthemedicaldomai n; i
ti sh ig
h l
yr egardedfordi agnosticpur poses
asthe3Dmodeldoesnots peculat
eonaddi t i
onaldat a( s
aybys urface_ tt
ing)nor
suppres sanyoft heorig
inaldatahowev erc onv oluted.

1
0.7Oc
trees
Oc tr
eesar eoneoft hedat astructuresus edforvol
umet ri
cmodel sthattesselate
ag iv
en3Ds pace.Theoriginalvolume, sayacube, i
spar t
iti
onedi nto8c ubes
i
fi tisnon-empty.Recursiv
ely,eac hsub-cubeispartit
ionedwh enev ernonempty ,
until
somemi nimums i
zeel ementi sreac hed.Si
nceempt ycubesarenots ub-di
vided,the
storages paceeffi
ciencyisincreas ed.Themaj orus
eofoc treesappear stobean
i
ndex ings ch
emef orac c
essef fi
cienc yinalarge3Dar ray.

1
0.8Cons
truc
tiv
esol
idg
eomet
ry
Oneoft hemos tef f
ici
entandpower fulmodel ingtechniquesisconstruc t
ivesoli
d
geomet ry.Unl i
ket hev oxelandoc t
rees ch emes ,t
hereisnor equi
rementt o
regularlytess elatetheent ires pace.Mor eov er ,t
heprimitiv
eobjec t
sar enot
li
mit edto( unifor m)cubes ;rathertherear eanynumberofs impl
epr i
mi t
ivessuch
asc ube,sph ere, cyl
inder,cone, hal
f-spac e, et
c .Eachprimiti
veistrans formedor
defor medandpos iti
onedi ns pace.Combi nat ionsofpr i
mi t
ivesorofpr ev i
ously
combi nedobj ect sarec r
eat edbyt heBool eanoper at
ions.Anobjectth erefore
existsasat rees tructur
ewh ichis\
ev aluat ed"dur i
ngrenderingormeas urement.

1
0.9Spec
ial
ized(
sing
lepr
imi
ti
ve)s
yst
ems
Theg ener
alityofthecons t
ruc t
ivesoli
dg eometrymet h
od{wi t
hitsmul tipli
cityof
primiti
veobjectsandex pens iv
eands lowray-t
racingdisplaymet hod{i sfrequently
reducedtog aine_ci
enc yinmodelc onstruct
ion,avoidBool eancombi nationsotherthan
union,andincreasedisplays peed.
Th eideaistor est
rictprimit
ivestoonet ypet h
en
designmanipulati
onanddi splayal
g ori
thmstot akeadvant ageoft h
euni f
ormi t
yofthe
repres ent at ion. Vox elsmi gh tbec ons ider eds uc has pec ialc as e,wh eret hepr imitives
areal lcoor di nat eax i
sal ig nedandi nt eg r al lypos itionedc ubes .Others c hemesar e
poss ible, forex ampl e,us ingel lips oids ,cy linder s ,super quadr ic s,ors pheres .Elli
ps oids
havebeenus edt omodelc ar t
oon- like_ gur es .Th eyar eg oodf orel ong ated, symmet ric,
roundedobj ec ts.Unf ortunat ely,th edi s playal g or ithmi snear lyt hes ameast heg ener al
ray-trac i
ngpr oc es s.Oneunus ualappl ic at ionof( ov erlappi ng )el li
ps oidsi stomodel
terrain, clouds , shr ubber y ,andt rees .Int hes ec as es , t
hec onv incingv isuale_ ec tofs emi-
solidityisac hiev edbys tat isticalt rans par enc ywh ic hh asal owr ender ingc os t
.Cy li
nder s
haveal sobeenus edt omodelel ong at ed, s ymmet ricobj ec t
s .An
appl icationt hatus esc y l
inder spl uss ph er eswi thas pec ial-pur pos edi splayal -
gorithmi sbal lands t i
ckmol ecul armodel ing .Super quadr i
csar eamat hemat i
cal
gener al i
zat ionofs ph ereswh i
chi nc ludeani nt eres ti ngc lassofs hapeswi thinas ing l
e
framewor k:s ph er es ,ellips oids ,andobj ec tswh ic har bitrarilyc loselyl ookl ikepr isms ,
cyli
nder s,ands tar s .Simpl epar amet er sc ont rolt hes hapes ot hatdef ormat ionst hroug h
member soft h ec l assar es impl eandnat ur al . Super quadr icsar eus edt omodelman-
madeobj ec ts ,butwh enov er lappedc ang ivet h eappear anc eoff acesand_ gures .
Spher esasas ing lepr i
mi tivef or mani nt rig uingc las s.Th eyar et heonl ymodel ing
primi tivet hati si s otr opic :thati s,ident icali nappear anc ef romanyv i
ew. Mor eov er,
spher esh av eas i
mpl icityofg eomet ryt hatr ival st hatofs impl epoi nt s:j
us taddar adius.
Ther ear et womet hodsofr ender ings ph er es .Th eyc anbedr awnasf ully3Dobj ec tsin
wh i
chc ases omes canc onv er si
ont ricksc anbeus edt os impl ifyt heg ener ationof
succ es si
veobj ec tpoi nts( bas ical l
yag ener al iz at i
onofi nc rement alc i
rcledr awi ng
algor ithms ).Anal ter nat i
v edi spl aymet hodt reat st h es ph eresasi ft heywer e\scales "on
themodel edobj ec t; inth i
sc aseas ph er ei sr ender edasaats hadeddi sk.Wi t
h
sufficientdens it yofov er lappi ngs ph er es, th er es ulti sas moot hlys hadeds olidwh ich
model sc urv edv ol umesr ath erwel l.Anat ur al istich uman_ gur emaybedonet hisway .

1
0.1
0Pot
ent
ialf
unc
tions
Ani nterestingg ener aliz
ationoft hes pheremodeli st ocons idert hev olumeas
apot entialf unc tionwi thac enterandaf ieldfunc ti
ont hatdec reasesmonot onical l
y( by
anex ponent ialorpol ynomi alf unct i
on)f romt hec enterout war d.Th ereisno\ r
adi us "or
sizeoft hepot ent ialfunction; rat her,thes i
zeors urf
ac eisdet erminedbys et tinga
thresholdval uef ort hefield. Wh atmakest hi
smor einterestingi st hatpot ential
functionsac tl ikeener gysour ces:adj acentpot entialfunctionsh aveov erlappi ngf ields
andt her esul t
antv alueatapoi ntins pac eisinf actthes umoft hef i
eldsac t i
veatt hat
point.Thusadj ac entf i
eldsbl ends moot hly,
unl i
ket he\creas es "thatar eobt ainedwi th
xedr adiuss ph er es .
Rec ently ,directionaldependenc eands el ectivef i
elds ummat ion
acrossmodel sh av ebeenaddedt oc reate\soft"model sthatbl endwi tht hems el v
esbut
notwi th
othermodel edobj ectsinth eenv ir
onment .Potent i
alfunc ti
onswer eor i
ginallyus edt o
modelmol ec ules , si
nceat omsex hibitex actlythi
sf ormoff i
el dbeh av i
or .
Ot herus esf or
potent i
alfunc t ionsar er ealandi mag inaryor g
ani cformsi ncludi ngh umanandani mal
_gures .
1
0.1
1Par
tic
les
yst
ems
Gener alizings ph er esi nadi ffer entdi rec ti
on, par t i
c les ystemsr educ et hes ph er e
toaz er or adi us .Av olumei st her efor ec h arac t er izedbyas etof( us ual lymov ing )
par tic l
eswh ichi ndi rec tl
ydef inet h es pac e.Eac hpar ticleh asi tsownc olor , pat h ,h istory,
andl ifet ime. Th ei rmot ioni st ypi cal lyc ont rolledbypr obabi li
s ti
cal g orith ms .Par tic l
e
sys temsh av ebeenus edt omodel_ re, gas es, ex pl os i
ons , _rewor ks ,andg ras sy_ el ds .
11Wh eredomodel sc omef rom?Th erear enumer ousmet h odsf org ener at ing3Ds urface
orv ol umemodel s.Webr ieyex ami nemanualdi g itizati on, s emi -dir ec tdat aac qui sit i
on,
andal gor ithmi cmet hods . Manualdi gitiz ationi st h ec ons truc t i
onofamodelf rom3D
coor di nat edat ameas ur edandt ypedi nordi rec tlyt rac edf rompl ans ,model s, ort he
realt hing .Sev er alf or msofdi git izer sex i
s ttoai dt hist as k. Of t
endat af r omt woormor e
viewsi sus eds ot h at2Ddat amaybeex trapol at edi nt o3D. I
ft wov iewsar e
takenwi thpr oper lypos itionedandc alibr atedc amer as , manuals el ec tionof
cor res pondi ngpoi nt sont h et wov iewsc anbei nputt oa3Dr ec ons truc tionf or mul a.
Thisi st hebas i
coper at i
oni nph ot ogr ammet ry. Ifmanuali nputi st oot edi us ,ort he
modelt ooc ompl ex ,semi -
aut omat eddat aac qui sitionmet hodsmaybeus ed. Pr inc ipals
amongs uc hmet hodsar edi rec t3Dc oor di nat ei nputv ia3Ddi gitizer s, las ers canni ngto
obt aing riddedr ang e( henc edept h )infor mat ion, vox elr ec ons t
ruc tionf rommul tipl e
planar( tomog raph i
c )imag es , anddi rec t( video)i mag eanal ysis.Th el as tmet h odi ss t
ill
inr el ativ einf anc y, butc omput erv isiont ec hniquesc anbeus edt oi nt ellig ent ly
rec ons t ructv ar iouss impl eg eomet ricmodel sbas edonpol yh edr a, cur veds ur fac es ,
sph er es ,c yl
inder s, ands uper quadr ics .
Al g orith mi candi nter ac tivedes ig nmet h odsf orm
al ar gec las sofmodelbui ldingt ec hniques .Ty pi cal lyani nter ac tiveg raph ic aledi t oris
usedt obui ldamodelwi thint hepar ticul armodel i
ngs y stemr epr es ent ati on. For
ex ampl e, pol yhedr almodel sar epr e-emi nenti nc omput er-ai deddes ignappl icat ions ,
withc ur veds urfac esandc ons tr uc tives ol idg eomet rymodel sbec omi ngwi del yav ai l
abl e.
Ot heri nter ac ti
v emodeledi t orsh av ebeenc ons truc tedf ort h eot hermodelt ypes . Often,
anal g or i
thmi cmet hodc anas sistt hedes i
g npr oc es sbyenf or cingc ons tr ai ntsont he
modelpr imi tives .Forex ampl e, us ingEul eroper ator swi lli ns uret h atapol y hedr al
modeli st opol og ical l
ywel l-de_ ned( thoug hitmays tillbeoddi nag eomet r ics ens e).
Ac tualc ons trai ntoper ator smayi nsur et h atf eat ureswi th i
namodelar epar all el,
tang ent ,hor izont al ,vertic al, non- inter sec ting ,et c .Final ly,s omeal g or i
th mi cmet hods
canbeus edt oi nt er polat eorg ener atedat af romf or msmor eeas i
lyi nput ; forex ampl e,
byc r eat ingr ec ons truc t
eds ur fac esbet weenpar allelc ont ours lic esofs ur f acepoi nts,by
_ttingc ur veds ur fac est oc ont rolv er ticesort ang enc yc ons t raints ,orbyus ingf or c eor
ener gymet hodst oc aus emodelpi ec est os elf -as sembl e.

AGl
oss
aryf
orModel
ingandAni
mat
ion
Nor
manI
.Badl
er_Andr
ewS.
Glas
snery

Adaptiv
esampli
ngEval
uatinganimag ecoars
elyat_rst
,andt h
enmore
_nel
yinregi
onswi
thdetai
l.Caus
esf ortaki
ngmores amplesincl
ude
complexsh
apes,t
ext
ures,andshadows.Seesuper-
sampling.
Ali
asingArti
fac t
sinanimagegener
atedwh enweonl yus
ea_ ni
tenumber
ofequall
y-s
pac edsamples
,suchasatpixelcent
ers.Theart
ifactsi
nclude
bandi
ng,Moi r
_ee_ ec
ts,
jaggi
esonpolyg
onedg es,andregul
arpat t
ernsthat
don'
tbelong.

Algebrai
cSurfac
eAs urf
acede_nedbyanequation.
Thesur
facecont
ainsal
l
pointswhi
cheval
uatetoaparti
cularv
alueofth
eequation,
usual
lyzero.
Algebrai
csurf
acesareac l
assofi
mplici
tsur
faces.

AmbientTh
eamountofill
uminati
oninas cenewhic
hisass
umedtoc
ome
fr
omanydirecti
onandist
husindependentofthepres
enc
eofobj
ect
s,t
he
vi
ewerposi
tion,
oract
ualli
ghtsourc
esinth esc
ene.

AnimationTh
epr oc
essofcr
eati
ngandrecor
dingimageswhic
hc h
angeover
ti
me. Th
oughofteni
nter
pret
edasimply
ingonlyt
wo-di
mensionalimag
e
changes,i
tmaybeappli
edingener
altoanymodelorscenechangesin
th
r eedi
mensionsaswell
.

Anti
-aliasingTh
epr oces
sofr educing
,remov i
ng,
oravoi
ding
Theproc essofreduci
ng,removing
,orav oi
dingal
ias
ingarti
fac
ts.
Seealias i
ng.

Arti
culati
onTheabi
li
tyofonepartofanobj
ectt
omov
ewi
thr
espec
tto
another,
suchasahi
ngeorsli
dingjoi
nt.

BackgroundA_ x
edimagewh
ichisi
nter
pretedasexi
sti
ngin_ni
tel
yfarfr
om
theobserver
.Iti
sdi
splay
edwhennoobjec
ts ur
fac
efall
sintoanimage
sample.

BendDef
ormi
ngas
hapeal
ongal
ine.

BinarySpacePar tit
ioning(BSP)Amet hodfordivi
dingupa3Ds pace
withas etofplanes.Foranyobserverandplane,eachobjectint
he
envir
onmentei therstraddl
estheplane,isonthesames ideastheobser
ver
,
oront heotherside.Objectsonth
es ames i
deof( andinfrontof
)the
observerwillbl
oc kvi
sibil
it
yofobjectsontheotherside.

Biv
ariatepatchAvect
or-v
aluedfunct
ionoft wovari
ables
,typi
call
ycall
edu
andv.Theseparamet
ersgeneratesomel oc
al l
ytwo-
dimensi
onalsurf
ace
inspace.Oft
entheparametersareassumedt oli
eintherange[0,
1].
See
patch.

BlobAni
mpl
ici
tsur
fac
e.Bl
obsar
eus
ual
lyr
adi
all
ysy
mmet
ricGaus
sianf
unc
-
ti
ons.

Bool
eanoper
ati
onAnal
gebr
aicc
ombi
nat
ion(
uni
on,
int
ers
ect
ion,
ordi
_er
-
enc
e)oft
wog
eomet
rics
oli
ds.

B-RepSeeBoundar yrepresentati
on.
Boundar yrepres
entationTheg eomet r
yofanobjectasdescri
bedbyitss
ur-
face.Thesurfaceoftheobjec tisapproximatedbyorpar
tit
ionedint
o
(non-overl
apping)0-,
1 -
,or2-dimensionalpri
miti
ves(poi
nts,c
urves,
and
surfaces).

BSPSeeBi
nar
ySpac
ePar
tit
ioni
ng.

B-Spl
ineAtypeofspli
ne,wherethecurveisinuencedbythecontr
olpoints,
butgener
all
ydoesnotpas sth
roughthem. SeeSpli
nes.
BumpmappingAt ech
niqueforfaki
ng_ ne-lev
elgeometryonas ur
face,suc
h
asthebumpsonanor angeri
nd.Ratherthanac t
uallyc
reatemanylit
tle

piecesofs urf
acetomodelt h
ebumps ,atexturemapi splacedonthe
surfacewh i
chdesc
ribeshowthesurfac
enor malwouldv aryift
heshape
reallyh adbumps.Byshadingwit
htheseimag i
narynormal s
,thesur
face
appear stobebumpy .
Drawbackstothetechniquearerelatedtothe
factth atonlysmal
lbumpsc anbefakedt hi
sway :t
hesilhouett
eisnot
chang ed,bumpsdon'toccl
udeotherbumps ,andusuall
yt hebumpsdon'
t
casts hadows .

CADComput er -AidedDes ign.Th epr oces sofmanual lyc reati

ngg eomet ric
objectmodel swh i
c hmeetdes ireddes i
g nc r
iter i
a.
Catmul l
-RomSpl i
neAt ypeofs pl i
ne, wh eret hec urvei sinuenc edbyt h e
controlpoi nts ,butg eneral l
ydoesnotpas sth r
oug ht hem. SeeSpl i
nes .
Celani mat ionAni mat i
onpr oduc edbys tackingmul tiple2Ddr awings( c al
led
cels),eac hc ont ainingaf rag mentofas ceneorc harac t
eronat r
ans par ent
backg round. Inaph ysicalenv ironmentt h ecel sarel itandph otograph ed
tocreat eonef r ameofani mat ion. Celsar ebas icallyal abor -
savingdev ic
e.
Ifac harac t eriss tandings til
landt alking ,t
h enac eloft hec haracter
withoutamout hmi g htbepl acedov erabac kg round. Thenf oreac h
frameac elc ont ainingt heappr opriatemout himag eispl acedont opof
thes t
ac k, av oidi ngt henec es sitytor e-drawt heent irec haracterforev ery
frame. Of tent her ear es ever alindependentpl anesofi mag eswh i
chmay
bes uper i
mpos edi nas o-cal l
ed" twoandah alf-di
mens ional"animat ion.

Clippingpl anesBoundar yplanesintheworldcoor di

nat espacewhichdel imi
t
thepor t
ionoft heg eomet ri
cmodelwh i
chwillberender ed.Typi
call
yt op,
bottom, ands i
dec lippi
ngpl anesar ebasedont h
ewi ndowandc amera
pos i
ti
on. Thefrontandbac k(orh i
therandyon)clippingplanesareus ed
tos el
ectt heobjectdept hlimitstoremov eundesi
r edforeground(clos er
)
orbac kground( fur
th er
)obj ectsorsli
cetheobjectper pendicul
artot he
vi
ewt oex poseinnerdet ail
.
Compos iti
ngThepr ocessofc reatingani mag ebyc ombi ningmul tipleinde-
pendentimag es.Amat tedes cri
besh owmuc heac hpointofeac himag e
contr
ibutestothe_ nalpicture.Toani mateas paceship, f
orex ampl e,one
mightrenderthes paceship,andt henc ompos i
teitov erabac kgroundof
star
s.Thes hi
pwoul dc ompletelyreplac ethes tar_
eld,buts omeoft he
star
smi ghtbev i
siblethrought h
eex haus t
.Inani mation, t
h ebackg round
ands t
ationaryobjectsarer enderedonc e,wh i
lec hangingormov ingob-
ject
sar erenderedonaperf ramebas isandc ombi nedwi tht hestaticpart
oftheimag e.

Constr
aintsVal
ues,r
elati
onsh
ipsorc ondit
ionswhicharetomaintai
nedwhil
e
oth
erc h
angesarebeingmade.Forexample, aconst
raintcanh
oldal i
ne
hori
zontal
,keeptwoobjec
tsac ons
tantdistancefr
omoneanot herasthey
moving,
orc anc
auseoneobjecttobeat t
ractedtoanot h
er.

Construc
tiv
eSolidGeomet ry(
CSG)Amodelr epresentationbasedona
setofprimi
tiv
eg eometr
icshapeswhi
chmaybet ransformeds pati
all
yand
combinedinatree-s
truc
turedfash
ionbyBooleanoper at
ions.Theresul
t-
ingmodelisatreewiththepri
miti
vesastheleaves,Booleanoperati
ons
asthenon-l
eafnodes,andedgesastrans
formations.

ContourTypic
allyacurv
einspac
e. Commoncontoursar
eth
oseusedast
he
cros
s-secti
onforsweptsur
fac
es,andthesi
lhouett
eofsomeobj
ectfr
oma
part
icularpoi
ntofview.

Cont
rolPoi
ntOneofth
epoi
ntst
hati
nuenc
est
hes
hapeofas
pli
neorpat
ch.
Seespl
ines
,pat
ches
.

Conc
aveSeeConv
ex.

Cont inuit
yWh ent wopat chess hareanedg e,thatedgemaybemor eorl ess
noticeable.Thec ontinuit
yoft heedgeac rossth epatchesisamat hemat i-
calmeas ureofhows moot ht heedgeis .
Iftheedg eh as0-degreec ontinuit
y,
thent hereares i
mpl ynoh oles,buttheremaybeac rease.Iftheedg eh as
_rst-degreecontinuity,th
ent he_rstderivativ
eoft hes urf
ac eiscontinuous
acros stheedg e.Thisderivat i
vemaybemeas ur edeith
erpar amet ri
cal l
y
org eomet ri
cally
; th
ef ormeri sder i
vedf romt henaturalmat hemat i
calde-
scri
pt ionofthepat ch,andt helatt
erc omesf romt heappar ent(orv i
sible)
natur eofthepat ches .
Hig herdeg r
eesofc onti
nuityrefertoev er
-smoot her
blendsoft hes urfaceac rosst heedge.

ConvexApr opertyofashapeinanydimension.Ashapei sconv

exi fyou
canpickanytwopoi nt
sintheshape,andconnectthemwi thastrai
g ht
li
nethatneverg oesoutofth
es hape.I
n3D,as phereisconvex,buta
winegl
assisnot( pi
ckapointinthebowlandapoi ntonth ebase;t
h eli
ne
betweenthemi snotenti
rel
ywithinth
ewineglass).
As hapethatisnot
c
onv
exi
sconc
ave.

ConvexHul lThes mallestpol y

hedr ont
h atencl
osesag ivenobject
.Thecon-
vexhullforapolyg onc anbec reatedin2Dbydr awingthepol y
gonona
woodenboar d,andt hendr i
vinginanai lateachvertex.Str
etcharubber
bands oitencl
osesal lth
enai ls,andthenreleaseit;t
hes hapeformedby
therubberbandi st h
ec onvexh ull.

CSGSeeCons
truc
tiv
eSol
idGeomet
ry.

Databaseampl i
_cati
onTh enot
ionthats
imples etsofparameter sstoredi
n
adatas t
ructur
eordat abasemayexpandintoquitecomplexobj ectsby
th
eappl i
cati
onofs ui
tableproc
eduralprocess
es .Ex
amplesar ef r
ac t
aland
graf
talshapeswhichrequir
eonlyafewnumber sorsyntac
ticrulestoenable
th
eg enerat
ionofintr
icateanddetai
ledmodels .

Datast
ruct
ur eAcomput
eriz
edmet
hodofs
tor
ingi
nfor
mat
ions
uchasal
ist
,
ar
ray,
graph,net
work,
etc.

Decompositi
onBr eaki
nganobjec
tint
osmall
erors
implercomponent
s,t
ypi
-
call
ytofacil
it
atethealgor
ith
micproc
ess
ingoft
heori
ginalobj
ectbymor
e
generalpr
ocedures.

Dept hcueingTheproc essofr educingtheappar entbrig

ht nessofanobject
thefurtherawayfromt h evieweritispos i
tioned.Thi
softenenh ancesthe
perceptionofdepth.Th eideai sth
atonei sviewi
ngth esceneunderf oggy
conditi
ons,wherethec ontrastdiminisheswi t
hdistance.Ifthebackground
isblack,thi
sdiminis
h edc ontrasti
sequi valenttoreducedbr i
g h
tness;i
f
thebac kgr
oundisg r
ay ,itisreduceds aturati
onoft hecolor.

DifferenceABool eanoperationov ertwoobj ectswher et

her esulti
ngobjecti
s
thes etofpointsi
nt h
e_ r
stobjectth atarenotc ontainedinthes econd
objec t
.Diff
erenceisals
oc all
eds ubt r
action.
Th i
sisc ommonl yusedt o
dri
llh olesi
napar tbysubtract
ingac yl
inderfromt hepart.Sees ymmet r
ic
dif
fer ence.

Dif
us eLi
ghtwh
ichisreectedequallyinalldir
ecti
onsfromapoi ntonthe
surf
aceofanobject
.Thebr i
ghtnes
soft hesurfacedoesnot,
t h
erefore,
dependontheobserv
er'sposit
ionrelati
vet othesur
facepoint.

Digit
izati
onTh eprocessofbui
ldingag eometr
icmodelfromex i
sti
ngdraw-
ings,
ph y
sicalobjec
ts,ormodelsofobjects.
Theth
ree-dimensi
onalcoor-
dinatesares ens
ed,locat
ed,ty
ped, orother
wis
eobtainedfromthes our
ce.

Dil
ationAscali
nggeometr
ictr
ansfor
mation.
Scali
ngofapoi
ntcoor
dinate
(mult
ipli
cat
ionbyaconst
ant)canoccuri
ndependent
lyi
neachdi
mension.
Ditheri
ngTh eproces
soft radingspatialresolutionforenhancedi ntensi
tyor
col
orr esol
utionbydist
ributingfewers h
adesov eraregionlar
g ert h
an
asinglepixel.Th
eeyeper formst h
es patialintegrat
iontoproduc et he
additi
onalshadesorintens i
ties
.Caremus tbet akentoavoidpr oduc i
ng
dis
tracti
ngs pati
alpatt
er ns.

Dynami
csAnimati
onbythes
peci_
cati
onoffor
c esandtorquesappli
edt
o
masses
,rat
herth
anthei
rposi
ti
ons,v
eloc
iti
es,andaccelerati
ons.

EasingAr educt
ionint
h eaccel
erat
ionordecel
erat
ionofamoti
ont
opresent
asmoot her,morecont
inuousmovement.
Alsousedtoblendt
womot
ions
tog
et her.
Envi
r onment

Envir
onmentMapApic
tureofth
eenvir
onments
urr
oundi
ngamodel
.Iti
s
usuall
yusedt
oappr
oximater
eecti
onso_th
emodel.

Fi
eldofvi
ewTh esoli
dangl
eacrossthevi
ewingpy
rami
datt
heapex(
cent
er
ofpr
ojec
tion)wher
et h
ecamer
ai sloc
ated.

FractalAg eometri
cstructur
eh avi
ngthepr opertyt
h atit
sfrequencydist
ribu-
ti
onist hesamenomat terwhatmag ni
_cation(scal
ing )f
actorisappli
ed
toit
.Ex amplesty
picall
yc i
tedarec oast
linesandc l
ouds .I
npr act
ice
thetermi sals
oappl i
edt oanycurveors urfaceobtainedbythes t
ochas-
ti
c(random)per tur
bationofpoi nt
sonori nt h
es ur
fac eastheshapeis

subdiv
idedt o(
s ub-
)pix
elsizef
orrenderi
ng.Tech
nical
ly,
afrac
tali
sa
mathemat i
calconstr
ucti
onthati
sin_ni
te;anycomputermodel
sbasedon
t
h eseideasarejustapprox
imati
onsoftruefract
als
.

Geomet r
iceditorAninteract
iveprogr
amwh i
chpermitsthecr
eati
onand
modi_cati
onofs hapesbydisti
nctoperati
onsont hegeometr
icdat
astruc
-
turerepr
es ent
inganobject,f
orexampl e,i
nstanci
ngapr i
mit
ivesh
ape,
trans
formingit,c
ombiningitwit
hothers h
apes ,
etc.

Goniometricdi
agramTh i
sisadiagramthatgi
vestheint
ensit
yandcolorof
theenergyleav
ingapointli
ghtsour
c ef
oreverydi
rect
ioninwhic
hli
ght
canleavethesource.
Thediagrammaybeus edwit
happr ox
imati
onsfor
l
ightsourceswit
h_nit
es i
ze.

GraftalAnyc ur
ve,shape,orsurf
aceobtainedbyth
edet er
mi nis
t i
cors y
ntac
tic
perturbati
onofpointsonori nsomes t
arti
ngprimi
tiv
east hes hapeis
subdividedto(sub-
)pixelsi
zeforrenderi
ng.Oft
engraft
alsar eg ener
ated
byas hapeg r
ammar .
HierarchyAtreeinwh i
chnodestypical
lyrepresentobj
ectsandedg esrepr
e-
sentrelati
onships.
Forexample,ahumanbodyc ouldberepresentedbya
hier
archywiththetors
oatt hetop(orroot),and_vechil
drenrepresent
ing
thearms ,l
egs,andhead.Eacharmc oul
dbeas mallerhi
erarchyofupper
arm,lowerarm, andhand;th
eh andisah i
erarchyof_ve_ngers
, andso
on.

HitherClippingPlaneSeec l
ippingpl
anes.
Implici
tsurfaceAs urfacedes cri
bedbyanequati
on.Th
es ur
fac
ec ont
ainsal
l
pointswhichevaluatetoag i
vennumberwhenpluggedint
otheequati
on;
usuallyth
isvalueisz ero.Oftenthesol
uti
onsmustbefoundbynumerical
searchtechniques.
In-
betweeningSeei nterpolation.

Inter
pol ati
onThepr oces
soff ill
ingininformat i
onf r
omas etofv al
ues.Inter
-
polati
onc anbeus edtogener ateplausiblesur f
acepoi nt
sf orsparsedat aor
forcurvesorsurf
ac esdefinedwi thasmal lsetofc ontr
olpoi nt
s.Iti
sals oa
waytog enerat
epos iti
onsf orobject
sbet weenkeyf r
amesi nanani mation.
Inter
sec ti
onABool eanoper ati
onov ertwoobj ectswherether esult
ingobject
isthesetofpointswh i
chlieinbot hobjectssimultaneously.

KeyframesCellani
mati
onimagesdrawnbyth
eanimatorwh
ichar
eint
erpo-
lat
edalgori
thmic
all
ybythecomput
ertogenerat
einbet
weenimag
esatthe
desi
redframerate.

Keyparamet
ersValuesofmot i
onorot
heratt
ributeparamet
erswhi
chare
speci
fi
edbytheani
mat or
.Inbet
weenparamet
erv aluesar
etypi
cal
lygen-
erat
edbyspl
inecur
v es
.

Kinematic
sMot i
onspecif
icati
onbyposit
ion,
v el
ocit
y,andaccel
erati
on,rat
her
thanbyforces
.Usual
lyappli
edtothemanipulati
onofmec hani
smsinvol
vingar
tic
ulat
ed
partsandvari
ousjoi
ntsandconst
raint
sbetweent h
epar t
s.

Levelsur
fac
eAllsol
uti
onstotheimpl
ici
tequat
ionF(
x)=kf
ors
omec
on-
stantkandmul
ti
dimensi
onalv
ec t
orx

Li
ghts
our
ceTh
ecol
loqui
alnamef
oral
umi
nai
re.

LuminaireAnobjectwhi
c hemit
slig
h tonitsown,evenifnootherl
igh
tis
str
iki
ngi tf
romelsewhere.
Localcoordi
natesyst
emAr ef
erencef r
amepos it
ionedwit
hinoronanob-
jec
ttof aci
li
tateit
sde_niti
on,manipulati
on,andplacementinascene.
Thelocalcoordi
natesys
temmov esal ongwit
ht h
eobj ec
t.

Mat
teAtwo-di
mensi
onalregi
on(notnec
essari
lyc
onnec
ted)ofani
mag
ewh
ich
i
ndi
cat
eswhereonei
mag eistobecomposi
tedwit
hanot
her.
Mip-MapApre-proc
essedfor
mofat ext
urewhi
chcani
ncreaserender
ing
speed.
\Mi
p"st
andsf ormul
tuminparvo,
Lati
nfor\
manyth
ingsin
asmallpl
ace"
.Simil
artoth
es umtabl
e.

OctreeAdatas t
ructur
eforas pat
ialpart
iti
onwhichdiv
idess
paceint
oeig
ht
equalcubes(
octants).
Eachcubeisrecur
sivel
ysubdiv
idedasneededto
thelev
elofdetailofthemodelorthepix
els i
ze.

Or
thogonalAs
ynony
mforper
pendi
cul
ar.

Or t
h ogonalproj
ectionThev i
ewc r
eat edbypl
ac i
ngthecamer aatani n_
nite
distancef r
omt hescene.Theviewdirecti
oncreatesaviewingparall
elop-
ipedwh osec r
oss-s
ec t
ionisdeterminedbythewindow.Objectsofthe
sames iz
eatdi _
erentdepthsfromt hecameraappearth esames iz
ei nthe
imag e.

Parti
clesyst
emAr epr
esentati
onwh i
chcons i
stsof(l
arge)setsofindiv
idual
point
sunderalgori
thmicorstochas
ticcontrol.
Thepointshavecolor,a
tr
ajector
y,andah i
stor
y.Theymaybec reated,destr
oyed,orspli
t.The
hi
storymaybeus edinas i
ngleimagetoc r
eateac urvi
li
nearshape, ori
n

ananimationtocreat
eamov ingpoint
.
PatchAlmostany3Dcurvedsurfac
emaybet h
oughtofasapatch.Ty
pic
all
y
th
eyares mallbi
tsofsur
facethatarecreatedbyames hofc
ontrolpoi
nts
,
whic
hinuencethepatchi
nthes amewayt h
attheyi
nuenceas pl
ine.

Patchesar
eusuall
yconnect
edt
opr
eser
vedi
_er
entt
ypesofc
ont
inui
ty.
See
cont
rolpoi
nts
,conti
nui
ty.

PathsThevalueofaparametr
icfunc
tionov
ertime.Ty
pic
allyusedtodescri
be
th
es pat
ialpositi
onofanobj
ect,
butalsoappl
icabl
etoitsori
entat
ion,or
evenatt
ribut
ev alues
.

Perspecti
veprojec t
ionTh eviewc reatedbyplac i
ngthecamer aata_nit
e
posi
tion(thecenterofpr ojecti
on)wi t
h i
nthescene.Thec amerafor
ms
theapexofav iewingpyrami dwh osecross-sec
tionisdeterminedbythe
vi
ewdi r
ectionandt hewindowor_ eldofv i
ew.Objectsofthesames i
ze

Transmitt
anceApropertyofas ur
fac
edes cr
ibi
ngthefrac
tionoflig
htthat
willpasscompl
etel
ythroughit
.Thusatrans
mi t
tanceof0isopaque,1is
invis
ibl
e,andval
uesinbetweeno_ervar
iousdegreesoftr
anslucency.

Transl
ati
onAg eometr
ict
rans
for
mati
onwh i
chc
aus
espoi
ntst
obemov
edby
const
antdis
placement
sineachc
oor
dinat
e.
UnionABooleanoper
ati
onovertwoobjec
tswh
eret
her
esul
ti
ngobj
ecti
sth
e
setofpoi
ntswhi
chli
eineit
herobjec
t.

Viewdir
ecti
onThedir
ecti
oni nwh
icht
hecamerai
spoi
nti
ng.I
tist
heper
-
pendi
culart
othevi
ewingplaneonwhi
cht
h ei
mageisf
ormed.

Vi
ewupdi r
ecti
onTheworl
ddir
ect
ionwhi
chwhenv
iewedbyt
hec
amer
a
wi
llappeartobev
erti
cal(
up)int
heimage.

Vi
ewportThesub-ar
eaofth
escreenorimagef
ormat
iondev
icei
ntowh
icha
gr
aphi
calrenderi
ngorot
herdr
awingisplac
ed.

Vi
sibl
esur
fac
eTh
atpar
tofanobj
ectwh
ichi
svi
sibl
efr
omag
ivenpoi
ntof
vi
ew.

Vis
iblelineTheprocessofrender
ingordrawi
ngonl
ylinearfeat
ures(
lines)
toshowatl eas
tonlythoseedgesorcont
ourli
nesofsurfac
eswh i
char
ei n
factvi
sibletothecamera.

VoxelAthr
ee-dimensi
onalvolumeel
ement
,usual
lyacubeorbox.I
thasa
val
ue(ordensit
y)andsixsidesandr
epr
esent
sa_ ni
tesi
zedpointi
na
reg
ulardecomposit
ionofspace.

WindowAr ectang
ularsubregionoft
hev iewingplanewh i
c hde_ nesth
et op,
bottom,andsidecli
ppingplanesrel
ati
vet ot h
ec amer apos i
tion.
Wire-f
rameTh edrawingofamodelbytr acinglinearfeaturess uchasedg es
orcontourli
neswithoutat
tempt i
ngtoremov einv i
sibl
eorh iddenpar t
s.

Worl
dc oordi
nat
esThearbi
trar
ycoordi
nat
eref
erenc
efr
amei
nwh
ichobj
ect
modelsareposi
ti
onedt
oc r
eateas c
ene.

Worl
dproj
ect
ionAspeci
albackg
roundimagebuil
tfromth
eobjec
tsi
nth
e
sc
eneit
sel
fwhic
hisusedtocr
eateconv
inci
ngreecti
onsonobj
ect
s.

YonCl
ippi
ngPl
aneSeec
lippi
ngpl
anes
.

Z-buzzerAr enderi
ngtechni
queinwh ichobject
sarescan-convert
edtopix
el
dataanddept hval
uesandt henins
ertedintoapixelarrayandadept h
array.Wheretheobject'
spixeldepthislessthanth
atoft h
ec urr
ently
storedpixel/depthpai
r,thenewdat aiswritt
enin,r
eplacingtheold.

AdvantagesoftheZ-bu_erar
ethattheobj
ectsneednotbeproc
essedin
anyspecialor
der,t
hei magecanbeincr
ementall
yupdatedwi
thaddit
ional
obj
ectsatanyt i
me,andt hes
cenecancontai
nanunlimit
ednumberof
obj
ects
Keywor
ds
3DRec
ons
truc
tion:
I
nc omput erv
isi
onandc omputergraphics
,3Dr ec onst
ructi
onist heproc
essof
captur
ingthes hapeandappearanceofr ealobject
s.Thispr oc
esscanbeac c
omplis
hed
[1
]
eit
herbyac ti
veorpas si
vemethods. I
fthemodeli sallowedt ochangeitsshapein
[2]
ti
me, t
hisisreferredtoasnon-
rigi
dors patio-
temporalr econstr
ucti
on.

Pr
obl
ems
tat
ementandbas
ics
Theapproac
hofus ingBinocul
arstereovis
iont oac qui
reobj
ect'
s3Dg eometri
c
[1
6]
inf
ormati
onisont h
ebas isofvi
sualdispari
ty. Thefoll
owi
ngpict
ureprovi
desas i
mpl
e
schemati
cdiagramofh ori
zont
all
ys ig
htedBinoc ularSter
eoVis
ion,wherebisthe
baseli
nebetweenpr oj
ecti
vecentersoftwocamer as.

Geomet
ryofas
ter
eos
copi
csy
stem

Theor i
g i
noft hec amer a'scoordi nates ystemi satt heopticalcenteroft hecamer a'
s
lensass h
owni nth efig ure.Actual ly,thecamer a'simag eplanei sbeh i
ndt heoptical
centeroft hecamer a'sl ens.Howev er,tosimpl i
fyt hecalc
ul ati
on,imag esaredr awni n
frontoft heopticalc ent erofthel ensbyf .Theu- axisandv -
ax i
soft heimag e'
s
coordinates ys
temO1 uvar eint h es amedi recti
onwi thx-axisandy -
ax i
soft hec amera's
coordinates ys
temr es pec t
ively.Th eor i
ginoft heimag e'
sc oordinatesys t
emi slocated
ont heintersecti
onofi mag i
ngpl aneandt heopt i
calax i
s.Suppos esuc hworldPpoi nt
wh os
ec orrespondi ngimag epoi nt sar eP1(u1 ,
v1)andP2( u2,v2)respec tiv
elyont heleft
andr ig
h timagepl ane. As sumet woc amer asar ei nthesamepl ane,theny -
coordinates
P1 &P2ofandar ei dent ical,i.
e.
,.V1 =V2Ac cordingt otri
gonomet ryrelations,
u1
=f(
xp/z
p)

u2=
f(x
b-b/z
p)

v
1=v
2=f
(yp/z
p)

where(xp,
yp,
zp)arecoordi
nat
esofPint h
el eftc
amer a'
scoordi
nat
es y
stem,i
sFfoc
al
lengt
hofthecamera.
Visualdi
spar
ityi
sdef i
nedast h
ediffer
enceinimagepoi
nt
locat
ionofacert
ainworl
dpointac
quiredbyt wocameras,

d=
u1-
u2=
f(b/z
p)

bas
edonwhic
ht h
ecoordinat
esofPc anbeworkedout
.Th
erefor
e,onc
ethe
c
oordi
nat
esofimagepointsi
sknown,besi
dest
heparameter
softwocamer
as,
the3D
c
oordi
nat
eofthepointc
anbedet er
mined.

Xp=
bu1
/d

Yp=
bv1
/d

Zp=
bf/d

Th
e3Dr
econs
truc
tionc
ons
ist
soft
hef
oll
owi
ngs
ect
ions
:

Def
ormabl
eModel
 Adef ormabl emodeli sag eomet ricobjectwhoseshapec anc hang
eoverti
me.
 Th edef ormat i
onbeh av i
orofadef ormablemodelisg overnedbyvari
ati
onal
pr inc i
ples( VPs)and/orpar t
ialdi f
ferent
ialequati
ons( PDEs )
.
 Cons istsofdef ormabl ecurves,def or
mabl esurf
aces,anddef ormabl
esoli
ds.
Ver
y
s ucc es sf
ulforvisualc omputing.
 Combi nesknowl edg efrommat hemat i
cs,physi
csandmec hani
cs.
 I nh eriteds moothness, morerobus ttonoise.
 Bui lt‐
indy namicbeh av i
or,
wel lsuitedfortimevary
ingPh enomena.
 I mpl icitrepresentation
A0X2+A1 Y2+ A2Z2+ A3XY+ A4XZ+ A5YZ+A6 X+A7Y+A8Z+A9=0
 Sph er e
X2+Y2+Z2+ R2= 0
 El lips oid
X2/A2+Y2/B2+ Z2/A2-1=0

Gr
adi
entv
ect
orf
low(
GVF)
,
 [
1][2]
Ac omputerv i
sionf ramewor ki
ntroducedbyCh enyangXuandJerryL.Pri
nce, isthe
vectorfi
eldthati spr oducedbyapr ocessthatsmoothsanddiff
usesani nputv ec tor
fi
eld.Iti
sus uallyus edt ocreateav ectorf
ieldf r
omimag est
hatpointstoobj ec tedges
fromadi st
anc e.Itiswi delyusedi nimageanal ys
isandcomputervis
ionappl icat i
onsfor
objecttr
acking, shaper ecog ni
ti
on, s
egmentation,andedgedetecti
on.Inpartic ular,i
tis
commonl yusedi nc onjunctionwithactivec
ont ourmodel.

Vi
sualh
ull
Av i
sualhul
lisag eomet r
icent
itycreatedbyshape-f
rom- si
lhouett
e3D
reconst
ructi
ontechni
quei ntr
oducedbyA. Laurenti
ni.
Thist
ec hni
queas sumest he
for
eg r
oundobjectinanimag ecanbes eparatedfromt h
ebac kground.Underthis
assumption,t
heorigi
nalimag ecanbethres h
oldedintoaforeground/backgroundbinary
image,whichwecallasil
h ouett
eimage.Th eforeg
roundmas k,knownasas i
lhouett
e,is
the2Dpr oject
ionoft hec orr
es pondi ng3Df oregroundobj ect.Alongwi ththecamer a
viewingpar amet er s
, t
h esil
houet tedef i
nesabac k-projectedgener aliz
edc onet hat
containst h
eac tualobj ect.
Thisc onei scaledas
l ilhouettecone. Theupperr ight
thumbnai lshowst wos uchconespr oducedf romt wos il
houetteimag estakenfr om
[
1]
diff
er entviewpoi nts.Theintersec t
ionoft het woc onesi scall
edav isualhull
, wh i
chis
aboundi ngg eomet ryoftheac tual3Dobj ect(seet hebot t
omr i
ghtth umbnail)
. Whenth e
recons t
ructedg eomet ryisonlyus edf orrenderingf romadi ff
er entv i
ewpoint,the
[
2]
implicitreconstruc ti
ont ogetherwi threnderingc anbedoneus i
ngg raph i
cshardwar e.
VI
SUALHULLOFASPEAREVI
EW

VI
SUALHULLOFHUMAN:
3Drec
ons t
ruc
tionfr
ommultipleimagesi
sthecr
eati
onofth
ree-
dimensi
onalmodels
fr
omas etofi
mag es
.Iti
sth
er evers
eproces
sofobt
aini
ng2Dimagesfr
om3Ds cenes.

Theess enc eofani mag eisapr oject i

onfroma3Ds c eneont oa2Dpl ane, duringwhich
processt hedept hislost.Th e3Dpoi ntcorrespondingt oas pecif
ici magepoi ntis
constr
ai nedt obeont helineofs i
g ht.
Fromas ingleimag e,itisimpos sibl
et odet ermine
whichpoi ntont h
islinec orrespondst otheimag epoint.Iftwoi mag esareav ailable,t
hen
thepos i
tionofa3Dpoi ntc anbef oundast h eintersecti
onoft hetwopr ojectionrays.
Thisproc essisreferredt oast ri
ang ulati
on.Thekeyf orthisproc essisther elati
ons
betweenmul ti
plev i
ewswh ichconveyt heinformat i
ont hatcor r
es pondingset sofpoints
mus tcont ai
ns omes tructureandt hatt hi
sst r
uc t
urei srelatedt othepos esandt he
cali
brationoft hec amer a.

Inrecentdec ades, thereisani mportantdemandf or3Dc ontentf orc omputergraphics,

vir
tualreal i
tyandc ommuni cat i
on,tr
iggeri
ngac h
ang einemph as isforth
er equir
ement s.
Manyex istingsystemsf orc ons tr
ucti
ng3Dmodel sar ebuiltarounds peciali
zed
hardware( e.g.
stereor igs)res ult
inginah i
ghcos t,whichcannots atisf
ytherequirement
ofitsnewappl ications.Thisg aps ti
mulatestheus eofdi git
alimag i
ngfacil
iti
es(likea
[2]
camer a).
Anear l
ymet hodwaspr oposedbyTomas iandKanade. Theyusedanaf f
ine
factori
zationappr oac htoex tract3Df romimag ess equences.Howev er,t
h eassumpt i
on
ofor t
hographicproj ecti
onisas i
gnif
icantli
mitationoft hi
ss ystem.
Tex
tur
es

Textur
esareimageswhichar
eappli
edtot
he3Dobject
’sf
aces,
e.g
.togiv
ecolourt
oth
e
surf
ace.Moredet
ailsaboutt
ext
uresar
edesc
ribedint
hechapt
eraboutComputer
Graphic
s.

UV-
Unwr
appi
ng

Sinceimag etext
uresar etwo-
di mensi
onaldata,amappi ngisneededf romt h
e3Dr oom
withtheobjecttothe2Ds pacewi tht
het ext
ure.I
tdefineswhichtexturepixel,als
o
called“texel
”,i
sdis pl
ayedont hemes h’
sface.Theprocessofcreatingthismappi ngis
calledUV-unwrapping.I
ntuit
ively,
themes hiscutopenats omeedg ess othatits3D
shapec anbeunfoldedi nt
oaf latfor
m. Theimagetextureisappliedtot h
isflat
representati
on.

Mor etec hni

cal,eachv ertexofthemes hi smappedt oa2Dv ertexinthes pac eoft h
e
textureandwi thnewc onnecti
vit
ydata.Th e2Dt exturespac eisalsocalledUV- space,
thetwoax esarelabel ledwit
hUandVi nsteadofXandY. Thisdist
inguishest heUV-
coordinateswh i
char eas soci
atedwi t
hat exturewithpointsi nthe3Dr oom. UV-
coordinatesrang efrom0t o1withthepoi nt(0,0)att heleftbottomoft hei mag eand( 1
,1)
atthet opr i
ght.
Th i
sal someanst hatUV- pointsar eindependentoft hetex ture
resolutionandar esc aledwiththeimag e.Wh enUV- unwrappinganobj ect, i
tsfac esare
lai
doutont hetex t
ur e.GoodUV- unwrappi ng
smi ni
miset heamountofs t
r etchi
ng .
Stretchi
ngoc cursifthes hapeofaf acei sdistortedint h
eUV- s
pace.Espec iall
yc urved
surfaceslikeas pherear ediffi
culttoUV- unwr appingwithouts t
retch
edt ex tures.

AUVmapwh
ichpr
ojec
tsac
ubet
oat
ext
ure.
Anot
herUVmapwh
ichals
oproj
ect
sthecubetoat
ex t
ure.I
nth
eUVmap,t
hefacesdo
noth
aveth
esameshapeasi
n3D.Thi
sleadst
oastretch
edtex
tur
eont
hecube.

Tomi nimi zethes tretchedt extures,edgesoft hemes hcanbemar kedass eams.This

meanst hatthef aceswh ichar enotc onnectedint hetexturespace.Subs equently
,a
vertexint he3Ds pac ec anbemappedt omul t
ipleverticesinth
eUV- s
pac eifthe
adjacentedg esar es eams .Thepr oblemofs eamsi sthattheycreatedis conti
nuiti
esin
thetext ureont hef aces .Fori nstance,ifabrickwalltextureisappliedt oac ube,itcan
happent hatthe“ li
nes ”ofbr i
ckwor kdonotmat chupatt heseams .Henc e,a3Dar tist
hast odomanualc orr ectionsbymov ingtheunwr appedf aceonthet extureuntilthe
seamsmat ch.Forth i
sr eas on, seamsar eoftenpl acedinareaswh ic
har el essvi
sibleto
thev i
rtualc amer a.
Ex
ampl
esh
owi
ngt
ext
uredi
scont
inui
ti
esatas
eam.
Thedi
sconti
nui
ti
esc aus
edbytheseamc anbefi
xedbymanuall
yal
igni
ngt
headj
acent
fac
esintheUVlayoutorbypl
acingseamsonles
svisi
bleedg
es.

Tex
tur
eRes
our
ces
Therear evariousonl
inereposi
tori
eswh ichoff
erphototext
uresforvary
ingpurposes
andwi t
hdi f
ferentquali
ti
es,pr
icesandl ic
ences.Alar
geamountoft heseonl
ine
resourcesoffersPBR-compati
bletextureswhichmeansthatitdoesnotonl
yprovide
imaget ext
uresbutal s
ootherdatalikenor malmaps,metal
licmaps,
roughness/smoot h
nessmapsandambi entoccl
usionmaps.

Conc
lus
ion
Iti
sunreasonabl
et otalkofac oncl
usionwhent
h emethodsdi
scuss
edint h
istextfor
documenti
ngoutstandingsi
tesar e,
ass h
ownhere,inaperi
odofrapidevolut
ion.The
equipment
,wheth
erl aserorphoto,isbecomi
ngli
g h
ter,
moreeffec
tive,
lessexpensive
andeas iertomanipulat
e.Thes oftwarei
sbecomingmoreautomaticandcanbetur
ned
toinmor eandmor ecompl exsi
t uati
ons.I
tcanbeusedtodayonsiteswhi
chwoul
dh av
e
unti
lr ecent
lybeenthoughtofasi mpossi
bletoworkwi
th.However
, nev
er-
endi
ng
progressinthefi
eldofc omputers ci
encemeansthatt
histypeofcomplex
ityc
annowbe
proces s
ed.

Nev erth
el es s,theper formanc eandt hemoder ni
tyoft hes emethodsh avenotc hanged
theunder ly i
ngv al ues :
in[ SAI92] ,wear er emi ndedt hatt hesurvey“mus testabl i
s hthe
shapesoft heobj ec t,t
h eirpr opor t
ionsanddi mensions ,andc l
earlydesc r
ibet he
multipler elat i
onsbet weent hev ari
ousel ement s”.
Sai nt-Aubinhighl
ightsthef actt h
at
“anobs erv ationoft her epr esent ati
oni sins uffic
ientfores t
ablish
ingwh atiss urveyed
andwh ati si nterpol atedand, evenmor es eriously,whati sac c
urateori naccurat einthe
element ss ur veyed” .Thisc allf orcaution,g ivenin1 992att h
edawnoft hef i
rstlas er
scannersandwel lbef or eth eaut omat ionofph otogrammet ry,
ishighl
yr elevantt oday.
Datapr oduc edus ingth ec urrentmet hods, nowav ailablet othoseouts i
det hes elected
groupofs urv eyors ,areonl yus efulifac c
ompani edbymet adatawhichqual i
fiesi tina
met hodicalway ,