Database Reference
In-Depth Information
1. For each gesture class
k
,anHMM
ʻ
k
is built, i.e., the system estimates the model
that optimize the likelihood of the training set observation
vectors for the
k-
th gesture.
2. For each unknown gesture
G
i
to be recognized, the system carries out processing
to measure the trajectories,
Tr
i
=
{
parameters
(
A
,
B
,
ˀ
)
u
1
,...,
u
m
,...,
u
M
},
K
, where
Tr
i
k
=
1
,...,
is
a
Weight Vectors
60
50
40
30
20
10
0
-10
-20
-30
-40
-60
-40
-20
0
20
40
60
W(i,1)
b
c
Weight Vectors
Weight Vectors
40
10
35
30
0
25
-10
20
15
-20
10
-30
5
0
-40
-5
-50
-10
-20
-10
0
10
20
30
40
-30
-20
-10
0
10
20
30
40
50
60
W(i,1)
W(i,1)
d
e
Weight Vectors
Weight Vectors
60
50
-10
40
-12
30
20
-14
10
0
-16
-10
-18
-20
-30
-20
-40
-60
-40
-20
0
20
40
-22
-20
-18
-16
-14
-12
-10
-8
W(i,1)
W(i,1)
Fig. 11.10
2, trained separately by
the data samples belonging to the same class; (
a
) the resulting prototypes of each sub-codebook
overlaying in the input data; (
b
-
e
) the results of classification of input data in each class
Data clustering obtained by multicodebooks, each of size 2
×
Search WWH ::
Custom Search