<s>
Articulated	B-General_Concept
body	I-General_Concept
pose	I-General_Concept
estimation	I-General_Concept
in	O
computer	B-Application
vision	I-Application
is	O
the	O
study	O
of	O
algorithms	O
and	O
systems	O
that	O
recover	O
the	O
pose	B-Architecture
of	O
an	O
articulated	O
body	O
,	O
which	O
consists	O
of	O
joints	O
and	O
rigid	O
parts	O
using	O
image-based	O
observations	O
.	O
</s>
<s>
It	O
is	O
one	O
of	O
the	O
longest-lasting	O
problems	O
in	O
computer	B-Application
vision	I-Application
because	O
of	O
the	O
complexity	O
of	O
the	O
models	O
that	O
relate	O
observation	O
with	O
pose	B-Architecture
,	O
and	O
because	O
of	O
the	O
variety	O
of	O
situations	O
in	O
which	O
it	O
would	O
be	O
useful	O
.	O
</s>
<s>
Thus	O
pose	B-General_Concept
estimation	I-General_Concept
is	O
an	O
important	O
and	O
challenging	O
problem	O
in	O
computer	B-Application
vision	I-Application
,	O
and	O
many	O
algorithms	O
have	O
been	O
deployed	O
in	O
solving	O
this	O
problem	O
over	O
the	O
last	O
two	O
decades	O
.	O
</s>
<s>
Pose	B-General_Concept
estimation	I-General_Concept
is	O
a	O
difficult	O
problem	O
and	O
an	O
active	O
subject	O
of	O
research	O
because	O
the	O
human	O
body	O
has	O
244	O
degrees	O
of	O
freedom	O
with	O
230	O
joints	O
.	O
</s>
<s>
Finally	O
,	O
most	O
algorithms	O
estimate	O
pose	B-Architecture
from	O
monocular	O
(	O
two-dimensional	O
)	O
images	O
,	O
taken	O
from	O
a	O
normal	O
camera	O
.	O
</s>
<s>
These	O
images	O
lack	O
the	O
three-dimensional	O
information	O
of	O
an	O
actual	O
body	O
pose	B-Architecture
,	O
leading	O
to	O
further	O
ambiguities	O
.	O
</s>
<s>
The	O
typical	O
articulated	B-General_Concept
body	I-General_Concept
pose	I-General_Concept
estimation	I-General_Concept
system	O
involves	O
a	O
model-based	O
approach	O
,	O
in	O
which	O
the	O
pose	B-General_Concept
estimation	I-General_Concept
is	O
achieved	O
by	O
maximizing/minimizing	O
a	O
similarity/dissimilarity	O
between	O
an	O
observation	O
(	O
input	O
)	O
and	O
a	O
template	O
model	O
.	O
</s>
<s>
Long-wave	B-Algorithm
thermal	O
infrared	O
imagery	O
,	O
</s>
<s>
The	O
above	O
equation	O
simply	O
represents	O
the	O
spring	O
model	O
used	O
to	O
describe	O
body	O
pose	B-Architecture
.	O
</s>
<s>
To	O
estimate	O
pose	B-Architecture
from	O
images	O
,	O
cost	O
or	O
energy	O
function	O
must	O
be	O
minimized	O
.	O
</s>
<s>
oriented	O
(	O
deformed	O
)	O
parts	O
match	O
,	O
thus	O
accounting	O
for	O
articulation	O
along	O
with	O
object	B-General_Concept
detection	I-General_Concept
.	O
</s>
<s>
Given	O
a	O
transformation	O
matrix	O
,	O
the	O
joint	O
position	O
at	O
the	O
T-pose	O
can	O
be	O
transferred	O
to	O
its	O
corresponding	O
position	O
in	O
the	O
world	O
coordination	O
.	O
</s>
<s>
Since	O
about	O
2016	O
,	O
deep	B-Algorithm
learning	I-Algorithm
has	O
emerged	O
as	O
the	O
dominant	O
method	O
for	O
performing	O
accurate	O
articulated	B-General_Concept
body	I-General_Concept
pose	I-General_Concept
estimation	I-General_Concept
.	O
</s>
<s>
The	O
first	O
deep	B-Algorithm
learning	I-Algorithm
models	O
that	O
emerged	O
focused	O
on	O
extracting	O
the	O
2D	O
positions	O
of	O
human	O
joints	O
in	O
an	O
image	O
.	O
</s>
<s>
Such	O
models	O
take	O
in	O
an	O
image	O
and	O
pass	O
it	O
through	O
a	O
convolutional	B-Architecture
neural	I-Architecture
network	I-Architecture
to	O
obtain	O
a	O
series	O
of	O
heatmaps	O
(	O
one	O
for	O
each	O
joint	O
)	O
which	O
take	O
on	O
high	O
values	O
where	O
joints	O
are	O
detected	O
.	O
</s>
<s>
Using	O
these	O
fields	O
,	O
joints	O
can	O
be	O
grouped	O
limb	O
by	O
limb	O
by	O
solving	O
a	O
series	O
of	O
assignment	B-Algorithm
problems	I-Algorithm
.	I-Algorithm
</s>
<s>
In	O
the	O
second	O
,	O
"	O
top-down	O
"	O
approach	O
,	O
an	O
additional	O
network	O
is	O
used	O
to	O
first	O
detect	B-General_Concept
people	I-General_Concept
in	O
the	O
image	O
and	O
then	O
the	O
pose	B-General_Concept
estimation	I-General_Concept
network	O
is	O
applied	O
to	O
each	O
image	O
.	O
</s>
<s>
With	O
the	O
advent	O
of	O
multiple	O
datasets	O
with	O
human	O
pose	B-Architecture
annotated	O
in	O
multiple	O
views	O
,	O
models	O
which	O
detect	O
3D	O
joint	O
positions	O
became	O
more	O
popular	O
.	O
</s>
<s>
Such	O
approaches	O
often	O
project	O
image	O
features	O
into	O
a	O
cube	O
and	O
then	O
use	O
a	O
3D	O
convolutional	B-Architecture
neural	I-Architecture
network	I-Architecture
to	O
predict	O
a	O
3D	O
heatmap	O
for	O
each	O
joint	O
.	O
</s>
<s>
Most	O
of	O
the	O
work	O
is	O
based	O
on	O
estimating	O
the	O
appropriate	O
pose	B-Architecture
of	O
the	O
skinned	O
multi-person	O
linear	O
(	O
SMPL	O
)	O
model	O
within	O
an	O
image	O
.	O
</s>
<s>
To	O
address	O
this	O
issue	O
,	O
computer	B-Application
vision	I-Application
researchers	O
have	O
developed	O
new	O
algorithms	O
which	O
can	O
learn	O
3D	O
keypoints	O
given	O
only	O
annotated	O
2D	O
images	O
from	O
a	O
single	O
view	O
or	O
identify	O
keypoints	O
given	O
videos	O
without	O
any	O
annotations	O
.	O
</s>
<s>
For	O
these	O
robots	O
,	O
high-accuracy	O
human	O
detection	O
and	O
pose	B-General_Concept
estimation	I-General_Concept
is	O
necessary	O
to	O
perform	O
a	O
variety	O
of	O
tasks	O
,	O
such	O
as	O
fall	O
detection	O
.	O
</s>
<s>
However	O
,	O
poses	B-Architecture
can	O
be	O
synced	O
directly	O
to	O
a	O
real-life	O
actor	O
through	O
specialized	O
pose	B-General_Concept
estimation	I-General_Concept
systems	O
.	O
</s>
<s>
Recent	O
advances	O
in	O
pose	B-General_Concept
estimation	I-General_Concept
and	O
motion	B-Application
capture	I-Application
have	O
enabled	O
markerless	O
applications	O
,	O
sometimes	O
in	O
real	O
time	O
.	O
</s>
<s>
As	O
such	O
,	O
an	O
intelligent	O
system	O
tracking	O
driver	O
pose	B-Architecture
may	O
be	O
useful	O
for	O
emergency	O
alerts	O
.	O
</s>
<s>
Along	O
the	O
same	O
lines	O
,	O
pedestrian	B-General_Concept
detection	I-General_Concept
algorithms	O
have	O
been	O
used	O
successfully	O
in	O
autonomous	O
cars	O
,	O
enabling	O
the	O
car	O
to	O
make	O
smarter	O
decisions	O
.	O
</s>
<s>
Commercially	O
,	O
pose	B-General_Concept
estimation	I-General_Concept
has	O
been	O
used	O
in	O
the	O
context	O
of	O
video	O
games	O
,	O
popularized	O
with	O
the	O
Microsoft	B-Algorithm
Kinect	I-Algorithm
sensor	O
(	O
a	O
depth	O
camera	O
)	O
.	O
</s>
<s>
These	O
systems	O
track	O
the	O
user	O
to	O
render	O
their	O
avatar	O
in-game	O
,	O
in	O
addition	O
to	O
performing	O
tasks	O
like	O
gesture	B-General_Concept
recognition	I-General_Concept
to	O
enable	O
the	O
user	O
to	O
interact	O
with	O
the	O
game	O
.	O
</s>
<s>
Pose	B-General_Concept
estimation	I-General_Concept
has	O
been	O
used	O
to	O
detect	O
postural	O
issues	O
such	O
as	O
scoliosis	O
by	O
analyzing	O
abnormalities	O
in	O
a	O
patient	O
's	O
posture	O
,	O
physical	O
therapy	O
,	O
and	O
the	O
study	O
of	O
the	O
cognitive	O
brain	O
development	O
of	O
young	O
children	O
by	O
monitoring	O
motor	O
functionality	O
.	O
</s>
<s>
Other	O
applications	O
include	O
video	O
surveillance	O
,	O
animal	O
tracking	O
and	O
behavior	O
understanding	O
,	O
sign	O
language	O
detection	O
,	O
advanced	O
human	O
–	O
computer	O
interaction	O
,	O
and	O
markerless	O
motion	B-Application
capturing	I-Application
.	O
</s>
<s>
A	O
commercially	O
successful	O
but	O
specialized	O
computer	O
vision-based	O
articulated	B-General_Concept
body	I-General_Concept
pose	I-General_Concept
estimation	I-General_Concept
technique	O
is	O
optical	O
motion	B-Application
capture	I-Application
.	O
</s>
<s>
A	O
number	O
of	O
groups	O
and	O
companies	O
are	O
researching	O
pose	B-General_Concept
estimation	I-General_Concept
,	O
including	O
groups	O
at	O
Brown	O
University	O
,	O
Carnegie	O
Mellon	O
University	O
,	O
MPI	O
Saarbruecken	O
,	O
Stanford	O
University	O
,	O
the	O
University	O
of	O
California	O
,	O
San	O
Diego	O
,	O
the	O
University	O
of	O
Toronto	O
,	O
the	O
École	O
Centrale	O
Paris	O
,	O
ETH	O
Zurich	O
,	O
National	O
University	O
of	O
Sciences	O
and	O
Technology	O
(	O
NUST	O
)	O
,	O
the	O
University	O
of	O
California	O
,	O
Irvine	O
and	O
Polytechnic	O
University	O
of	O
Catalonia	O
.	O
</s>
<s>
At	O
present	O
,	O
several	O
companies	O
are	O
working	O
on	O
articulated	B-General_Concept
body	I-General_Concept
pose	I-General_Concept
estimation	I-General_Concept
.	O
</s>
