<s>
ViBe	B-General_Concept
is	O
a	O
background	B-Algorithm
subtraction	I-Algorithm
algorithm	O
which	O
has	O
been	O
presented	O
at	O
the	O
IEEE	O
ICASSP	O
2009	O
conference	O
and	O
was	O
refined	O
in	O
later	O
publications	O
.	O
</s>
<s>
ViBe	B-General_Concept
is	O
patented	O
:	O
the	O
patent	O
covers	O
various	O
aspects	O
such	O
as	O
stochastic	O
replacement	O
,	O
spatial	O
diffusion	O
,	O
and	O
non-chronological	O
handling	O
.	O
</s>
<s>
ViBe	B-General_Concept
is	O
written	O
in	O
the	O
programming	B-Language
language	I-Language
C	I-Language
,	O
and	O
has	O
been	O
implemented	O
on	O
CPU	B-Device
,	O
GPU	B-Architecture
and	O
FPGA	B-Architecture
.	O
</s>
<s>
Many	O
advanced	O
techniques	O
are	O
used	O
to	O
provide	O
an	O
estimate	O
of	O
the	O
temporal	O
probability	O
density	O
function	O
(	O
pdf	O
)	O
of	O
a	O
pixel	O
x	O
.	O
ViBe	B-General_Concept
's	O
approach	O
is	O
different	O
,	O
as	O
it	O
imposes	O
the	O
influence	O
of	O
a	O
value	O
in	O
the	O
polychromatic	O
space	O
to	O
be	O
limited	O
to	O
the	O
local	O
neighborhood	O
.	O
</s>
<s>
In	O
practice	O
,	O
ViBe	B-General_Concept
does	O
not	O
estimate	O
the	O
pdf	O
,	O
but	O
uses	O
a	O
set	O
of	O
previously	O
observed	O
sample	O
values	O
as	O
a	O
pixel	O
model	O
.	O
</s>
<s>
ViBe	B-General_Concept
ensures	O
a	O
smooth	O
exponentially	O
decaying	O
lifespan	O
for	O
the	O
sample	O
values	O
that	O
constitute	O
the	O
pixel	O
models	O
.	O
</s>
<s>
This	O
makes	O
ViBe	B-General_Concept
able	O
to	O
successfully	O
deal	O
with	O
concomitant	O
events	O
with	O
a	O
single	O
model	O
of	O
a	O
reasonable	O
size	O
for	O
each	O
pixel	O
.	O
</s>
<s>
To	O
ensure	O
the	O
spatial	O
consistency	O
of	O
the	O
whole	O
image	O
model	O
and	O
handle	O
practical	O
situations	O
such	O
as	O
small	O
camera	O
movements	O
or	O
slowly	O
evolving	O
background	O
objects	O
,	O
ViBe	B-General_Concept
uses	O
a	O
technique	O
similar	O
to	O
that	O
developed	O
for	O
the	O
updating	O
process	O
in	O
which	O
it	O
chooses	O
at	O
random	O
and	O
update	O
a	O
pixel	O
model	O
in	O
the	O
neighborhood	O
of	O
the	O
current	O
pixel	O
.	O
</s>
<s>
By	O
denoting	O
NG(x )	O
and	O
p(x )	O
respectively	O
the	O
spatial	O
neighborhood	O
of	O
a	O
pixel	O
x	O
and	O
its	O
value	O
,	O
and	O
assuming	O
that	O
it	O
was	O
decided	O
to	O
update	O
the	O
set	O
of	O
samples	O
of	O
x	O
by	O
inserting	O
p(x )	O
,	O
then	O
ViBe	B-General_Concept
also	O
use	O
this	O
value	O
p(x )	O
to	O
update	O
the	O
set	O
of	O
samples	O
of	O
one	O
of	O
the	O
pixels	O
in	O
the	O
neighborhood	O
NG(x )	O
,	O
chosen	O
at	O
random	O
.	O
</s>
<s>
As	O
a	O
result	O
,	O
ViBe	B-General_Concept
is	O
able	O
to	O
produce	O
spatially	O
coherent	O
results	O
directly	O
without	O
the	O
use	O
of	O
any	O
post-processing	O
method	O
.	O
</s>
<s>
Since	O
no	O
temporal	O
information	O
is	O
available	O
prior	O
to	O
the	O
second	O
frame	O
,	O
ViBe	B-General_Concept
populates	O
the	O
pixel	O
models	O
with	O
values	O
found	O
in	O
the	O
spatial	O
neighborhood	O
of	O
each	O
pixel	O
;	O
more	O
precisely	O
,	O
it	O
initializes	O
the	O
background	O
model	O
with	O
values	O
taken	O
randomly	O
in	O
each	O
pixel	O
neighborhood	O
of	O
the	O
first	O
frame	O
.	O
</s>
