<s>
Activity	B-Application
recognition	I-Application
aims	O
to	O
recognize	O
the	O
actions	O
and	O
goals	O
of	O
one	O
or	O
more	O
agents	O
from	O
a	O
series	O
of	O
observations	O
on	O
the	O
agents	O
 '	O
actions	O
and	O
the	O
environmental	O
conditions	O
.	O
</s>
<s>
Since	O
the	O
1980s	O
,	O
this	O
research	O
field	O
has	O
captured	O
the	O
attention	O
of	O
several	O
computer	B-General_Concept
science	I-General_Concept
communities	O
due	O
to	O
its	O
strength	O
in	O
providing	O
personalized	O
support	O
for	O
many	O
different	O
applications	O
and	O
its	O
connection	O
to	O
many	O
different	O
fields	O
of	O
study	O
such	O
as	O
medicine	O
,	O
human-computer	O
interaction	O
,	O
or	O
sociology	O
.	O
</s>
<s>
Due	O
to	O
its	O
multifaceted	O
nature	O
,	O
different	O
fields	O
may	O
refer	O
to	O
activity	B-Application
recognition	I-Application
as	O
plan	O
recognition	O
,	O
goal	B-Application
recognition	I-Application
,	O
intent	B-Application
recognition	I-Application
,	O
behavior	B-Application
recognition	I-Application
,	O
location	O
estimation	O
and	O
location-based	B-Application
services	I-Application
.	O
</s>
<s>
Sensor-based	O
activity	B-Application
recognition	I-Application
integrates	O
the	O
emerging	O
area	O
of	O
sensor	O
networks	O
with	O
novel	O
data	B-Application
mining	I-Application
and	O
machine	O
learning	O
techniques	O
to	O
model	O
a	O
wide	O
range	O
of	O
human	O
activities	O
.	O
</s>
<s>
smart	O
phones	O
)	O
provide	O
sufficient	O
sensor	O
data	O
and	O
calculation	O
power	O
to	O
enable	O
physical	O
activity	B-Application
recognition	I-Application
to	O
provide	O
an	O
estimation	O
of	O
the	O
energy	O
consumption	O
during	O
everyday	O
life	O
.	O
</s>
<s>
Sensor-based	O
activity	B-Application
recognition	I-Application
researchers	O
believe	O
that	O
by	O
empowering	O
ubiquitous	B-Architecture
computers	I-Architecture
and	O
sensors	O
to	O
monitor	O
the	O
behavior	O
of	O
agents	O
(	O
under	O
consent	O
)	O
,	O
these	O
computers	O
will	O
be	O
better	O
suited	O
to	O
act	O
on	O
our	O
behalf	O
.	O
</s>
<s>
Visual	O
sensors	O
that	O
incorporate	O
color	O
and	O
depth	O
information	O
,	O
such	O
as	O
the	O
kinect	B-Algorithm
,	O
allow	O
more	O
accurate	O
automatic	O
action	B-Application
recognition	I-Application
and	O
fuse	O
many	O
emerging	O
applications	O
such	O
as	O
interactive	O
education	O
and	O
smart	O
environments	O
.	O
</s>
<s>
Multiple	O
views	O
of	O
visual	O
sensor	O
enables	O
the	O
development	O
of	O
machine	O
learning	O
for	O
automatic	O
view	O
invariant	O
action	B-Application
recognition	I-Application
.	O
</s>
<s>
More	O
advanced	O
sensors	O
used	O
in	O
3D	B-Application
motion	I-Application
capture	I-Application
systems	O
allow	O
highly	O
accurate	O
automatic	O
recognition	O
,	O
in	O
the	O
expenses	O
of	O
more	O
complicated	O
hardware	O
system	O
setup	O
.	O
</s>
<s>
Sensor-based	O
activity	B-Application
recognition	I-Application
is	O
a	O
challenging	O
task	O
due	O
to	O
the	O
inherent	O
noisy	O
nature	O
of	O
the	O
input	O
.	O
</s>
<s>
Recognition	O
of	O
group	O
activities	O
is	O
fundamentally	O
different	O
from	O
single	O
,	O
or	O
multi-user	O
activity	B-Application
recognition	I-Application
in	O
that	O
the	O
goal	O
is	O
to	O
recognize	O
the	O
behavior	O
of	O
the	O
group	O
as	O
an	O
entity	O
,	O
rather	O
than	O
the	O
activities	O
of	O
the	O
individual	O
members	O
within	O
it	O
.	O
</s>
<s>
Group	O
activity	B-Application
recognition	I-Application
has	O
applications	O
for	O
crowd	O
management	O
and	O
response	O
in	O
emergency	O
situations	O
,	O
as	O
well	O
as	O
for	O
social	O
networking	O
and	O
Quantified	O
Self	O
applications	O
.	O
</s>
<s>
Lesh	O
and	O
Etzioni	O
went	O
one	O
step	O
further	O
and	O
presented	O
methods	O
in	O
scaling	O
up	O
goal	B-Application
recognition	I-Application
to	O
scale	O
up	O
his	O
work	O
computationally	O
.	O
</s>
<s>
Furthermore	O
,	O
they	O
introduced	O
compact	O
representations	O
and	O
efficient	O
algorithms	O
for	O
goal	B-Application
recognition	I-Application
on	O
large	O
plan	O
libraries	O
.	O
</s>
<s>
Another	O
approach	O
to	O
logic-based	O
activity	B-Application
recognition	I-Application
is	O
to	O
use	O
stream	O
reasoning	O
based	O
on	O
answer	B-Application
set	I-Application
programming	I-Application
,	O
and	O
has	O
been	O
applied	O
to	O
recognising	O
activities	O
for	O
health-related	O
applications	O
,	O
which	O
uses	O
weak	O
constraints	O
to	O
model	O
a	O
degree	O
of	O
ambiguity/uncertainty	O
.	O
</s>
<s>
Probability	O
theory	O
and	O
statistical	O
learning	O
models	O
are	O
more	O
recently	O
applied	O
in	O
activity	B-Application
recognition	I-Application
to	O
reason	O
about	O
actions	O
,	O
plans	O
and	O
goals	O
under	O
uncertainty	O
.	O
</s>
<s>
In	O
IEEE	O
Pervasive	B-Architecture
Computing	I-Architecture
,	O
pages	O
50	O
–	O
57	O
,	O
October	O
2004.Dieter	O
Fox	O
Lin	O
Liao	O
,	O
Donald	O
J	O
.	O
Patterson	O
and	O
Henry	O
A	O
.	O
Kautz	O
.	O
</s>
<s>
The	O
use	O
of	O
temporal	O
probabilistic	O
models	O
has	O
been	O
shown	O
to	O
perform	O
well	O
in	O
activity	B-Application
recognition	I-Application
and	O
generally	O
outperform	O
non-temporal	O
models.TLM	O
van	O
Kasteren	O
,	O
Gwenn	O
Englebienne	O
,	O
BJA	O
Kröse	O
.	O
</s>
<s>
"	O
Human	B-Application
activity	I-Application
recognition	I-Application
from	O
wireless	O
sensor	O
network	O
data	O
:	O
Benchmark	O
and	O
software.	O
"	O
</s>
<s>
Activity	B-Application
Recognition	I-Application
in	O
Pervasive	O
Intelligent	O
Environments	O
,	O
165	O
–	O
186	O
,	O
Atlantis	O
Press	O
Generative	O
models	O
such	O
as	O
the	O
Hidden	O
Markov	O
Model	O
(	O
HMM	O
)	O
and	O
the	O
more	O
generally	O
formulated	O
Dynamic	O
Bayesian	O
Networks	O
(	O
DBN	O
)	O
are	O
popular	O
choices	O
in	O
modelling	O
activities	O
from	O
sensor	O
data.Piyathilaka	O
,	O
L.	O
;	O
Kodagoda	O
,	O
S.	O
,	O
"	O
Gaussian	O
mixture	O
based	O
HMM	O
for	O
human	O
daily	O
activity	B-Application
recognition	I-Application
using	O
3D	O
skeleton	O
features	O
,	O
"	O
Industrial	O
Electronics	O
and	O
Applications	O
(	O
ICIEA	O
)	O
,	O
2013	O
8th	O
IEEE	O
Conference	O
on	O
,	O
vol.	O
,	O
no.	O
,	O
pp.567	O
,	O
572	O
,	O
19	O
–	O
21	O
June	O
2013TLM	O
van	O
Kasteren	O
,	O
Gwenn	O
Englebienne	O
,	O
Ben	O
Kröse	O
"	O
Hierarchical	O
Activity	B-Application
Recognition	I-Application
Using	O
Automatically	O
Clustered	O
Actions	O
"	O
,	O
2011	O
,	O
Ambient	O
Intelligence	O
,	O
82	O
–	O
91	O
,	O
Springer	O
Berlin/HeidelbergDaniel	O
Wilson	O
and	O
Chris	O
Atkeson	O
.	O
</s>
<s>
Discriminative	O
models	O
such	O
as	O
Conditional	B-General_Concept
Random	I-General_Concept
Fields	I-General_Concept
(	O
CRF	O
)	O
are	O
also	O
commonly	O
applied	O
and	O
also	O
give	O
good	O
performance	O
in	O
activity	O
recognition.TLM	O
Van	O
Kasteren	O
,	O
Athanasios	O
Noulas	O
,	O
Gwenn	O
Englebienne	O
,	O
Ben	O
Kröse	O
,	O
"	O
Accurate	O
activity	B-Application
recognition	I-Application
in	O
a	O
home	O
setting	O
"	O
,	O
2008/9/21	O
,	O
Proceedings	O
of	O
the	O
10th	O
international	O
conference	O
on	O
Ubiquitous	B-Architecture
computing	I-Architecture
,	O
1	O
–	O
9	O
,	O
ACMDerek	O
Hao	O
Hu	O
,	O
Sinno	O
Jialin	O
Pan	O
,	O
Vincent	O
Wenchen	O
Zheng	O
,	O
Nathan	O
NanLiu	O
,	O
and	O
Qiang	O
Yang	O
.	O
</s>
<s>
Real	O
world	O
activity	B-Application
recognition	I-Application
with	O
multiple	O
goals	O
.	O
</s>
<s>
In	O
Proceedings	O
of	O
the	O
10th	O
international	O
conference	O
on	O
Ubiquitous	B-Architecture
computing	I-Architecture
,	O
Ubicomp	B-Architecture
,	O
pages	O
30	O
–	O
39	O
,	O
New	O
York	O
,	O
NY	O
,	O
USA	O
,	O
2008	O
.	O
</s>
<s>
A	O
dataset	O
together	O
with	O
implementations	O
of	O
a	O
number	O
of	O
popular	O
models	O
(	O
HMM	O
,	O
CRF	O
)	O
for	O
activity	B-Application
recognition	I-Application
can	O
be	O
found	O
here	O
.	O
</s>
<s>
Conventional	O
temporal	O
probabilistic	O
models	O
such	O
as	O
the	O
hidden	O
Markov	O
model	O
(	O
HMM	O
)	O
and	O
conditional	B-General_Concept
random	I-General_Concept
fields	I-General_Concept
(	O
CRF	O
)	O
model	O
directly	O
model	O
the	O
correlations	O
between	O
the	O
activities	O
and	O
the	O
observed	O
sensor	O
data	O
.	O
</s>
<s>
Hierarchical	O
models	O
for	O
activity	B-Application
recognition	I-Application
.	O
</s>
<s>
Examples	O
of	O
such	O
a	O
hierarchical	O
model	O
are	O
Layered	O
Hidden	O
Markov	O
Models	O
(	O
LHMMs	O
)	O
and	O
the	O
hierarchical	O
hidden	O
Markov	O
model	O
(	O
HHMM	O
)	O
,	O
which	O
have	O
been	O
shown	O
to	O
significantly	O
outperform	O
its	O
non-hierarchical	O
counterpart	O
in	O
activity	B-Application
recognition	I-Application
.	O
</s>
<s>
Different	O
from	O
traditional	O
machine	O
learning	O
approaches	O
,	O
an	O
approach	O
based	O
on	O
data	B-Application
mining	I-Application
has	O
been	O
recently	O
proposed	O
.	O
</s>
<s>
In	O
the	O
work	O
of	O
Gu	O
et	O
al.	O
,	O
the	O
problem	O
of	O
activity	B-Application
recognition	I-Application
is	O
formulated	O
as	O
a	O
pattern-based	O
classification	O
problem	O
.	O
</s>
<s>
They	O
proposed	O
a	O
data	B-Application
mining	I-Application
approach	O
based	O
on	O
discriminative	O
patterns	O
which	O
describe	O
significant	O
changes	O
between	O
any	O
two	O
activity	O
classes	O
of	O
data	O
to	O
recognize	O
sequential	O
,	O
interleaved	O
and	O
concurrent	O
activities	O
in	O
a	O
unified	O
solution.Tao	O
Gu	O
,	O
Zhanqing	O
Wu	O
,	O
Xianping	O
Tao	O
,	O
Hung	O
Keng	O
Pung	O
,	O
and	O
Jian	O
Lu	O
.	O
</s>
<s>
epSICAR	O
:	O
An	O
Emerging	O
Patterns	O
based	O
Approach	O
to	O
Sequential	O
,	O
Interleaved	O
and	O
Concurrent	O
Activity	B-Application
Recognition	I-Application
.	O
</s>
<s>
of	O
the	O
7th	O
Annual	O
IEEE	O
International	O
Conference	O
on	O
Pervasive	B-Architecture
Computing	I-Architecture
and	O
Communications	O
(	O
Percom	O
'	O
09	O
)	O
,	O
Galveston	O
,	O
Texas	O
,	O
March	O
9	O
–	O
13	O
,	O
2009	O
.	O
</s>
<s>
At	O
each	O
stage	O
of	O
the	O
hierarchy	O
,	O
the	O
most	O
distinctive	O
and	O
descriptive	O
features	O
are	O
learned	O
efficiently	O
through	O
data	B-Application
mining	I-Application
(	O
Apriori	O
rule	O
)	O
.Gilbert	O
A	O
,	O
Illingworth	O
J	O
,	O
Bowden	O
R	O
.	O
Action	B-Application
Recognition	I-Application
using	O
Mined	O
Hierarchical	O
Compound	O
Features	O
.	O
</s>
<s>
Location-based	O
activity	B-Application
recognition	I-Application
can	O
also	O
rely	O
on	O
GPS	O
data	O
to	O
recognize	O
activities.Liao	O
,	O
Lin	O
,	O
Dieter	O
Fox	O
,	O
and	O
Henry	O
Kautz	O
.	O
</s>
<s>
"	O
Hierarchical	O
conditional	B-General_Concept
random	I-General_Concept
fields	I-General_Concept
for	O
GPS-based	O
activity	O
recognition.	O
"	O
</s>
<s>
The	O
primary	O
technique	O
employed	O
is	O
Computer	B-Application
Vision	I-Application
.	O
</s>
<s>
Vision-based	O
activity	B-Application
recognition	I-Application
has	O
found	O
many	O
applications	O
such	O
as	O
human-computer	O
interaction	O
,	O
user	O
interface	O
design	O
,	O
robot	O
learning	O
,	O
and	O
surveillance	O
,	O
among	O
others	O
.	O
</s>
<s>
Scientific	O
conferences	O
where	O
vision	O
based	O
activity	B-Application
recognition	I-Application
work	O
often	O
appears	O
are	O
ICCV	O
and	O
CVPR	O
.	O
</s>
<s>
In	O
vision-based	O
activity	B-Application
recognition	I-Application
,	O
a	O
great	O
deal	O
of	O
work	O
has	O
been	O
done	O
.	O
</s>
<s>
Researchers	O
have	O
attempted	O
a	O
number	O
of	O
methods	O
such	O
as	O
optical	O
flow	O
,	O
Kalman	O
filtering	O
,	O
Hidden	O
Markov	O
models	O
,	O
etc.	O
,	O
under	O
different	O
modalities	O
such	O
as	O
single	O
camera	O
,	O
stereo	B-Algorithm
,	O
and	O
infrared	O
.	O
</s>
<s>
Recently	O
some	O
researchers	O
have	O
used	O
RGBD	B-Algorithm
cameras	I-Algorithm
like	O
Microsoft	B-Algorithm
Kinect	I-Algorithm
to	O
detect	O
human	O
activities	O
.	O
</s>
<s>
Depth	B-Algorithm
cameras	I-Algorithm
add	O
extra	O
dimension	O
i.e.	O
</s>
<s>
Sensory	O
information	O
from	O
these	O
depth	B-Algorithm
cameras	I-Algorithm
have	O
been	O
used	O
to	O
generate	O
real-time	O
skeleton	O
model	O
of	O
humans	O
with	O
different	O
body	O
positions	O
.	O
</s>
<s>
This	O
skeleton	O
information	O
provides	O
meaningful	O
information	O
that	O
researchers	O
have	O
used	O
to	O
model	O
human	O
activities	O
which	O
are	O
trained	O
and	O
later	O
used	O
to	O
recognize	O
unknown	O
activities.Piyathilaka	O
,	O
L.	O
;	O
Kodagoda	O
,	O
S.	O
,	O
"	O
Gaussian	O
mixture	O
based	O
HMM	O
for	O
human	O
daily	O
activity	B-Application
recognition	I-Application
using	O
3D	O
skeleton	O
features	O
,	O
"	O
Industrial	O
Electronics	O
and	O
Applications	O
(	O
ICIEA	O
)	O
,	O
2013	O
8th	O
IEEE	O
Conference	O
on	O
,	O
vol.	O
,	O
no.	O
,	O
pp.567	O
,	O
572	O
,	O
19	O
–	O
21	O
June	O
2013	O
URL	O
:	O
http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6566433&isnumber=6566328Piyathilaka	O
,	O
L	O
.	O
and	O
Kodagoda	O
,	O
S.	O
,	O
2015	O
.	O
</s>
<s>
Human	B-Application
activity	I-Application
recognition	I-Application
for	O
domestic	O
robots	O
.	O
</s>
<s>
With	O
the	O
recent	O
emergency	O
of	O
deep	O
learning	O
,	O
RGB	O
video	O
based	O
activity	B-Application
recognition	I-Application
has	O
seen	O
rapid	O
development	O
.	O
</s>
<s>
Despite	O
remarkable	O
progress	O
of	O
vision-based	O
activity	B-Application
recognition	I-Application
,	O
its	O
usage	O
for	O
most	O
actual	O
visual	O
surveillance	O
applications	O
remains	O
a	O
distant	O
aspiration	O
.	O
</s>
<s>
Based	O
on	O
this	O
observation	O
,	O
it	O
has	O
been	O
proposed	O
to	O
enhance	O
vision-based	O
activity	B-Application
recognition	I-Application
systems	O
by	O
integrating	O
commonsense	O
reasoning	O
and	O
,	O
contextual	O
and	O
commonsense	B-General_Concept
knowledge	I-General_Concept
.	O
</s>
<s>
In	O
vision-based	O
activity	B-Application
recognition	I-Application
,	O
the	O
computational	O
process	O
is	O
often	O
divided	O
into	O
four	O
steps	O
,	O
namely	O
human	O
detection	O
,	O
human	O
tracking	O
,	O
human	B-Application
activity	I-Application
recognition	I-Application
and	O
then	O
a	O
high-level	O
activity	O
evaluation	O
.	O
</s>
<s>
In	O
computer	O
vision-based	O
activity	B-Application
recognition	I-Application
,	O
fine-grained	O
action	O
localization	O
typically	O
provides	O
per-image	O
segmentation	O
masks	O
delineating	O
the	O
human	O
object	O
and	O
its	O
action	O
category	O
(	O
e.g.	O
,	O
Segment-Tube	O
''	O
)	O
.	O
</s>
<s>
Techniques	O
such	O
as	O
dynamic	O
Markov	O
Networks	O
,	O
CNN	B-Architecture
and	O
LSTM	B-Algorithm
are	O
often	O
employed	O
to	O
exploit	O
the	O
semantic	O
correlations	O
between	O
consecutive	O
video	O
frames	O
.	O
</s>
<s>
Geometric	O
fine-grained	O
features	O
such	O
as	O
objective	O
bounding	O
boxes	O
and	O
human	O
poses	O
facilitate	O
activity	B-Application
recognition	I-Application
with	O
graph	B-Algorithm
neural	I-Algorithm
network	I-Algorithm
.	O
</s>
<s>
When	O
activity	B-Application
recognition	I-Application
is	O
performed	O
indoors	O
and	O
in	O
cities	O
using	O
the	O
widely	O
available	O
Wi-Fi	O
signals	O
and	O
802.11	O
access	O
points	O
,	O
there	O
is	O
much	O
noise	O
and	O
uncertainty	O
.	O
</s>
<s>
One	O
of	O
the	O
primary	O
thought	O
of	O
Wi-Fi	O
activity	B-Application
recognition	I-Application
is	O
that	O
when	O
the	O
signal	O
goes	O
through	O
the	O
human	O
body	O
during	O
transmission	O
;	O
which	O
causes	O
reflection	O
,	O
diffraction	O
,	O
and	O
scattering	O
.	O
</s>
<s>
In	O
the	O
paper	O
,	O
and	O
,	O
they	O
have	O
applied	O
the	O
Fresnel	O
model	O
to	O
the	O
activity	B-Application
recognition	I-Application
task	O
and	O
got	O
a	O
more	O
accurate	O
result	O
.	O
</s>
<s>
There	O
are	O
some	O
popular	O
datasets	O
that	O
are	O
used	O
for	O
benchmarking	O
activity	B-Application
recognition	I-Application
or	O
action	B-Application
recognition	I-Application
algorithms	O
.	O
</s>
<s>
One	O
can	O
find	O
applications	O
ranging	O
from	O
security-related	O
applications	O
and	O
logistics	O
support	O
to	O
location-based	B-Application
services	I-Application
.	O
</s>
<s>
Activity	B-Application
recognition	I-Application
systems	O
have	O
been	O
developed	O
for	O
wildlife	O
observation	O
and	O
energy	O
conservation	O
in	O
buildings	O
.	O
</s>
