Digital Signal Processing Reference
In-Depth Information
4.2.4 Clustering Procedure
The unsupervised hierarchical clustering is applied to the feature vectors
x
j
(
j
¼
1,
2,
,
n
). The clustering algorithm is listed below:
...
3. Regard each feature vector
x
j
as each cluster
C
j
, i.e., each cluster consists only
of one feature vector. Calculate the dissimilarity
D
p
,
q
between any two clusters
C
p
and
C
q
by using the following dissimilarity measure:
2
T
R
p;q
1
D
p; q
¼ x
p
x
q
¼ðx
p
x
q
Þ
ðx
p
x
q
Þ
(4.14)
R
p;q
1
Where
R
p; q
1
R
p
1
R
q
1
:
¼
þ
(4.15)
4. Unify two clusters
C
x
and
C
y
which shows the smallest
D
x
,
y
. The unified cluster
is denoted by
C
r
. If all clusters are unified, terminate the algorithm. Otherwise,
go to step 3.
5. Calculate the dissimilarity
D
x
,
y
between
C
r
and
C
t
for all
t
(
t
6¼
r
) by using the
following dissimilarity measure:
X
X
2
n
r
n
t
n
r
þ
D
r;t
¼
x
i
r
x
i
t
(4.16)
R
i
r
;i
t
1
n
t
:
x
i
r
2C
r
x
i
t
2C
t
6. Where
n
r
and
n
t
are numbers of feature vectors belonging to clusters
C
r
and
C
t
,
respectively. Go to step 2.
After this clustering procedure, the classification of the feature vector space is
achieved together with a dendrogramwhich shows the hierarchical classification for
different number of modes. Since the transformation from the feature vector (
)
space to the original observed data (
y
,
r
) space is straightforward, the mode segmen-
tation of the observed data is obtained together with the hierarchical structure.
Note that once mode segmentation of the data is achieved, the identification of
the parameters
x
y
i
and the partitions of the subspaces
R
1
,
,
R
s
in the PWARX
...
model [
2
] is straightforward.
4.3 Analysis of Driving Behavioral Data
4.3.1 Driving Environment
In this chapter, the following driving environment on the highway was designed on
the driving simulator which provides a stereoscopic immersive vision [
7
].
Search WWH ::
Custom Search