Biomedical Engineering Reference
In-Depth Information
Fig. 1 The human stick figure models
assumption that only one side of the body is visible during jumping process from a
lateral view.
Each stick is represented by
S ¼
{
p
t
,
p
b
,
a
,
b
} where
p
t
and
p
b
are the stick end
points,
a
and
b
are the lengths of the major and minor axes, respectively. Figure
1
shows an illustration of the stick figure models.
3 The Proposed Approach
3.1
Image Segmentation and RAG Modeling
The mean shift segmentation method [
10
] is used to robustly cluster pixels into
meaningful blobs in each frame. Specifically, video frames are converted into
Luv
color space, and the mean shift vector is defined as
P
i¼
1
x
i
gðð
x
x
i
Þ
2
k
=
h
k
Þ
m
h;G
ð
x
Þ¼
Þ
x
(1)
P
i¼
1
gðð
x
x
i
Þ
2
k
=
h
k
for kernel
G
, where x is the kernel center,
g
is the negative derivative of the kernel
profile, and
h
is the bandwidth (see [
10
] for more details).
After mean shift segmentation, the image is partitioned into different blobs.
Each blob is represented by a feature vector
f ¼
[
a
,
m
L
,
m
u
,
m
v
,
m
x
,
m
y
], where
a
is
the blob size,
m
L
,
m
u
,
m
v
are the modes of the blob in
Luv
color space,
m
x
and
m
y
are