Indonesian J
ournal of Ele
c
trical Engin
eering and
Computer Sci
e
nce
Vol. 2, No. 3,
Jun
e
201
6, pp. 625 ~ 63
5
DOI: 10.115
9
1
/ijeecs.v2.i3.pp62
5-6
3
5
625
Re
cei
v
ed Ma
rch 3, 2
016;
Re
vised
Ma
y 18, 2016; Accepted Ma
y 29
, 2016
Pattern Recognition of Overhead Forehand and
Backhand in Badminton Based on the Sign of Local
Euler Angle
Muhammad Ilhamdi Rus
y
di
*1
, S
y
amsul Huda
2
, Febdian Rusy
d
i
3
, Muhammad Hadi
Sucipto
1
, Minoru Sasa
ki
4
1
Departme
n
t of Electrical En
gi
neer
ing, An
dal
as Uni
e
rs
it
y
,
L
i
mau Man
i
s, Padan
g Cit
y
,
2
5
1
63, Indo
nesi
a
2
Departme
n
t of Mechan
ical E
ngi
neer
in
g, Andal
as Uni
e
rs
it
y, Limau Man
i
s, Pada
ng
Cit
y, 2
516
3, Indon
esi
a
3
Departme
n
t of Ph
y
s
ics, Airl
a
ngg
a Univ
ersit
y
, Sura
ba
ya C
i
t
y
,
601
11, Indo
nesi
a
4
Departme
n
t of Mechan
ical E
ngi
neer
in
g, Gifu Univ
er
sit
y
, 1-
1 Yana
gi
do, Gifu Cit
y
,
50
1-11
93, Japa
n
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: rilhamd
i
@ft.unan
d.ac.id
A
b
st
r
a
ct
Studyin
g the b
a
d
m
i
n
ton skil
l
base
d
on th
e a
r
m
move
m
ent i
s
a chall
e
n
ge s
i
nc
e the l
i
m
itati
on of the
sensor suc
h
as camera to record the
movement
par
a
m
eter. T
h
is study prop
ose
d
a new
met
hod
to
deter
mi
ne the
pattern of ar
m move
ment for forehan
d an
d
backha
nd stro
kes in ba
dmint
on bas
ed on t
he
sign of the l
o
ca
l Euler a
ngl
e gr
adi
ent from fo
u
r
point
s of right
arm se
gments
.
Each seg
m
e
n
t
s w
a
s identifie
d
by
motio
n
se
ns
or attach
ed to
the d
o
rsa
l
surfa
c
e of
the
ha
nd
(sensor
1), w
r
ist (sensor
2), e
l
bow
(sens
or 3
)
and sh
oul
der
(sensor 4). T
h
ree certifie
d coach
e
s parti
ci
pated i
n
this researc
h
to de
termi
ne the a
r
m
mov
e
me
nt patterns for foreha
nd an
d back
h
a
nd strokes.
Skil
ls in foreh
a
n
d
and b
a
ckh
and
strokes from e
i
gh
t
professi
ona
l p
l
ayers a
n
d
ei
gh
t amateur
pl
ay
ers w
e
re
o
b
ser
v
ed to
deter
mi
ne th
e p
a
ttern.
T
he res
u
lt sh
o
w
ed
that the loc
a
l
Euler a
n
g
l
e ca
n be us
ed to r
e
cog
n
i
z
e
th
e a
r
m
mov
e
ment
pattern. Base
d
on the o
b
serv
ed
patterns, the p
r
ofessio
nal
pla
y
ers had
a hi
gher si
mila
r
i
ty to the coach
e
s
’
patter
n
s th
an thos
e a
m
at
e
u
r
play
ers to the coach
e
s
’
.
Ke
y
w
ords
: ba
dminto
n, local
Euler a
ngl
e, ba
ckhan
d, foreha
nd
Copy
right
©
2016 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Indone
sia
ha
s b
een
rekno
w
n fo
r
pro
d
u
c
ing
world
cl
ass b
adminto
n athlete
s
. B
adminton
is a
prim
ary
sp
ort
cou
r
se in Ind
one
sian ele
m
enta
r
y scho
ol
cu
rri
culum. Alm
o
st eve
r
y si
n
g
le
Indone
sia
n
can play b
a
d
m
inton well. Thus,
an ex
p
e
rime
ntal stu
d
y of sen
s
o
r
movement u
s
ing
motion in
ba
dminton i
s
a
n
interi
sting to
pic
sin
c
e it gi
ves wi
der im
pact to In
don
esia
n ba
dmin
ton
hobbi
est.
Today b
admi
n
ton i
s
playe
d
all ove
r
th
e
worl
d. It wa
s
an exhibitio
n
spo
r
t in
Olym
pic
1972
before it was officially played as the competitiv
e s
p
ort for the firs
t time in the Olympic
1992.
Although it i
s
a famo
us
g
a
me, but
bro
w
si
ng a
nd
se
arching
cite
d
refe
ren
c
e
s
a
bout this ga
me
take a relatively longer tim
e
than other
racket sp
ort
s
su
ch a
s
tenni
s.
Many of the previou
s
st
udie
s
con
d
u
c
t
ed came
ra
to evaluate the badmint
on game.
Wan
g
, Liu and Moffit [1]
recorded u
s
i
ng cam
e
ra
s
a numbe
r of stude
nts play
ing badmint
o
n to
study the
a
r
m an
d trun
k
movement i
n
overhea
d fo
reha
nd
st
r
o
k
e
s f
o
r
so
me
skill
lev
e
l
s
.
T
hey
divided the
seque
nces of
arm m
o
veme
nt into th
ree
step
s; elb
o
w flexion, elbo
w a
nd
hume
r
al
flexion, and
upward flexio
n wh
en som
eone p
e
rfo
r
m
ed the ove
r
h
ead
stro
ke. F
u
rthe
rmoe, th
ey
tennis. Fu
rth
e
rmo
r
e, they also ha
d thre
e s
egm
ents
of trunk mov
e
ment for overhe
ad fore
h
and
stro
ke
s,
whi
c
h comp
rise n
o
tru
n
k a
c
tion
, forward
-
ba
ckward
move
ment a
nd tru
n
k
actio
n
rota
tion.
The result showed that the stud
ents at
advanced skill performed
a
better
action in thi
s
st
roke
comp
ared to anothe
r level.
Mean
while, Z
hu [2] studie
d
the string t
ensi
on for fa
st swi
ng and
angle
d
stri
kin
g
. In this
resea
r
ch eig
h
t
different lev
e
l of stri
ng te
nsio
ns
we
re use
d
.
Some players
were recorded usi
n
g
a
came
ra
whil
e
stri
king
a
sh
uttleco
ck
wit
h
the racket
s of eight lev
e
l stri
ng ten
s
ions. T
he
re
sult
sho
w
e
d
that expert playe
r
s co
uld a
d
just the powe
r
b
e
longi
ng to the strin
g
tensi
on to stro
ke t
h
e
shuttlecock.
The player wi
th low level skill fail
ed to perform fast swing
and angled stri
king
with
variou
s types of string ten
s
ion.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
2, No. 3, Jun
e
2016 : 625
– 635
626
Nag
a
sawa et
al [3]
analyzed the
hu
ma
n motio
n
b
a
sed o
n
the
ba
dminton
sm
a
s
h i
m
age.
The h
u
man
motion in Sp
ace
-
G
wa
s mappe
d
into Space
-
V
us
ing
KL
tr
an
s
f
orm. T
h
is
me
tho
d
cla
ssifie
d
the motion rel
a
te
d to the cente
r
of the
body i
n
to cloo
se lo
op, cu
rve and
line. In [4] th
e
diffference of forehand
overhe
ad
smas
h performed by male
and female players
was
investigate
d
. The a
r
m
wa
s
segm
ented
in
to uppe
r a
r
m,
forea
r
m a
nd
wri
s
t. Oqu
s
camera sy
ste
m
s
recorded th
e
motion
starti
ng fro
m
the
positio
n of
th
e holdi
ng
ra
cket to the
smashi
ng m
o
tion.
Quali
s
ys Tra
ck M
anag
er
softwa
r
e
wa
s used to an
al
yze the motio
n
. The re
sult
sho
w
e
d
that th
e
male
su
bject
has hig
her ra
cqu
e
t g
r
ip vel
o
city
that
the female s
u
bjec
ts
. Us
ing
Qualys
is-MCU500
high
spe
ed
camera, the m
e
thod to
sta
b
i
lize
and
bala
n
ce
the
cent
e
r
g
r
avity of b
ody was stu
d
i
ed
sin
c
e this pl
a
y
s an impo
rta
n
t role in bad
minton athlet
es to re
gulate
the spiki
ng a
c
tion [5].
To tackle th
e limitations
of came
ra such
a
s
workspa
c
e a
r
ea
and complexi
ty of
the
nume
r
ical p
r
oce
s
s, lo
cal
sen
s
o
r
s
were devel
ope
d. Using
ele
c
t
r
ogo
niom
eter, Teu
et al
[6]
prop
osed du
al Euler angl
es to analyse arm move
ment. The b
ody wa
s seg
m
ented into thre
e
se
ction
s
. The
relation
shi
p
betwe
en seg
m
ent velo
city
and the racket velocity
wa
s dete
r
min
ed
usin
g kin
e
ma
tic equatio
ns.
The ra
cket velocity
wa
s a
l
so mea
s
u
r
e
d
using a
n
accele
rom
e
ter
as
the comp
ari
s
on of the simulation re
sult. In
[7],
the sma
s
h stro
ke in badmi
n
ton wa
s stud
ied.
Accel
e
romete
r and ea
rthqu
ake
sen
s
o
r
attache
d
to the badminton racq
uet. The Adaptive Neu
r
o
Fuzzy Inferen
c
e Syste
m
was
develop
ed
to combin
e t
h
e info
rmatio
n from
the
accou
s
tic emi
s
sion
and a
c
celeration inform
atio
n in orde
r to determi
ne the
the ball spe
e
d
.
Ha
stie et al
[8] studi
ed the
develop
ment
of sk
ill a
n
d t
a
ctical
kno
w
l
edge
of stu
d
e
n
ts after
the badminto
n sea
s
o
n
. Th
e re
sult sho
w
ed that a
fter the se
ason, st
udent
s impro
v
ed their abili
ty
to sen
d
the
shuttleco
c
k to
their d
e
si
red
locati
o
n
s. Stu
dents
we
re
more
agg
re
si
ve in hitting t
h
e
shuttle
c
o
ck.
Students cou
l
d de
cid
e
with the
re
aso
n
s
a
nd th
e ta
ctics that th
e
y
want to
u
s
e in
some giv
e
n c
a
se
s.
This resea
r
ch prop
osed t
he of the local Euler an
gl
e gradi
ent to
model the o
v
erhea
d
foreha
nd a
n
d
backha
nd
stroke in
badmi
n
ton. T
he pat
terns
of overhead fo
reh
a
n
d
and b
a
ckha
nd
stro
ke
s we
re determi
ned
from ce
rtified
coache
s. Som
e
playe
r
s
fr
om p
r
o
f
e
s
s
i
ona
l a
n
d
ama
t
eu
r
level participated in this
res
e
arc
h
. The patterns we
re used to inv
e
stigated the similarity of
skil
l
betwe
en play
ers a
nd the coache
s.
2. Motion Se
nsor
In this
re
se
arch, the
m
o
tion
was cacul
a
ted in
3-dim
e
si
onal
sp
ace by
an ine
r
tial
measurement
unit pro
d
u
c
ed by Motio
nnod
e. This
is a comp
act
sen
s
o
r
de
si
gned fo
r hu
man
motion tra
c
ki
ng. Thi
s
10
gram sen
s
or was e
a
sy to
use The
physi
c
al si
ze i
s
3
5
mm x 35 mm
x 15
mm, as sho
w
n by Figure 1
.
The samplin
g rate is 100
Hz an
d the error i
s
about 0
.
5
º
to 2
º
RM
S.
Another rese
arch [9], used
ac
celeromet
e
r as th
e motion se
nsor to detect road di
sea
s
e.
Figure 1. Inertial measu
r
e
m
ent unit by Motionno
d
e
The m
o
tion
wa
s in
dicated
by the
Euler angl
es. E
u
le
r a
ngle
s
are
the
su
ccesive
rotatio
n
to the moving referen
c
e p
o
int. It was the seq
uen
ce
of rotations a
b
out x
1
, y
2
and z
3
c
o
or
d
i
nate
,
as sho
w
n by Figure 2. The
first rotation
about the x-a
x
is by an an
gle
prod
uced the x
1
y
1
z
1
-
axis. The
se
cond rotation
about the y
1
-axis by an
a
ngle
gene
rat
ed the x
2
y
2
z
2
-ax
i
s.
Th
e la
st
rotation is a
b
out the z
2
-axi
s by an angl
e
constru
c
te
d the x
3
y
3
z
3
-ax
i
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Pattern Re
co
gnition of Overhe
ad Fo
reh
and an
d Backhand in Bad
m
inton …
(M. Ilham
di Rusydi)
627
x
y
z
1
x
1
y
1
z
2
y
2
x
2
z
3
z
3
x
3
y
Figure 2. Three rotation
s o
f
coordi
nate
3. Sign of Local Euler An
gle
The sig
n
s of the local Eule
r angle are d
e
term
in
ed from
the slop of the signal
s. Th
ere are
three proba
b
ilities of the sign:
positiv
e (+), negati
v
e (-) an
d st
ationery (0
). Two thre
sh
old
values—po
s
it
ive thre
shold
and
negativ
e thre
sh
old—
were a
pplied
in this
re
sea
r
ch.
The
r
e
were
movement
s if the sign
al was big
g
e
r
tha
n
the po
sitive thre
shol
d o
r
sm
aller th
a
n
the neg
ative
threshold. If the sign
wa
s stationa
ry, the play
e
r
s fi
nish
ed st
ro
ki
ng the racke
t. The thresh
old
values
we
re
alway
s
rene
wed if th
ere
wa
s a
ne
w lo
cal m
a
ximum
or lo
cal
mini
mum poi
nt of
the
local Eule
r an
gle. The local
minimum or l
o
cal maxim
u
m was
calle
d
as refe
ren
c
e
point. The new
threshold val
ue for po
sitive and neg
ative were cal
c
ul
ated by (1) a
nd (2
). Ru
syd
i
et al.,
[10] used
3the threshol
d value of a
biosginal
to i
ndicate
a human
activity. Figure
2 illustrates the process
to determin
e
the sig
n
of local Euler an
gle
.
Figure 3 give
s an
example
of local E
u
ler angle
of arm
movement. In this exa
m
pl
e, there
are
four a
r
ea
s of
the l
o
cal
Euler an
gle:
(a
), (b
),
(c)
and
(d
). Th
ere a
r
e fo
ur po
sitive an
d fo
ur
negative th
re
shol
ds. A
r
ea
a an
d
c h
a
ve
a p
o
sitive
(+)
sign
s of
the
gra
d
ient. T
h
ey are
differe
n
t
from a
r
e
a
(b) whi
c
h
ha
s
a
neg
ative si
g
n
of the
Eule
r a
ngle
gradi
ent. Area
(d)
is the
statio
n
a
ry
area
with its
gradi
ent is
eq
ual to zero. B
a
se
d on thi
s
method, the
pattern of lo
cal Euler a
ngl
e in
this example i
s
“+-+0”. Rusydi et
al.,
[11] briefly introd
uce
d
this
met
hod for the p
a
ttern of the arm
movement re
cog
n
ition sy
stem.
4. Method
In this stu
d
y, the pattern
of local Eul
e
r
a
ngle
was
est
ablished b
a
sed on th
ree
coache
s’
techni
que
s f
o
r fo
reh
and
and
ba
ckhan
d st
ro
ke
s. T
he
coa
c
h
e
s
were
ce
rtified by Ba
dmin
ton
Asso
ciatio
n o
f
Indonesi
a
. Each
coa
c
h
e
s
pe
rf
orm
ed
f
o
reh
and and backh
and stroke
s
ten
tim
e
s.
Eight profe
s
sional playe
r
s (from 1
4
to 1
7
y
ears old
)
and six a
m
at
eur
pl
ayers (about 20 ye
a
r
s
old) were ev
aluated ba
se
d on the simi
larity in t
he pattern of the local Eule
r a
ngle. All of the
coa
c
h
e
s a
nd
players we
re
right-han
ded.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
2, No. 3, Jun
e
2016 : 625
– 635
628
Figure 3. The
process to d
e
termin
e the sign of lo
cal
Euler an
gle
Figure 3. The
local Eule
r a
ngle wave
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Pattern Re
co
gnition of Overhe
ad Fo
reh
and an
d Backhand in Bad
m
inton …
(M. Ilham
di Rusydi)
629
The rig
h
t arm
s
of co
ache
s and playe
r
s
were
segme
n
t
ed into four
se
ction
s
, whi
c
h were
determi
ned b
a
se
d on the
kine
siol
ogy of the human a
r
m [12]. Four gyro se
nsors we
re atta
ch
ed
each on the
dorsal
surfa
c
e of hand
(se
n
so
r 1),
wri
s
t (se
n
sor 2
)
, e
l
bow
(sen
sor
3) an
d sh
oul
der
(se
n
sor 4) a
s
sh
own in
Fig
u
re
4. Thi
s
sensor
mea
s
u
r
ed th
e 3
-
dim
ensi
onal l
o
cal
Euler angl
e
of
each se
gme
n
t. Based on t
he previou
s
rese
arch by Ru
sydi et al. [11], the initial con
d
ition of the
s
e
ns
or
w
a
s
ve
r
y
impo
r
t
a
n
t
in
th
is
s
t
u
d
y
to
im
prove th
e sy
stem
perf
o
rma
n
ce. Th
e initial
po
sition
of the
sen
s
o
r
relative to
th
e world
coord
i
nate
wa
s set to sta
nda
rdi
z
e the
re
sult.
These p
o
sitio
n
s
are given in
Table 1. The
relation
shi
p
b
e
twee
n
the world coo
r
dinat
e to the sen
s
or co
ordinate
is
illustrated by
Figure 5. The symbol
is the an
gle ab
o
u
t the x-axis,
is a
bout y-axis, and
is
about z-axis.
Figure 4. Fou
r
sen
s
o
r
s attache
d
to the right hand
s
Figure 5. Sensor
coo
r
di
nat
e to the world
coordinate
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
2, No. 3, Jun
e
2016 : 625
– 635
630
Table 1. The i
n
itial position
of sen
s
ors
Sensor 1
150°
20°
85°
Sensor 2
110°
25°
120°
Sensor 3
-150°
30°
110°
Sensor 4
120°
25°
90°
Three coaches did
the
foreha
nd and backhand
st
ro
kes ten tim
e
s for
each skill.
Each
sen
s
o
r
had th
ree lo
cal Eule
r angle
s
, so there
we
re totally 12 local
Euler an
gle
s
for four
sen
s
ors.
The data
were analyzed to determi
ne the pattern of
local Eule
r a
ngle ba
se
d o
n
the sign. T
h
e
pattern
s prod
uce
d
by the coache
s we
re
calle
d refe
ren
c
e patte
rn.
The skill
s of eight
p
r
ofe
ssi
onal players and six
am
ateur playe
r
s
were co
mpa
r
e
d
ba
se
d
on thei
r
simil
a
rity to the
coache
s’. The
r
e
were
two method
s
to check
the
skill
of the pl
aye
r
s’.
First, dete
r
mi
ning the
pe
rcenta
ge of u
n
kn
own
arm movement was
p
r
o
p
o
s
ed.
Un
kno
w
n arm
movement
was th
e a
r
m
movement
of players th
at
dismi
s
s the
p
a
ttern p
r
o
d
u
c
ed by
coa
c
h
e
s’.
The hi
gher
percentage of
unknown
arm
movement i
s
the worse
of the pl
ayers’
skill.
Second,
the point p
r
o
duced by the
players was cal
c
ul
ated.
The poi
nt de
pend
ed on th
e perce
ntage
of
players p
e
rfo
r
med the
patt
e
rn
s a
nd the
weig
ht of
tho
s
e pattern
s. The weig
ht
of
the
patte
rn
s was
determi
ned b
y
normali
ze the perce
ntag
e t
he referen
c
e patte
rns from 0 to 1.
5. Result a
n
d Discus
s
io
n
Table 2 shows the pattern
of local Euler
angle
from th
ree certified coache
s for foreha
nd
stro
ke
s. The
pattern wa
s seen fro
m
the local Eule
r angle of 4
sen
s
o
r
locations. Th
e re
sult
sho
w
e
d
that there
we
re th
ree types
of the pattern
in
the x-axis
of all se
nsors. T
here
we
re th
ree
types of the
p
a
ttern o
n
the
y-axis for sen
s
or 1,
2 a
nd
3, yet there
were
only two t
y
pes of
se
nsor
2. The z-axis
of sen
s
or 1, 2
and 4 had al
so thre
e
patte
rns, ex
cept z-axis of sen
s
o
r
4 whi
c
h ha
s 2
pattern
s only. Pattern 1 for each sen
s
o
r
sugge
sted t
he coa
c
h
e
s’
mostly pro
d
u
c
ed patte
rn. The
average probability of all axes for
the pattern 1
of the
a
ll sensors
was
almost 0.6. It has twi
c
e
as
many a
s
patt
e
rn
2. Prob
ab
ility of pattern
3 was
sm
alle
r than
half of t
he patte
rn 2.
The lo
cal E
u
l
e
r
angle
s
of the pattern 1 for f
o
reh
and
stro
kes are illust
ra
ted by Figure
6.
Table 3
sho
w
s the Eul
e
r angle patte
rns an
d the p
r
oba
bility for backh
and
stroke. Th
e
averag
e p
r
ob
ability of the first patte
rn in
bac
kh
and
st
roke, whi
c
h
wa
s ab
out 0.
81, had
a hig
her
prob
ability than the first pa
ttern in foreh
and st
ro
ke. The avera
ge p
r
oba
bility was only about 0.1
3
for pattern 2
and 0.06 for
pattern 3. Th
e x-axis
on sensor 2, 3 an
d 4 had only 1 type of pattern.
The y-axis
on
sen
s
o
r
2 an
d 4 had o
n
ly two pattern
s
and al
so
z-ax
is of se
nsor 3
which ha
d o
n
ly
2
patte
rn
s. Only sensor 1
had 3 types of pattern
for all the axes. The local Eu
ler angl
es of the
pattern 1 for
backh
and
stroke
s are illust
rated by Figu
re 7.
Table 2. The
pattern of forehan
d stro
ke
s
Pattern
Sensor 1
Sensor 2
Sensor 3
Sensor 4
x y z
x y z x y
z x y z
1
-+0
(73%
)
-+0
(47%
)
-+0
(47%
)
-+0
(67%
)
-+0
(47%
)
-+0
(67%
)
+0
(67%
)
-+0
(47%
)
-0
(73%
)
+-0
(40%
)
-+0
(87%
)
-0
(53%
)
2
-+-0
(20%
)
-+-0
(33%
)
-+-0
(40%
)
-+-+0
(20%
)
-+-0
(40%
)
-+-+0
(20%
)
+-+0
(20%
)
-+-+0
(40%
)
-+0
(27%
)
-+-+0
(33%
)
-0
(13%
)
-+0
(33%
)
3
-+-+0
(7%)
-+-+0
(20%
)
-+-+0
(13%
)
-+-0
(13%
)
-+-0
(13%
)
-+-+0
(13%
)
-+-0
(13%
)
+-0
(13%
)
-+-0
(27%
)
-+-0
(13%
)
Table 3. The
pattern of Ba
ckhan
d stro
kes
Pattern
Sensor 1
Sensor 2
Sensor 3
Sensor 4
X
y
z
X
y
z
x
Y
z x y z
1
-+-0
(67%
)
+-+-0
(60%
)
-+-0
(60%
)
-+0
(100%
)
+-+0
(73%
)
-+0
(100%
)
-+0
(100%
)
+-+0
(67%
)
+0
(80%
)
-+0
(100%
)
+-0
(67%
)
-+0
(100%
)
2
-+0
(20%
)
+-+0
(20%
)
-+-+0
(20%
)
+-0
(27%
)
+-0
(20%
)
-+-0
(20%
)
+-+0
(33%
)
3
-+-+-
0
(13%
)
+-+-
+0
(20%
)
-+0
(20%
)
+-+-0
(13%
)
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Pattern Re
co
gnition of Overhe
ad Fo
reh
and an
d Backhand in Bad
m
inton …
(M. Ilham
di Rusydi)
631
Figure 6. Local Euler an
gle
from four se
ns
o
r
s fo
r first
pattern of forehan
d stro
ke
s
Figure 7. Local Euler an
gle
from four se
ns
ors
for firs
t
pattern of back
hand s
t
rokes
Table 4 sugg
ests
the simil
a
rity
between
pr
ofe
s
sion
al
players a
nd
coache
s for fo
reha
nd
skill. The athl
etes we
re a
s
ked to stri
ke
the s
huttleco
ck u
s
ing the f
o
reh
and
skill
ten times. Their
recorded
arm
movement p
a
tterns were
comp
ared to
those
of the coache
s. Arou
nd 37% of th
ei
r
movement
s were re
co
gni
zed a
s
patte
rn 1. In additi
on, the table of figure
s
sh
owe
d
that 27
% of
the moveme
nts were fo
u
nd for p
a
ttern 2, while
21
% for pattern
3. It demonstrated that ab
out
16% of their movement
s were not associate
d
with a
n
y types of the coa
c
h
e
s’.
Table
5
sh
o
w
s the
simil
a
rity bet
wee
n
am
ateur p
l
ayers an
d
coache
s fo
r f
o
reh
and
stro
ke
s. Th
e
distri
bution
of the p
a
ttern for
t
he a
m
ateur playe
r
s whil
e p
e
rf
ormin
g
foreh
and
stro
ke
s is the
same bet
we
en pattern 1
and pattern 2
,
which i
s
27.
5% in averag
e. It is slightly
highe
r tha
n
p
a
ttern 3
which is
only 2
0
%. The ove
r
all
even
chan
ce
for patte
rn
1, 2 an
d 3 i
s
ab
out
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
2, No. 3, Jun
e
2016 : 625
– 635
632
76% for am
ateur pl
ayers, while a
r
ou
nd 24% of
t
heir move
me
nts are un
re
cog
n
ized by
the
system d
e
riv
ed from the coache
s’ patte
rn.
Table 4. Prof
ession
al players’
a
r
m’m
o
vement for forehan
d stro
ke
Pattern
Sensor 1
Sensor 2
Sensor 3
Sensor 4
x y z
x y z
X
Y
z
x
y
z
1
57.5%
10%
12.5%
40%
45%
30%
40%
57.5%
7.5%
57.5%
70%
25%
2
17.5%
52.5%
30%
17.5%
15%
17.5%
25%
20%
52.5%
5%
5%
62.5%
3
15%
17.5%
22.5%
32.5%
32.5%
35%
35%
17.5%
32.5%
10%
Table 5. Ama
t
eur playe
r
s’
arm movem
e
nt for foreha
n
d
stro
ke
s
Pattern
Sensor 1
Sensor 2
Sensor 3
Sensor 4
x
y z x
y z x
Y
z
x y z
1
20%
17%
10%
40%
30%
27%
17%
40 23
43%
53%
10%
2
37%
20%
17%
3%
27%
17%
27%
13 37
30%
17%
87%
3
10%
27%
13%
40%
30%
27%
57%
20
27%
0
Table
6 di
spl
a
ys the
simil
a
rity between
pr
ofe
s
sion
al
players a
nd
coache
s for b
a
ckha
nd
stro
ke
s. The
profe
ssi
onal
players app
ro
ximately prod
uce 4
8
% of pattern 1. It is four times th
an
pattern 2
and about 12 ti
m
e
s than pattern
3. Th
e probability of unrecogni
zed strokes,
while the
profe
ssi
onal
players hit the shuttle
c
o
ck
with foreh
and
skill, is ab
out
one-thi
r
d.
Table 7 p
r
e
s
ents the pe
rcenta
ge of a
m
ateur
pl
aye
r
s’ a
r
m mov
e
ments,
which is the
same a
s
the
coa
c
he
s’ m
o
vement for backh
and
ski
ll. It points out that about 55% of arm
movement is identified by the system, while
4
5
% unre
c
og
nized
arm movem
e
nt. Furtherm
o
re,
Pattern 1 is a
r
oun
d 39% a
nd pattern 2
about 11%.
Table 6. Prof
ession
al players’ a
r
m mov
e
ment for ba
ckha
nd stroke
s
Pattern
Sensor 1
Sensor 2
Sensor 3
Sensor 4
x
y
z X
y
z
x Y
z
X
y
z
1
50%
60%
25%
27.5%
55%
52.5%
65%
20%
57.5%
62.5%
40%
65%
2
20%
10%
25%
17.5%
37.5%
25%
32.5%
3
0%
5%
45%
-
32.5%
-
Table 7. Ama
t
eur playe
r
s’
arm movem
e
nt for backha
nd stro
ke
s
Pattern
Sensor 1
Sensor 2
Sensor 3
Sensor 4
x
y
z
X
y
Z
x Y z
x
y
z
1
37%
50%
25%
27%
33%
53%
65%
23%
47%
43%
37%
2
7%
13%
23%
13%
37%
30%
13%
3
13%
13%
27%
-
0%
-
Figure 8
sh
o
w
s the
avare
ge p
e
rcenta
g
e
of u
n
known
arm
movem
ent from
profession
al
players and
amateu
r players. Unkno
w
n arm mo
ve
ment indicated any arm
movement that
prod
uced di
ssimila
r pattern with co
ach
e
s’ pattern. The blue b
a
r i
s
the un
kno
w
n arm move
ment
for
the profe
ssi
onal
playe
r
s and
the
red b
a
r i
ndi
cates th
e u
n
known a
r
m m
o
vement fo
r
the
amateu
r pla
y
ers. In a glan
ce, the figure indi
cat
e
s that prof
ession
al pla
y
ers ha
d higher
recogni
ze
d a
r
m moveme
nt
than am
ateu
r players fo
r
b
o
th st
roke typ
e
s. T
he u
n
kn
own
movem
e
n
t
for forehand
is less than
the
backhand for both types of pl
ay
ers. T
h
is data illustrated that
profe
ssi
onal and
amate
u
r players
fa
ce
d
more challe
n
ges to lea
r
n b
a
ckha
nd than
forehan
d.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Pattern Re
co
gnition of Overhe
ad Fo
reh
and an
d Backhand in Bad
m
inton …
(M. Ilham
di Rusydi)
633
Figure 8. The
percentag
e o
f
unkno
wn a
r
m movement
for foreh
and
and ba
ckha
n
d
stro
ke
Hastie et al. [8] said that the skill of play
er
s increased after traini
ng. The skill of players
in this rese
arch
correlate
d
to the simila
ri
ty of t
heir skill
to the coa
c
h
e
s’. To eval
u
a
te the simila
rity
of the playe
r
s’ techni
que, t
he patte
rn p
r
obability
p
r
od
uce
d
by the
coache
s were
weig
hted u
s
i
ng
norm
a
lization
.
Table 8 and
Table 9 sho
w
ed the weig
ht for all the
pattern
s. The
players got the
point by multipling their
pat
tern
probability to the weight value of
the pattern. Figure 9
shows the
points for fo
rehan
d
stro
ke
and
Figu
re
1
0
indi
ca
te
s th
e poi
nts fo
r b
a
ckha
nd
stro
ke. Th
e bl
ue
bar
is the the poi
nt for profe
s
i
onal playe
r
s
and the
re
d bar is the p
o
i
nt for amate
u
r playe
r
s. T
h
e
points eval
ua
ted for the en
tire ax
is at four sen
s
ors. T
he maximum
possibl
e poin
t
was 1.0
0
an
d
the minimu
m
wa
s 0.0
0
. T
he bette
r pl
a
y
ers
are th
e
highe
r p
o
int
.
The result
sho
w
e
d
that
in
averag
e profession
al players g
o
t highe
r point
than a
m
ateur pl
ayers for both
stro
ke types.
Table 8. The
weig
ht of pattern for fo
reha
nd stro
ke
s.
Pattern
Sensor 1
Sensor 2
Sensor 3
Sensor 4
x y
z
x
y
z
x
Y
z
X
Y
z
1
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
2
0.27
0.70
0.85
0.30
0.85
0.30
0.30
0.85
0.37
0.83
0.15
0.62
3
0.10
0.43
0.28
0.19
0.28
0.19
0.19
0.28
0.68
0.00
0.25
Table 9. The
weig
hting of pattern for b
a
c
khan
d stro
kes
Pattern
Sensor 1
Sensor 2
Sensor 3
Sensor 4
x
y
z x
y
z
x
Y
z
X Y
z
1
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
2
0.30
0.33
0.33
0.37
0.30
0.25
0.49
3
0.19
0.33
0.33
0.19
0%
10%
20%
30%
40%
50%
60%
Forehand
Backhand
Percentage
Stroke
Unknown
Arm
Mov
emen
t
Profesional
Amateur
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
2, No. 3, Jun
e
2016 : 625
– 635
634
Figure 9. Point of players’
arm movem
e
nt for foreha
n
d
stro
ke
Figure 10. Point of players’
ar
m moveme
nt for backha
nd stro
ke
6. Conclusio
n
There ma
ny
method
s
we
re u
s
ed i
n
p
a
ttern
re
cog
n
ition, such a
s
f
u
sio
n
of lo
cal
Gab
o
r
pattern
s [1
3]
and fu
zzy [14
]. In this
re
se
arch th
e p
a
ttern
of a
r
m
m
o
vement
wa
s dete
r
mine
d
by
local
Euler a
ngle. Th
e
re
sults indi
cate
d that the
lo
cal Eul
e
r an
gle g
r
adi
ent
coul
d b
e
u
s
e
d
to
con
s
tru
c
t the
arm m
o
veme
nt pattern
whi
l
e playing
ba
dminton fo
r f
o
reh
and
an
d
backh
and
sid
e
s.
The initial se
nso
r
po
sition
s set the sce
ne for three
pattern
s of fo
reha
nd
stro
kes for
all axe
s
in
sen
s
o
r
1 a
n
d
2.
Ho
weve
r, z-axis in
sensor
3
and
y-axis in
se
nso
r
4 h
ad
only 2
patterns.
Con
c
e
r
nin
g
b
a
ckha
nd stro
ke, only sen
s
or 1 h
ad th
ree types of
arm motion
s.
This co
nditi
o
n
sho
w
e
d
that
area
of d
o
rsa
l
hand
was the mo
st p
r
ef
erabl
e type o
f
arm m
o
vem
ent for fo
reh
a
nd
and b
a
ckh
a
n
d
strokes. Pat
t
ern 1
had th
e high
est ave
r
age
proba
bility for foreha
n
d
and
ba
ckha
nd
stro
ke
s. Cu
shione
d by the simila
rity of skill, professional playe
r
s
had a hig
her similarity to the
coa
c
h
e
s tha
n
the amateu
r players.
The profe
ssi
onal players
prove
d
that they could imitate the
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
xy
z
x
y
z
xy
z
x
y
z
Sensor
1S
e
n
s
o
r
2S
e
n
s
o
r
3S
e
n
s
o
r
4M
e
a
n
Point
F
o
r
e
hand
st
ro
k
e
Proesional
Amateur
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
xy
z
x
y
z
xy
z
x
y
z
Sensor
1S
e
n
s
o
r
2S
e
n
s
o
r
3S
e
n
s
o
r
4M
e
a
n
Point
Backhand
st
ro
k
e
Proesional
Amateur
Evaluation Warning : The document was created with Spire.PDF for Python.