TELKOM
NIKA
, Vol. 11, No. 6, June 20
13, pp. 3115
~ 312
2
e-ISSN: 2087
-278X
3115
Re
cei
v
ed
Jan
uary 18, 201
3
;
Revi
sed Ma
rch 3
0
, 2013;
Acce
pted April 16, 2013
Hybrid Collision Culling by Bounding Volumes
Manipulation in Massive Rigid Body Simulation
Norhaid
a
Mo
hd Suaib
*1
, Abdullah Bade
2
, Dz
ulkifli Mohamad
3
1,3
F
a
culty
of Co
mputin
g, Unive
r
siti T
e
knologi
Mala
ysi
a
, 813
1
0
UT
M Skudai, Johor, Mala
ys
i
a
.
2
School of Scie
nce & T
e
chnol
og
y, Univ
ersiti
Ma
la
ysi
a
Sab
a
h
, Kota Kina
bal
u, Sabah, Ma
la
ysi
a
.
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: haid
a
@utm.
m
y
*
, a
bad
e0
8
@
yah
oo.com, dzulkifl
i@utm.
m
y
A
b
st
r
a
ct
Collis
io
n dete
c
tion is an i
m
p
o
rtant aspect i
n
ma
ny
real-ti
m
e si
mulati
on
envir
on
me
nts. Due to its
nature of high Com
p
utation i
n
volv
ed,
collis
i
on
detection c
an c
ontribute t
o
the bottleneck on t
he syst
em
involv
in
g l
a
rg
e
nu
mb
er of
inter
a
cting
ob
jects.
This p
a
p
e
r foc
u
ses
on fi
nd
ing
opti
ons to
effic
i
ently c
u
l
l
aw
ay
obj
ect pa
irs th
at are
not
like
l
y to col
l
i
de
in
larg
e-
scal
e
dy
na
mic r
i
gi
d-b
o
d
y si
mu
lati
ons
invo
lvi
ng
n-b
o
dy
obj
ects. T
he
ma
in
id
ea
is to
perfor
m
ti
me
cr
itical c
o
mp
utin
g
co
ncept by ma
ni
pul
at
io
n of
pote
n
tial
b
oun
din
g
volu
me tech
niq
ues. In or
der t
o
take a
d
va
nta
ge of a
fa
st col
lisio
n test a
nd
a more
accurat
e
resu
lt, a hybr
i
d
collis
io
n cull
in
g appr
oac
h ba
sed on sp
her
e
-
or-Dops w
a
s
used. Base
d on in
itial res
u
lt
s, this approa
ch
show
s a potent
ial a
dapt
ation t
o
a massive ri
g
i
d bo
dy si
mul
a
tion.
Ke
y
w
ords
: col
lisio
n cul
lin
g, collis
io
n detecti
on, rigi
d bo
dy, bou
ndi
ng vo
lu
me, b
oun
dary r
epres
entati
o
n
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Colli
sion det
ection i
s
usually an integral
p
a
rt to many real
-time and in
teractiv
e
appli
c
ation
s
. There are m
any appli
c
ati
ons th
at re
lie
s on
colli
sion
detectio
n
su
ch a
s
in the f
i
eld
of virtual reali
t
y, computer
game
s
, anim
a
tion, training
, enginee
ring
and medi
cal
simulatio
n
. With
these va
st areas of
collisi
on dete
c
tion
appli
c
atio
n
s
, it is unde
rsta
ndabl
e that collision
dete
c
tion
research i
s
still an
active
research area.
Despit
e the advances in techno
logy, coll
ision
detection
still contri
bute
s
to the syste
m
bottlene
cks for mo
re th
an thirty years [1].
In many re
al-time and i
n
te
ractive
appli
c
ations
,
colli
si
on dete
c
tion
is a p
r
e
-
requi
site for
reali
s
tic
syste
m
re
sp
on
se.
A simpl
e
ex
ample i
s
i
n
volving a
com
p
uter g
ene
rate
d bou
nci
ng
b
a
ll.
Colli
sion
bet
wee
n
the
bal
l and
any
ob
stacl
e
s ne
ed
to be
d
e
tected so that
the
system
can
gene
rate a
p
p
rop
r
iate
re
spon
se li
ke t
he directio
n
and mo
me
ntum after i
m
pact.
Coll
ision
detection and collisi
on resp
onse usually are carried
out successively [2].
The m
a
in
purpose
of colli
sion detection is
to ensure that ther
e
will not be any two
obje
c
ts that o
c
cupy the
sa
me space at t
he same ti
me
. This i
s
eq
ui
valent to ho
w obje
c
ts b
eha
ve
in the real
worl
d–two o
b
ject
s cann
o
t
fit into
the sam
e
spa
c
e at exactly
the sa
me t
i
me
.
Dep
endin
g
o
n
the level
of accu
ra
cy
that is
need
ed o
n
the
a
pplication, co
llision
dete
c
tion
pro
c
e
ss
can
either si
mply flag interse
c
tion(s)
bet
we
en obje
c
ts, o
r
it might produ
ce detail
ed
repo
rt on the
event like time of conta
c
t, point of
contact and int
e
rpe
netration
depth. The firs
t
type of detection ca
n be
carri
ed out in
a very sh
ort
time but doe
s not yield m
u
ch i
n
form
ation
about the
coll
ision. Ap
plication like co
mputer
gam
e
s
that requi
re
s ap
proxim
ate but fa
st coll
ision
detectio
n
u
s
u
a
lly ado
pts t
hese type
s o
f
colli
sion
de
tection
app
ro
ach
e
s.
On
the oth
e
r han
d,
more computation n
eeds t
o
be
carried
out in
order t
o
get
detailed co
llisi
on i
n
formation needed
for a mo
re
seriou
s a
ppli
c
ation a
s
in m
edical sim
u
la
tor. The
r
efo
r
e there is
ge
nerally trade
-offs
betwe
en sp
e
ed and a
c
cu
racy in collisi
on detectio
n
.
At the same time, issue
s
like real-ti
m
e
perfo
rman
ce,
efficiency an
d robu
stne
ss need to be a
d
d
re
ssed [1].
Perha
p
s the
most ad
apted
appro
a
ch in
tr
ading
spe
e
d
and accu
ra
cy was introdu
ced by
Hub
bard a
s
can
be
se
en
in ma
ny re
search
pap
ers co
ncerni
ng
colli
sion
dete
c
tion. It works
based on th
e idea of time-critical com
puting: co
lli
si
on detectio
n
and re
sp
on
se for intera
ctive
grap
hics ap
pl
ication
s
ca
n be improve
d
by usi
ng a two-ph
ase pro
c
ess:
broa
d-ph
ase a
nd na
rrow-
pha
se [3]. Collision
cullin
g, which in e
s
sen
c
e deal
s with quickly removin
g
obj
ect pairs that
are
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No. 6, June 20
13 : 3115 – 3
122
3116
less likely to collide i
s
very much rel
a
ted to
the
broa
d-pha
se
level, as wi
ll be discu
ssed
thorou
ghly in the next secti
on. Th
is will
be the focu
s
of this pape
r.
The
re
st of this p
ape
r
will
be o
r
g
anize
d a
s
follows
:
related work
, partic
u
larly involving
boun
ding
vol
u
me i
n
b
r
oa
d
-
pha
se
collisi
on
cullin
g
will
be
discu
s
se
d in S
e
ctio
n I
I
followe
d by
th
e
idea of hybri
d
collisi
on cul
ling in Section III.
Experimental layout
will be outlined in Section IV.
Section V de
als with the result an
d discussion, while
Section VI co
nclu
de
s this p
aper.
2. Related Work
As mentio
ne
d in p
r
eviou
s
sectio
n, a
two-pha
se
approa
ch i
s
usu
a
lly ad
opted in
minimizi
ng
collision
test
s due to
the
nature of
collision
dete
c
tion that is
comp
utationa
ll
y
intensive.
G
enerally, the
broa
d-
pha
se colli
sion dete
c
tion
a
c
ts
m
o
re
li
ke
a filter that id
entifies
only pairs of
objects that
are m
o
re li
kely to co
llide. This
step is responsi
ble to cull
away
unrel
ated
pai
rs i
n
the whole
collisi
on
detection pr
ocess, and
thus
synonymous with
the
t
e
rm
‘colli
sion
culli
ng’. These identified pai
rs are t
hen fe
d to the narro
w-p
h
a
s
e colli
sion d
e
tectio
n for
further collisi
on te
sts
(ple
ase
refe
r to
[4] for a
n
el
aborate di
scussion
on
co
llision
dete
c
tion
approa
che
s
a
nd appli
c
atio
ns).
Different approaches
can
be implem
ent
ed duri
ng col
lision culling
pr
ocess
such as the
brute fo
rce
(all-pai
r te
st),
swee
p an
d
prun
e
(S
aP)
and
hierarchi
c
al
ha
sh tabl
e [4]. Meth
od
impleme
n
ted
in this
study i
s
ba
se
d o
n
b
oundi
ng
volu
me techniq
u
e
whi
c
h i
s
ve
ry conventio
n
a
l
with the first
approa
ch. Convention
a
l b
oundi
ng
volu
mes that a
r
e
comm
only used in the bro
a
d
-
phase level will be present
ed next.
2.1. Conv
entional Boundi
ng Volumes
Boundin
g
vol
u
me i
s
o
ne
of the mo
st
comm
only u
s
ed b
r
oa
d-p
h
a
s
e
colli
sion
d
e
tection
approa
ch in
simulatio
n
s i
n
volving n-bo
dy obje
c
ts
.
Based
on it
s
popul
arity, we will n
e
xt ou
tline
some of the p
opula
r
bou
ndi
ng volume
s.
Boundin
g
vol
u
me
algo
rith
ms
en
comp
a
s
ses techniq
ues like b
o
u
nding
sphe
re
, Axis-
aligne
d boun
ding box (AA
BB), oriented
boundin
g
bo
x (OBB) and
discrete o
r
ien
t
ation polytop
es
(
k
-
D
O
P
s)
(s
ee
F
i
g
u
r
e
1)
.
O
r
ie
n
t
ed
-D
ops
(o
r-
Do
ps),
a co
mbinatio
n of OBB an
d k-Do
ps wa
s
introdu
ce
d to overcome the
update cost
of k-Dop
s
[5] (sh
o
wn as Fi
gure 2
)
.
Figure 1. Con
v
entional Bou
nding Volu
me
s
Figure 2. Orie
nted-Dop
s
o
r
also
kno
w
n a
s
or-
Dop
s
2.2. Boundin
g
Volumes: Simplicit
y
Versus
Accu
r
a
c
y
Based
on previous re
se
a
r
ch,
th
ese
b
oundi
ng
volu
mes had
it
s own advanta
ges and
disa
dvantag
e
s
. In short,
a simple b
o
undin
g
volu
me req
u
ire
s
less co
mputa
t
ion in terms o
f
con
s
tru
c
tion,
update
s
an
d tests. Th
erefo
r
e it
is
expecte
d tha
t
simulation
s or appli
c
ati
ons
utilizing
simpl
e
bounding volumes
c
an
give higher frame rates but
with the
cost of more fal
s
e
positive te
st outcom
e
. A false
po
sitive colli
si
on
occu
rre
d when
col
lision te
st ba
sed on
bou
ndi
ng
volume gives a positive result (c
olli
sion detected) but the actual
object di
d not
collide. This
is
due to la
rg
e
empty co
rne
r
s cau
s
ed
by
simple
bou
nd
ing volume.
On the
other hand, a
mo
re
accurate
bou
nding
volume
re
quires mo
re
com
putati
on a
n
d
mo
re
time, resulting in
le
ss fra
m
e
rates.
Th
e p
r
eviou
s
b
oun
ding volu
me
s ca
n be
r
oug
hly arrang
ed
according
to
simpli
city versu
s
accuracy a
s
in Figure 3.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Hybrid
Colli
si
on Culli
ng by
Bounding Vol
u
m
e
s Manipulation... (Norhaida Mohd Suaib)
3117
Figure 3. Simplicity versu
s
Accu
ra
cy
2.3. Time-Cri
tical Collision Detection
The ide
a
of
time-critical
collision
dete
c
ti
on, as int
r
o
duced by
Hu
bbard [3], ha
s be
en
adapte
d
by
many researche
r
s.
Altho
ugh it i
s
quit
e
impo
ssible
to incl
ude
all
refe
ren
c
e
s
i
n
this
category, some
of highly
relat
ed work
will be
discussed here.
Most
of the approaches use
boun
ding vol
u
me hie
r
arch
y (BVH) in o
ne way o
r
an
other. One o
f
them utilize
d
the tree cal
l
ed
averag
e-distri
bution tree o
r
ADB-tre
e
[6] that
is a combinatio
n o
f
bounding v
o
lume hie
r
a
r
chy
(BVH) with
an estimated
probability of
colli
sion
occurrence
to reduce co
llisi
o
n tests. S
phere-
tree
was used for time-cri
tical
collisi
o
n detection
by
at least two resear
chers;
sphere-tree
on
redu
ce
d mo
d
e
l for d
e
form
able o
b
je
cts
[7] and
sph
e
r
e-t
r
ee
with
close
s
t featu
r
e map
s
(CF
M
s)
applie
d to refinable
collisi
o
n respon
se [8
, 9].
An AABB-tree with
re
duced defo
r
m
able mod
e
l was
also
used for self-colli
sion cullin
g [10]. An event-based
scheduli
ng that adapt
ively prioriti
zi
ng
colli
sion te
sts and p
e
rfo
r
m
s
collisi
on te
sts at
differe
nt time interv
al wa
s int
r
od
uce
d
by Com
i
ng
and Staa
dt [
11, 12],
whil
e a BV
H
wa
s u
s
e
d
with
certificates th
at indi
cate a
b
se
nce
of se
lf-
colli
sion [10].
2.4. Total Co
st Be
nchmar
king
Total co
st
be
nchm
arkin
g
(Equation 1)
for colli
sion
d
e
tection
that
wa
s p
r
op
ose
d
by [13]
will be used as a basis:
T = Nu x
Cu
+ Nv
x
Cv
+ Np x
Cp
+ Co
(1)
Whe
r
e:
T: total c
o
s
t
func
tion for interferenc
e
det
ec
tion,
Nv: numbe
r o
f
boundin
g
volume pai
r ove
r
lap test
s
Cv: cost of testing a pai
r of boundi
ng vol
u
mes fo
r overlap
Np: is the nu
mber p
r
imitive pairs tested
for interferen
ce,
Cp: co
st of testing a pai
r of primitives for interfere
n
ce,
Nu: numb
e
r o
f
boundin
g
volumes u
pdate
d
,
Cu: co
st of averag
e bou
ndi
ng volume up
date,
Co: indi
cate
s co
st for one ti
me pro
c
e
s
sin
g
, where ne
cessary
Ho
wever, not
all param
ete
r
s mu
st be i
n
volved;
they depen
d on the problem
and the type
of
colli
sion invol
v
ed.
3. H
y
brid Co
llision Cullin
g Method
The m
a
in
pu
rpo
s
e
of colli
sion
culling i
s
to
red
u
ce
colli
sion te
st
s by i
dentifying a
nd
cullin
g away unne
ce
ssary
pairs. The b
a
si
c colli
si
o
n
detectio
n
pro
c
e
ss
whi
c
h u
s
ually is
ca
rri
ed
out in
a t
w
o-pha
se
pro
c
e
s
s (bro
ad
-ph
a
se
an
d n
a
rrow-pha
se
) i
n
spires a
two
-
pha
se
colli
si
on
cullin
g pro
c
e
s
s in ord
e
r to a
c
hieve a
simp
le, fast and re
liable collisio
n detectio
n
.
The culling
p
r
ocess
will b
e
ba
sed
on t
he bo
undi
n
g
volume ap
proach. Since
there a
r
e
quite a num
b
e
r of pop
ular
boun
ding vol
u
me, a que
sti
on will ne
ed
s to be add
re
ss is the type
o
f
boun
ding
volumes that i
s
suitabl
e for th
at pu
rpo
s
e.
If there
are
su
itable
can
d
id
ates, i
s
the
r
e
any
way that perf
o
rmance of the
culling process can be i
m
proved?
There are two main issue
s
that need to be co
n
s
id
e
r
ed in the qu
estion ab
ove. On the
one h
and, a
simple
and fa
st colli
sio
n
te
st is n
eed
ed
but on the
other h
and, it a
l
so n
eed
s to
be
Si
m
p
le
r B
V
M
o
re
ac
c
u
ra
t
e
BV
Evaluation Warning : The document was created with Spire.PDF for Python.
TEL
K
3118
relia
b
their
to le
a
into t
fals
e
4. E
x
st
ag
e
one
a
the e
a se
t
and
O
orien
and
e
4.1.
H
Duo
C
Micr
o
usin
g
4.2.
E
suit
a
b
part
w
4.2.1
the
c
use
d
1346
T
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
next
e
differ
obje
c
capt
u
K
OM
NIKA
V
b
le
. T
he c
a
n
simpl
e
repr
e
a
d to false c
o
he possi
bilit
y
d
e
t
ec
tion
.
x
perimental
For th
e
m
e
,
we are ta
r
a
not
her, like
mpty sp
ace
Since m
o
t
of
simp
le
e
O
BB) whic
h
ted
D
ops
(
O
e
ac
h
ob
je
c
t
w
H
ard
w
a
r
e U
s
A
ll of the
C
P
U
E7
400
o
soft Windo
w
g
the sa
me h
E
xperiment
s
There ar
e
b
l
e
boun
din
g
w
as
to expe
r
. Part I
Four set
s
c
ommonl
y u
s
to bo
un
d 1
5
ver
t
ic
es as
T
able 1. List
List
of
1
[
h
2
[t
o
3
4
5
6
7
8
9
0
1
2
3
4
5
[p
e
[upper
[uppe
[low
e
r
R
[low
e
r
[le
f
[rig
h
[uppe
r
[upper
R
[low
e
r
[low
e
r
R
[lef
t
[Rig
h
Average
e
xperim
ents
ent velo
city
c
ts
. Figure
u
ring th
e
av
e
V
ol. 11, No.
6
n
didate
s
f
o
r
s
e
se
ntation
a
n
o
lli
sion de
te
c
y
of manipul
a
La
y
out
m
om
ent we
a
r
getin
g a
sui
in a crowd
e
is
not too la
r
o
vem
ent of
V
e
xperim
ent
s
w
ar
e
most
c
r-
Dop
s
). At
t
w
as
as
sume
d
s
ed
exp
e
rime
nt
s
@2.
8
0G
Hz
w
s
XP P
r
of
e
ardware co
n
s
e
m
a
inly tw
o
g
volumes t
h
r
iment on th
e
s
of
experim
s
e
d
b
oun
din
g
5
different
o
b
listed in T
a
b
of 3D Obje
c
3D O
b
jects
F
iles
h
ead.tri]
o
rso.tri]
e
lvis.tr
i]
RightL
eg.t
ri]
rLeft
Leg.tri]
R
ightLeg.t
ri]
r
LeftLeg
.tri]
f
tFoot.tri]
h
tFoot.t
ri]
r
LeftA
rm.tri
]
R
ightA
rm.t
ri]
r
LeftArm.
t
ri]
R
ightArm.t
ri]
t
Hand.tri]
h
tHand.t
ri]
b
o
u
nding v
o
, all obje
c
ts
assi
gne
d
a
s
4 sho
w
s
on
e
r
a
ge
co
nstr
u
6
, June 20
1
3
s
imp
l
e a
nd
f
n
d
t
e
st
s.
H
o
c
tion d
ue to
a
ting a m
o
r
e
a
re ex
perim
e
tabl
e BV to
e
d environ
m
e
r
ge
) but co
ul
d
V
E inh
abitan
t
w
here we t
e
c
ommonly f
o
t
his stage, a
d
to repres
e
n
s
outli
ned h
e
processo
r
w
e
ssio
nal S
P
3
n
figuration a
s
o
exp
e
rime
n
t
h
at could b
e
e
perfo
rma
n
c
ents were
i
n
g
volumes
–
b
je
ct
s.
Obj
e
le 1.
c
ts Us
ed
Total Verte
x
542
841
74
706
706
706
706
122
122
1346
1346
1138
1138
589
589
o
lum
e
const
r
are ra
ndo
ml
s
in rigid b
o
d
e of
the ex
p
u
c
t
ion time, t
3
: 3115 – 3
1
f
ast collisi
o
n
o
wev
e
r, it i
s
large empty
e
ac
cu
r
a
te
b
o
nting o
n
diff
e
be u
s
ed
on
e
nt. Theref
o
d
su
ppo
rt re
a
t
s coul
d
not
b
e
st four
diffe
o
und in the
singl
e boun
d
n
t an avata
r
.
e
re w
e
re
do
n
w
ith 2 GB
R
A
3
. Different
s
will be expl
t
s involved:
e
used fo
r t
h
c
e of the p
r
o
p
n
volved in t
h
–
sph
e
r
e
,
A
A
cts used
a
r
e
Figur
e
r
uc
tion time
s
y trans
form
e
d
y motion, s
o
p
erim
ents c
o
here a
r
e th
r
e
1
22
t
e
st
a
r
e
sp
h
ty
p
i
c
a
l
f
o
r
t
h
cor
n
e
r
s.
T
h
o
undi
ng vol
u
e
rent BVs
to
many avat
a
o
re, a BV th
a
a
l-time ap
pli
c
b
e anticipat
e
rent types o
l
iterature, pl
d
ing volume
n
e on a CPU
A
M onbo
ard,
types of b
o
ained in the
the first
par
t
h
e hybrid co
l
p
osed hybri
d
h
e firs
t part.
A
BB, OBB,
k
e
based o
n
p
e
4. Sample
e
s
for thes
e o
e
d insid
e
a v
o
that there
w
o
nd
uc
te
d
o
n
e
e other
exp
e
-
h
ere and
AA
h
e
s
e ‘
s
impl
e’
h
e
r
efore, it i
s
u
me so that i
be
us
ed
o
n
a
rs and
prob
a
t fits
an av
a
c
ation is ver
y
e
d mo
st of t
h
f BVs
(sphe
us
a
n
o
t
he
r
t
wa
s as
sig
n
e
ru
nnin
g
on
a
and NVidia
o
undin
g
vol
u
nex
t
sub-
se
c
t
w
a
s
de
sig
n
l
li
sion cullin
g
d
method.
All four ex
p
k
-Do
p
s an
d
o
p
o
i
nt-clo
ud,
r
e
xper
iment
u
sph
e
re
s
bj
ect
s
wil
l
b
e
olume (box)
w
ill be collis
n
bo
undin
g
s
eri
m
ents th
a
-
ISSN: 2087
A
BB, mainly
d
b
o
undi
ng v
o
s
d
e
sirable t
o
t coul
d red
u
c
the avatar.
A
ab
ly very cl
o
a
tar (me
anin
y
much pref
e
h
e time, we
d
re,
AABB, k
t
ype of BV
e
d
for each
o
a
n Intel®
C
o
Ge
F
o
rc
e 2
1
u
mes were
t
c
ti
o
n
.
n
ed to ide
n
t
i
g
,
and the s
e
p
eri
m
ents in
v
o
r-
D
o
ps tha
t
r
an
ging fro
m
u
s
i
ng
b
o
un
d
i
e
logged.
F
with differe
n
i
o
ns am
ong
s
pheres.
B
e
a
t were c
o
nd
u
-278X
d
ue
to
o
lume
o
loo
k
c
e the
A
t this
o
se to
g that
e
rred.
d
esign
-Do
p
s
cal
l
ed
o
bjec
t,
o
re™2
0 with
t
e
s
ted
i
fy the
e
co
nd
v
o
l
ved
t
wer
e
74
to
i
ng
or the
n
t with
th
e
s
e
e
sides
u
ct
e
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Hybrid
Colli
si
on Culli
ng by
Bounding Vol
u
m
e
s Manipulation... (Norhaida Mohd Suaib)
3119
The first experiment was
conduc
ted to test collisi
o
n time using different types of
boundi
ng
volumes
whil
e the se
con
d
experime
n
t wa
s co
ndu
ct
ed to test the
average
upd
ate time. Results
from the
s
e
experim
ents
were fed i
n
to the To
tal
Co
st Ben
c
h
m
arking fu
nction. The l
a
st
experim
ent was im
pleme
n
ted to
roug
hly sho
w
th
e ov
erall
perfo
rm
ance ba
se
d
on fra
m
es-p
e
r
-
se
con
d
(FPS) count
s.
4.2.2. Part
II
The second
part of this t
e
st deal
s
with fu
rthe
r exp
e
rime
nts on
the pro
p
o
s
ed
hybrid
colli
sion
culli
ng meth
od.
As m
ention
ed e
a
rlie
r, th
e hybri
d
colli
sion
culling i
s
a
two
-
ph
a
s
e
colli
sion culli
ng
process d
e
sig
ned
to achieve
a simp
l
e
, fast and
re
liable
collisi
o
n dete
c
tion.
The
scene involve
s
a co
mbinati
on of ran
dom
ly moving
objects a
nd som
e
static obj
ects insid
e
a bo
x
(se
e
Fig
u
re
5
a
and
5b for
a sam
p
le of l
oade
d obje
c
t
s
). Simila
r to
experim
ents
in Part I, obje
c
ts
are
allo
wed
to go throug
h othe
r obj
e
c
ts
but w
ill
simply ch
ange
dire
ction
on
ce they
hit the
boun
dary of t
he box to e
n
s
ure that n
o
obje
c
t will
wa
nder off the b
ound
arie
s.
Once collisi
o
n is
detecte
d, a chang
e of col
our will visua
lly i
ndicate collision b
e
tween two obj
e
c
ts an
d relat
e
d
information will be logged.
While
experi
m
ents in P
a
rt
I only involved rel
a
tively small number of
obje
c
ts, the p
r
opo
se
d met
hod will b
e
tested ag
ain
s
t conve
n
tional
techni
que
s in
a massive ri
gid
body simul
a
tion und
erg
o
in
g rigid
-
bo
dy simulation.
Figure 5a. Sample Experim
ent usin
g 50
Obje
ct
s wit
h
o
u
t
B
V
Figure 5b. Sample Experim
ent usin
g 50
Obje
cts with
Boundin
g
Sp
here
5. Results a
nd Analy
s
is
5.1. Part I
C
o
ns
tr
uc
tio
n
T
i
me
. Con
s
truction
times
for ea
ch
obj
ects u
s
ed
(a
s p
r
eviou
s
ly
listed in
Table 1
)
wa
s
recorded a
n
d
repeate
d
10
0 times. Average con
s
tru
c
tion time is sh
own in Fig
u
re
6.
Figure 6. Effects of Sele
cting Differe
nt Switchi
ng un
de
r Dynami
c
Co
ndition
Num
ber of detected collisions
. Initially,
accumul
a
ted
numbe
r of col
lision
s
dete
c
ted wa
s
recorded for
100, 200, 30
0
,
400 and 50
0
frames a
s
in
dicate
d in Ta
ble 2 and Fig
u
re 7.
0
1000
Sphere
AABB
O
BB
k
‐
Dops
orDops
Av
e
r
a
g
e
c
o
ns
truction
time
Evaluation Warning : The document was created with Spire.PDF for Python.
TEL
K
3120
Ta
Bou
n
g
Volu
Sp
h
AA
O
B
k-
D
o
or-
D
mea
s
inclu
d
sin
c
e
the v
a
Whe
r
Re
s
u
belo
w
K
OM
NIKA
V
ble 2. Nu
m
b
n
din
g
mes
10
0
2
h
ere
36
7
5
BB 15
9
2
B
B
o
ps
D
op
s
82
71
59
9
8
Average
s
ure for
per
f
d
ed for i
ndic
a
Total Co
s
e
colli
sio
n
t
e
s
a
lues of
Np
a
T = Nu x
r
e
T: total c
o
Nv: num
b
Cv
:
cost
o
Nu: num
b
Cu:
co
st
o
Co: indi
c
a
u
lts from
ex
p
w
. Calculati
o
S
A
O
k
-
o
r
V
ol. 11, No.
6
ers of Colli
si
2
0
0
Frame
s
300
5
7
2
872
2
5
5
387
13
4
9
9
8
6
207
165
140
fram
es pe
r
s
f
or
ma
nc
e
,
t
h
a
tive purpos
Figure
8
s
t Be
nchm
a
r
s
t
s
cond
uct
e
a
nd
C
p
we
r
e
Cu
+ Nv
x
C
o
s
t
func
tion
f
b
e
r
o
f
bound
i
o
f testing a
p
b
er o
f
bound
o
f averag
e
b
a
t
e
s co
st
f
o
r
p
e
r
iment
s f
o
o
ns
for tot
a
l
c
T
a
BV
Nv
phere
5239
5
A
ABB 5239
5
O
BB
5239
5
-
Dops 5239
5
r
-Do
p
s 5239
5
6
, June 20
1
3
ons Dete
cte
400
50
0
109
1
144
1
497 625
308
230
202
395
316
267
s
e
c
on
d (FP
S
h
e fps valu
e
s
es. Figu
re
8
8
. Frames
p
e
r
kin
g
.
A
mo
d
e
d in
th
ese
e
e
not in
clud
e
d
C
v +
C
o
f
or interfere
n
i
ng
vo
lu
me
p
p
ai
r of boun
d
ing volumes
b
ou
ndi
ng vol
u
one ti
me pr
o
o
r
the firs
t 5
c
os
t were b
a
a
ble 3. Nu
m
b
Cv
5
0.00024
5
0.00024
5
0.100106
5
0.000249
5
0.001567
0
100
10
0
3
0
fps
Numb
e
A
v
3
: 3115 – 3
1
d
0
F
S
).
A
lth
ough
s
for
differe
n
8
illustrates t
h
e
r
second fo
r
d
ified b
ench
m
e
xpe
r
iment
s
d
and total c
n
ce det
ectio
n
p
ai
r ove
r
lap
t
d
i
ng vol
u
me
s
u
pdate
d
,
u
me up
date,
o
cessing, in
t
00 frame
s
a
a
sed on Eq.
2
b
ers of Colli
s
Values
Nu
104790
0.
0
104790
0.0
104790
0.0
104790
0.0
104790
0.0
T
otal
number
of
k
‐
Dops
0
0
0
500
e
r
of
frames
t
e
v
er
ag
e
F
1
22
igure 7. Nu
m
fram
es
per
n
t types of
b
h
e fps value
s
r
the firs
t
50
0
m
a
r
ki
ng fun
c
d
o
not i
n
vo
l
os
t func
tion
n
,
t
est
s
s
(ov
e
rl
a
p
)
t
his ca
se
,
t
h
e
a
re summar
i
2
mentioned
s
i
ons Dete
ct
e
Cu C
o
0
0012
135.0
49377
90.63
50053
367.6
00124
574.2
00784
369.4
0
2000
100
collisions
N
Numb
e
e
sted
F
PS
k
A
O
e
-
m
ber of collis
second (FP
S
b
oundi
ng v
o
s
.
0
frames
c
tion ba
se
d
o
l
ve primitive
only involve
:
e
const
r
u
c
ti
o
i
zed in Tabl
above.
e
d
Total
Cost
o
783
160.212
781
5277.36
315
10857.7
866
600.328
002
533.635
100
200
300
400
500
N
umber
of
fr
a
e
r
of
co
k
‐
Dops
A
ABB
O
BB
-
ISSN: 2087
i
ons d
e
tecte
S
) is not a d
o
lume
s
us
ed
o
n Eq. 1
wa
s
t
e
st
s.
The
r
:
(
o
n co
st
e
3 and Fi
g
6
4
2
8
6
500
a
mes
llisions
Sp
h
AA
OB
k
‐
D
-278X
d
efinite
wer
e
s
us
ed
r
ef
ore,
(
2)
g
ure 9
h
ere
BB
B
D
ops
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Hybrid
Colli
si
on Culli
ng by
Bounding Vol
u
m
e
s Manipulation... (Norhaida Mohd Suaib)
3121
Figure 9. Total co
st based
on first 500 frames
The total
co
st benchma
r
ki
ng p
r
o
c
e
ss
wa
s al
so
ca
rried
out for
1
000 a
nd 5
0
0
0
frame
s
simulatio
n
s.
Experime
n
ts involving 10
00 fram
es
g
a
ve unexp
e
ct
ed result wh
ere th
e value
of
total cost for k-Dops
skyrock
eted above OBB. Based on the
first 500 fram
es t
e
sted, AABB and
k-Dop
s
gave
frame rat
e
s
way belo
w
int
e
ra
ctive fr
am
e rates
(~60 fps), an
d this
might cont
rib
u
te
to the lapse. These two b
oundi
ng volu
mes are excl
uded in the 5
000 frame
s
test. Result
s for
the final test (500
0 fram
es) are
as exp
e
c
ted; sp
he
re
gave the lowest co
st, follo
wed by o
r
-Do
p
s
(se
e
Figu
re 1
0
).
Figure 10. To
tal Cost ba
se
d on
Shortliste
d BV and 5000 F
r
ame
s
Figure 13. Fp
s Perfo
r
man
c
es for
Differe
nt BV Approa
ch
for 500 O
b
jects (first 10
0 frames)
5.2. Part II
Hybrid
Colli
si
on Culli
ng.
Re
sults from the experim
e
n
ts in
Part I show that com
b
ination
of sp
here a
n
d
or-Dop
s b
oundi
ng volu
mes
offer th
e pote
n
tial fo
r an
optio
n t
o
wa
rd
s a
bet
ter
colli
sion
culli
ng process. Hybrid
colli
si
on culli
ng tha
t
is a combi
n
ation of boun
ding sphe
re
and
Oriente
d
-Di
s
crete O
r
ientati
on Polytope
s
(Or-Dop
s)
wa
s imple
m
ente
d
on m
u
ltiple
*.tri obje
c
ts a
s
outlined in
previous
se
ctio
n. Figure
11
sho
w
s
hybri
d
colli
sion
cul
ling implem
e
n
ted on a
pai
r of
collidin
g o
b
je
ct – thi
s
i
s
d
one to
pu
rpo
s
ely hig
h
light
the con
c
ept.
It wa
s the
n
tested
agai
n
s
t
conve
n
tional
boun
ding vol
u
me app
ro
ach.
Figure 11. Hy
brid Collisi
on
cullin
g–note t
he
Pairs
with Sphere a
nd o
r
-Dop
s
Figure 12. Hy
brid Collisi
on
Culling
on 350
Obje
cts
0
20000
To
t
a
l
C
o
s
t
TotalCost
0
200000
S
…
O
…
or
…
To
t
a
l
Cos
t
fo
r
5000
Fr
ames
TotalCost
0
20
1
14
27
40
53
66
79
92
FPS
Frame
number
FPS
perf
ormance
No
BV
Sphere
Or
‐
Dops
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No. 6, June 20
13 : 3115 – 3
122
3122
Initial result based o
n
th
e frame
s
p
e
r se
con
d
(fp
s
) of 50
0 obj
ects
und
erg
o
i
ng rigi
d
motion i
s
sh
o
w
n
as Fig
u
re
13. Simul
a
tion
without
bo
undin
g
volum
e
(th
u
s no
col
lision
dete
c
tio
n
employed) was
i
m
plem
ent
ed
as a c
ont
rol experiment. If hybrid co
llision culling
is employed
on
a particular o
b
ject of intere
st (label
ed as ‘1ObjHyb
rid’ in the graph
),
the frame rates ca
n rea
c
h
to
nearly the pe
rforman
c
e of the co
ntrol ex
perim
ent,
sim
ilar to the perf
o
rma
n
ce of sphere bou
ndi
ng
volume. Fps perform
ance dropped if all objects
em
ployed hybrid collision
cull
ing, but it stil
l
outperfo
rm
s t
he h
o
mog
e
n
eou
s o
r
-Do
p
s
im
pleme
n
tation. An
oth
e
r
app
roa
c
h
wa
s al
so
tested
whe
r
e
a t
w
o-pass te
st
wa
s e
m
ploye
d
(l
abele
d
a
s
‘S
eq Sp
-O
rD’
i
n
the
graph
).
Th
e first te
st to
identify pairs
that are li
kely
to
collide
ba
sed
on
bo
un
ding sp
here
t
e
sts.
Th
ese pairs are
sen
t
to
the second pass
where
or-Dops te
st
s
will be conducted. It
show
s an improved perform
a
nce if
all obje
c
ts ne
eds to empl
o
y
the hybrid collision
cullin
g method.
6. Conclusio
n
Re
sults from
the experim
e
n
ts
shows th
at the hybrid
colli
sion
culli
ng metho
d
shows the
potential of an option for a better colli
sion culli
ng t
e
ch
niqu
e for massive rigid
body simula
tion
comp
ared to
the ho
mog
e
n
eou
s b
oun
din
g
volum
e
s.
Ho
wever a
d
e
tailed te
st li
ke th
e total
cost
ben
chma
rkin
g test need
s to be don
e to sy
stem
aticall
y
evaluate this re
sult.
Ackn
o
w
l
e
dg
ment
This
re
sea
r
ch is supp
ort
ed by Unive
r
siti
Teknolo
g
i
Malaysia
(UTM) an
d Min
i
stry of
High
er E
d
u
c
ation (MO
H
E
)
, in
colla
bo
ration
with
Re
sea
r
ch
M
ana
gement Ce
ntre (RMC),
UTM.
This pa
pe
r is finan
cially sup
porte
d by Resear
ch
University Grant (R
UG
) Prog
ram, Tier 2
(
r
e
s
e
a
r
c
h
gr
an
t 0
5
J
21
)
.
Referen
ces
[1]
Q Avril, V Gouranto
n
, B Arnal
di. Fast Co
llis
i
on C
u
ll
in
g i
n
Lar
ge-Sca
l
e
Environm
ents
Using GPU
Mapp
ing F
u
nction.
Euro
gra
phi
cs Sympos
iu
m
on Para
lle
l Graphics a
nd Vis
u
ali
z
a
t
i
on.
2
012.
[2]
D Harmo
n, D
Panozz
o
, O Sorkin
e, D Z
o
ri
n. Interference
-
a
w
are
ge
ome
t
ric model
in
g.
ACM T
r
ans.
Graph.
201
1; 3
0
: 1-10.
[3]
PM Hub
bar
d. Col
lisi
on
Det
e
ction f
o
r Int
e
ractive Gra
p
h
ics Ap
plic
atio
ns.
IEEE Transactions on
Visua
l
i
z
a
t
i
on a
nd Co
mputer
Graphics
. 19
95
; 1: 218-23
0.
[4]
NM Suaib, A Bade, D Moh
a
mad.
Co
llis
io
n Detectio
n: A Survey of
Techni
ques a
nd App
licati
o
n
.
Coll
isio
n Detec
t
ion for Rea
l
-T
i
m
e Grap
hics:
Series of T
e
ch
niq
ues.
Joh
o
r; UT
M. 2009; 1: 1-18.
[5]
NM Suai
b, A Bade, AM Z
i
n,
T
M
T
Sembok.
Oriented co
n
v
ex poly
h
e
d
ra
for collis
io
n de
tection i
n
3
D
computer a
n
i
m
ati
on.
Proce
edi
ngs of the 4th Internati
o
n
a
l Conf
erenc
e
on Computer
Graphics and
Interactive T
e
chni
ques i
n
Aus
t
ralasi
a and S
o
utheast Asia K
ual
a Lump
u
r, Mala
ysi
a
: ACM
,
2006.
[6]
J Klein, G Zachmann.
T
i
me-c
ritical c
o
ll
isio
n
detectio
n
usin
g
an
aver
ag
e-c
a
se
appr
oac
h
. Procee
din
g
s
of the ACM s
y
mposi
u
m on Vi
rtual rea
lit
y
s
o
ftw
a
r
e a
nd tech
nol
og
y Osaka,
Japa
n: ACM, 2003.
[7]
C Mendoz
a, C O'Sullivan.
An Interruptib
l
e
Algorit
hm fo
r Collis
ion D
e
tection b
e
tw
een Defor
m
a
b
l
e
Objects
. Intern
ation
a
l W
o
rksh
op on V
i
rtual
Real
it
y
an
d Ph
ysic
al Sim
u
lati
on (VRIPHYS'
05). 200
5: 73-
80.
[8]
T Giang, C O'Sulliva
n
.
Cl
osest Feature
Maps for Time-
C
ritica
l Co
llisi
on H
and
lin
g
. International
W
o
rkshop o
n
Virtual R
eal
it
y
and Ph
ys
ical S
i
mulati
on (VRI
PHYS'
05). 200
5: 65-72.
[9]
T
Giang, C O
'
Sulliva
n
. Virtu
a
l R
eal
it
y
I
n
te
racti
on
an
d P
h
y
s
ical
Simu
la
tion: Ap
pro
x
im
ate col
lisi
o
n
respo
n
se usi
n
g
closest feature
maps.
Comput
. Graph
. 2006;
30: 423-
43
1.
[10]
J Barbic, DL Ja
mes. Subspac
e self-col
lisi
on
culli
ng. ACM T
r
ans. Graph. 2
010; 29: 1-
9.
[11]
DS Comin
g
, OG Staadt. Stride sched
uli
ng fo
r time-critical c
o
llis
io
n detecti
on VRST
: AC
M. 2007.
[12]
DS Comi
ng, OG Staadt.
Stride Sche
du
ling for
T
i
me-Critical C
o
ll
isi
on Detecti
on.
T
he F
ourt
h
Internatio
na
l S
y
mp
osi
u
m o
n
3D D
a
ta P
r
ocessi
ng, Vis
ualiz
atio
n an
d
T
r
ansmissio
n
(3DPVT
'
08).
Atlanta, Georg
i
a, 2008.
[13]
GVD Bergen.
Coll
isio
n Detec
t
i
on in Interacti
v
e 3D Envir
o
n
m
ents.
Elsevier.
2004.
Evaluation Warning : The document was created with Spire.PDF for Python.