Database Reference
In-Depth Information
In Fig.
11.3
, the system utilizes a reference template for recognition. Based on
the SC method, a sparse code template (SCT) over the set of gesture instances for
the
i
-th class can be computed as the reference template, i.e.,
N
i
n
=
1
SC
i
,
n
=
n
=
1
SC
i
,
n
SCT
i
(11.8)
N
i
where
N
i
is the number of gesture instances in the
i
-th gesture class, and
SC
i
,
n
is the
sparse code vector of the
n
-th gesture instance belonging to the
i
-th class.
11.5.2
Posture Occurrence
Posture occurrence (PO) is analogous to the popular bag-of-words (BOW) approach
adopted for the task of pattern recognition [
346
,
347
,
350
]. In regard to the BoW
approach used in information retrieval, each posture on the SSOM can be considered
a unique word, while each gesture is a collection of individual words—structured
according to a particular grammar (e.g. set, sequence, etc.). By aggregating the
occurrence of postures in a gesture against the indexed set of nodes on the map,
a histogram can be formed (over a single gesture, or set of similar gestures), thus
forming a template that is used in recognition. In this method, a histogram is formed
for each gesture instance
n
,
PO
n
t
. The value of the
=[
H
(
1
)
,...,
H
(
c
)
,...,
H
(
C
)]
c
-th component is calculated by:
T
t
=
1
ʴ
(
u
(
t
)
−
c
)
,
c
∈{
1
,...,
C
}
H
(
c
)=
(11.9)
where
is the Kronecker delta function, and
T
is the total number of indices in the
sequence of node indices
S
discussed in Eq. (
11.5
).
A reference template for the
i
-th gesture class can be formed by summing over
the set of
PO
i
,
n
in this class:
ʴ
N
i
n
=
1
PO
i
,
n
n
=
1
PO
i
,
n
POT
i
=
(11.10)
N
i
where
N
i
is the number of gesture instances in the
i
-th gesture class and
PO
i
,
n
is the
posture occurrence vector of the
n
-th gesture instance belonging to the
i
-th class.
Search WWH ::
Custom Search