<s>
Sensor	B-General_Concept
fusion	I-General_Concept
is	O
the	O
process	O
of	O
combining	O
sensor	O
data	O
or	O
data	O
derived	O
from	O
disparate	O
sources	O
such	O
that	O
the	O
resulting	O
information	O
has	O
less	O
uncertainty	O
than	O
would	O
be	O
possible	O
when	O
these	O
sources	O
were	O
used	O
individually	O
.	O
</s>
<s>
The	O
term	O
uncertainty	O
reduction	O
in	O
this	O
case	O
can	O
mean	O
more	O
accurate	O
,	O
more	O
complete	O
,	O
or	O
more	O
dependable	O
,	O
or	O
refer	O
to	O
the	O
result	O
of	O
an	O
emerging	O
view	O
,	O
such	O
as	O
stereoscopic	B-Algorithm
vision	O
(	O
calculation	O
of	O
depth	O
information	O
by	O
combining	O
two-dimensional	O
images	O
from	O
two	O
cameras	O
at	O
slightly	O
different	O
viewpoints	O
)	O
.	O
</s>
<s>
Direct	O
fusion	O
is	O
the	O
fusion	O
of	O
sensor	O
data	O
from	O
a	O
set	O
of	O
heterogeneous	B-General_Concept
or	O
homogeneous	B-General_Concept
sensors	O
,	O
soft	O
sensors	O
,	O
and	O
history	O
values	O
of	O
sensor	O
data	O
,	O
while	O
indirect	O
fusion	O
uses	O
information	O
sources	O
like	O
a	O
priori	O
knowledge	O
about	O
the	O
environment	O
and	O
human	O
input	O
.	O
</s>
<s>
Sensor	B-General_Concept
fusion	I-General_Concept
is	O
also	O
known	O
as	O
(	O
multi-sensor	O
)	O
data	B-General_Concept
fusion	I-General_Concept
and	O
is	O
a	O
subset	O
of	O
information	B-General_Concept
fusion	I-General_Concept
.	O
</s>
<s>
Sensor	B-General_Concept
fusion	I-General_Concept
is	O
a	O
term	O
that	O
covers	O
a	O
number	O
of	O
methods	O
and	O
algorithms	O
,	O
including	O
:	O
</s>
<s>
Two	O
example	O
sensor	B-General_Concept
fusion	I-General_Concept
calculations	O
are	O
illustrated	O
below	O
.	O
</s>
<s>
In	O
sensor	B-General_Concept
fusion	I-General_Concept
,	O
centralized	O
versus	O
decentralized	O
refers	O
to	O
where	O
the	O
fusion	O
of	O
the	O
data	O
occurs	O
.	O
</s>
<s>
Complementary	O
features	O
are	O
typically	O
applied	O
in	O
motion	O
recognition	O
tasks	O
with	O
Neural	B-Architecture
network	I-Architecture
,	O
Hidden	O
Markov	O
model	O
,	O
Support-vector	B-Algorithm
machine	I-Algorithm
,	O
clustering	O
methods	O
and	O
other	O
techniques	O
.	O
</s>
<s>
Cooperative	O
sensor	B-General_Concept
fusion	I-General_Concept
uses	O
the	O
information	O
extracted	O
by	O
multiple	O
independent	O
sensors	O
to	O
provide	O
information	O
that	O
would	O
not	O
be	O
available	O
from	O
single	O
sensors	O
.	O
</s>
<s>
Cooperative	O
information	B-General_Concept
fusion	I-General_Concept
can	O
be	O
used	O
in	O
motion	O
recognition	O
,	O
gait	O
analysis	O
,	O
motion	O
analysis	O
,,	O
.	O
</s>
<s>
There	O
are	O
several	O
categories	O
or	O
levels	O
of	O
sensor	B-General_Concept
fusion	I-General_Concept
that	O
are	O
commonly	O
used	O
.	O
</s>
<s>
Sensor	B-General_Concept
fusion	I-General_Concept
level	O
can	O
also	O
be	O
defined	O
basing	O
on	O
the	O
kind	O
of	O
information	O
used	O
to	O
feed	O
the	O
fusion	O
algorithm	O
.	O
</s>
<s>
More	O
precisely	O
,	O
sensor	B-General_Concept
fusion	I-General_Concept
can	O
be	O
performed	O
fusing	O
raw	O
data	O
coming	O
from	O
different	O
sources	O
,	O
extrapolated	O
features	O
or	O
even	O
decision	O
made	O
by	O
single	O
nodes	O
.	O
</s>
<s>
It	O
is	O
the	O
most	O
common	O
sensor	B-General_Concept
fusion	I-General_Concept
technique	O
in	O
many	O
fields	O
of	O
application	O
.	O
</s>
<s>
Data	O
level	O
fusion	O
algorithms	O
usually	O
aim	O
to	O
combine	O
multiple	O
homogeneous	B-General_Concept
sources	O
of	O
sensory	O
data	O
to	O
achieve	O
more	O
accurate	O
and	O
synthetic	O
readings	O
.	O
</s>
<s>
Data	O
level	O
information	B-General_Concept
fusion	I-General_Concept
tends	O
to	O
generate	O
big	O
input	O
spaces	O
,	O
that	O
slow	O
down	O
the	O
decision-making	O
procedure	O
.	O
</s>
<s>
Typically	O
decision	O
level	O
sensor	B-General_Concept
fusion	I-General_Concept
is	O
used	O
in	O
classification	O
an	O
recognition	O
activities	O
and	O
the	O
two	O
most	O
common	O
approaches	O
are	O
majority	O
voting	O
and	O
Naive-Bayes	O
.	O
</s>
<s>
It	O
also	O
allows	O
the	O
combination	O
of	O
heterogeneous	B-General_Concept
sensors	O
.	O
</s>
<s>
One	O
application	O
of	O
sensor	B-General_Concept
fusion	I-General_Concept
is	O
GPS/INS	O
,	O
where	O
Global	O
Positioning	O
System	O
and	O
inertial	O
navigation	O
system	O
data	O
is	O
fused	O
using	O
various	O
different	O
methods	O
,	O
e.g.	O
</s>
<s>
Another	O
example	O
is	O
using	O
the	O
data	B-General_Concept
fusion	I-General_Concept
approach	O
to	O
determine	O
the	O
traffic	O
state	O
(	O
low	O
traffic	O
,	O
traffic	O
jam	O
,	O
medium	O
flow	O
)	O
using	O
road	O
side	O
collected	O
acoustic	O
,	O
image	O
and	O
sensor	O
data	O
.	O
</s>
<s>
In	O
the	O
field	O
of	O
autonomous	O
driving	O
,	O
sensor	B-General_Concept
fusion	I-General_Concept
is	O
used	O
to	O
combine	O
the	O
redundant	O
information	O
from	O
complementary	O
sensors	O
in	O
order	O
to	O
obtain	O
a	O
more	O
accurate	O
and	O
reliable	O
representation	O
of	O
the	O
environment	O
.	O
</s>
<s>
Although	O
technically	O
not	O
a	O
dedicated	O
sensor	B-General_Concept
fusion	I-General_Concept
method	O
,	O
modern	O
Convolutional	B-Architecture
neural	I-Architecture
network	I-Architecture
based	O
methods	O
can	O
simultaneously	O
process	O
very	O
many	O
channels	O
of	O
sensor	O
data	O
(	O
such	O
as	O
Hyperspectral	B-Application
imaging	I-Application
with	O
hundreds	O
of	O
bands	O
)	O
and	O
fuse	O
relevant	O
information	O
to	O
produce	O
classification	O
results	O
.	O
</s>
