<s>
In	O
computer	B-Application
vision	I-Application
,	O
the	O
term	O
cuboid	B-General_Concept
is	O
used	O
to	O
describe	O
a	O
small	O
spatiotemporal	O
volume	O
extracted	O
for	O
purposes	O
of	O
behavior	B-Application
recognition	I-Application
.	O
</s>
<s>
The	O
cuboid	B-General_Concept
is	O
regarded	O
as	O
a	O
basic	O
geometric	B-Algorithm
primitive	I-Algorithm
type	O
and	O
is	O
used	O
to	O
depict	O
three-dimensional	O
objects	O
within	O
a	O
three	O
dimensional	O
representation	O
of	O
a	O
flat	O
,	O
two	O
dimensional	O
image	O
.	O
</s>
<s>
Cuboids	B-General_Concept
can	O
be	O
produced	O
from	O
both	O
two-dimensional	O
and	O
three-dimensional	O
images	O
.	O
</s>
<s>
One	O
method	O
used	O
to	O
produce	O
cuboids	B-General_Concept
utilizes	O
scene	O
understanding	O
(	O
SUN	O
)	O
primitive	B-Algorithm
databases	O
,	O
which	O
are	O
collections	O
of	O
pictures	O
that	O
already	O
contain	O
cuboids	B-General_Concept
.	O
</s>
<s>
By	O
sorting	O
through	O
SUN	O
primitive	B-Algorithm
databases	O
with	O
machine	O
learning	O
tools	O
,	O
computers	O
observe	O
the	O
conditions	O
in	O
which	O
cuboids	B-General_Concept
are	O
produced	O
in	O
images	O
from	O
SUN	O
primitive	B-Algorithm
databases	O
and	O
can	O
learn	O
to	O
produce	O
cuboids	B-General_Concept
from	O
other	O
images	O
.	O
</s>
<s>
RGB-D	O
images	O
,	O
which	O
are	O
RGB	O
images	O
that	O
also	O
record	O
the	O
depth	O
of	O
each	O
pixel	O
,	O
are	O
occasionally	O
used	O
to	O
produce	O
cuboids	B-General_Concept
because	O
computers	O
no	O
longer	O
need	O
to	O
determine	O
the	O
depth	O
of	O
an	O
object	O
,	O
as	O
they	O
typically	O
do	O
because	O
depth	O
is	O
already	O
recorded	O
.	O
</s>
<s>
Cuboid	B-General_Concept
production	O
is	O
sensitive	O
to	O
changes	O
in	O
color	O
and	O
illumination	O
,	O
blockage	O
,	O
and	O
background	O
clutter	O
.	O
</s>
<s>
This	O
means	O
that	O
it	O
is	O
difficult	O
for	O
computers	O
to	O
produce	O
cuboids	B-General_Concept
of	O
objects	O
that	O
are	O
multicolored	O
,	O
irregularly	O
illuminated	O
,	O
or	O
partially	O
covered	O
,	O
or	O
if	O
there	O
are	O
many	O
objects	O
in	O
the	O
background	O
.	O
</s>
<s>
This	O
is	O
partially	O
due	O
to	O
the	O
fact	O
that	O
algorithms	O
for	O
producing	O
cuboids	B-General_Concept
are	O
still	O
relatively	O
simple	O
.	O
</s>
<s>
Cuboids	B-General_Concept
are	O
created	O
for	O
point	O
cloud-based	O
three-dimensional	O
maps	O
and	O
can	O
be	O
utilized	O
in	O
various	O
situations	O
such	O
as	O
augmented	O
reality	O
,	O
the	O
automated	O
control	O
of	O
cars	O
,	O
drones	O
,	O
and	O
robots	O
,	O
and	O
object	O
detection	O
.	O
</s>
<s>
Cuboids	B-General_Concept
allow	O
for	O
software	O
to	O
identify	O
a	O
scene	O
through	O
geometric	O
descriptions	O
in	O
an	O
“	O
object-agnostic	O
”	O
fashion	O
.	O
</s>
<s>
Interest	O
points	O
,	O
locations	O
within	O
images	O
that	O
are	O
identified	O
by	O
a	O
computer	O
as	O
essential	O
to	O
identifying	O
the	O
image	O
,	O
created	O
from	O
two-dimensional	O
images	O
can	O
be	O
used	O
with	O
cuboids	B-General_Concept
for	O
image	O
matching	O
,	O
identifying	O
a	O
room	O
or	O
scene	O
,	O
and	O
instance	O
recognition	O
.	O
</s>
<s>
Interest	O
points	O
created	O
from	O
three	O
dimensional	O
images	O
can	O
be	O
used	O
with	O
cuboids	B-General_Concept
to	O
recognize	O
activities	O
.	O
</s>
<s>
RGB-D	O
images	O
and	O
SLAM	B-Application
systems	O
are	O
used	O
together	O
in	O
RGB-D	O
SLAM	B-Application
systems	O
,	O
which	O
are	O
employed	O
by	O
Computer-aided	B-Application
design	I-Application
systems	O
to	O
generate	O
point	O
cloud-based	O
three-dimensional	O
maps	O
.	O
</s>
<s>
Most	O
industrial	O
multi-axis	O
machining	O
tools	O
use	O
computer-aided	O
manufacturing	O
and	O
subsequently	O
work	O
in	O
cuboid	B-General_Concept
work	O
spaces	O
.	O
</s>
