Biomedical Engineering Reference
In-Depth Information
Input : # sensor (
L
)
set
Gfi t
(
L
);
//
first
(
L
) : generate a first candidate solution for cluster number selection
Tem p
=
Infinity
;
while ( G ≠
L
)
//
L
: # sensors
{
*
;
*
: the set of finite mixture model parameter
estimate Θ
// Θ
CV
=
) : discrepancy of consecutive log-likelihood functions
if
valid
(
L
,
G
) // check whether candidate
G
is an appropriate cluster number for
L
then break
Δ
(
G
);
//
Δ
(
⋅
else
if
Temp
>
CV
then
Tem p
=
CV
;
O
t
=
G
;
GG
+ 1
// update the next candidate
G
for
L
}
Output : cluster number (
O
t
)
Fig. 3.3 Brute-force search algorithm to select the group number, This figure shows the Brute-
force search algorithm to select the group number (G) based on the log-likelihood function
sequential data because of the lack of adaptive parameter for the time sequential
data. To overcome such static grouping limitations, we introduce distributed
grouping using the multi-channel (MC) selection for the time sequence data of
distributed body sensors in the next Chapter.
3.3 Proposed Grouping Criteria with Distributed Sensors
The distributed measurements can provide more reliable estimation of target
tracking. The motivation of this Chapter is to prepare interactive relationships with
distributed sensory data for clustering, i.e., how to collaborate with distributed
measurements to achieve better performances compared to the single measure-
ment. In
Sect. 3.3.1
, we will show how to initialize the hyper-parameter presenting
a hypothetical prior probability for background knowledge, and can select the
collaborative cluster number using EM iteration. In
Sect. 3.3.2
, we will calculate
switching probability representing the conditional transition probability from
channel a to channel b within a cluster number.
3.3.1 Collaborative Grouping with Distributed Body Sensors
The cluster number selection using GMM works well in the distributed means
model as well as in the static data model. But it only works within limitation of the
initial dataset. The tracking estimate system with distributed body sensors has time
sequential data. That means the measured information from each sensor can be
Search WWH ::
Custom Search