Co
m
pu
ter Sci
ence a
nd Inf
or
mat
i
on
Tec
h
no
lo
gies
Vo
l.
1
, No
.
3
,
Novem
ber
2020
, p
p.
11
6
~
12
0
IS
S
N:
27
22
-
3221
,
DOI: 10
.11
591
/
csi
t.v
1i
3
.p
11
6
-
12
0
116
Journ
al h
om
e
page
:
http:
//
ia
esprime
.com/i
ndex.
php/csit
Hand
gesture re
cogniti
on u
sing machine
learning
algorithm
s
Ab
hishe
k B
1
, Kan
ya Kris
hi
2
, M
e
ghan
a M
3
, Mohamme
d
Daan
i
yaal
4
, A
nup
ama H
S
5
1, 2, 3,4
B.
E
,
Com
pute
r
Sc
ie
nc
e and
Engi
n
ee
ring
,
B
MS
Instit
ute of Te
chno
log
y
,
Ba
ngal
ore
,
Ind
ia
5
As
socia
te
Profe
ss
or,
BMS
Institute
of
T
ec
hnolo
g
y
,
Bang
al
or
e, I
ndia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
A
pr
24
, 20
2
0
Re
vised
Jun
14
, 20
2
0
Accepte
d
J
un
2
9
, 20
2
0
Gesture
re
cogni
t
ion
is
an
emergi
ng
topi
c
in
tod
a
y
’s
t
ec
hnolog
ie
s
.
The
m
ai
n
foc
us
of
thi
s
is
to
rec
ogni
ze
the
hu
m
an
gesture
s
using
m
at
hemati
c
al
al
gorit
hm
s
for
hum
an
com
pute
r
intera
ct
io
n.
Onl
y
a
few
m
odes
of
hum
a
n
-
computer
int
er
ac
t
ion
(HCI)
ex
ist,
th
e
y
are:
through
k
e
y
boar
d,
m
ouse,
tou
ch
scre
ens
etc.
Ea
ch
of
th
ese
de
vic
es
h
as
their
o
wn
li
m
it
ations
when
it
comes
to
a
dapt
ing
m
or
e
ver
satile
h
ard
wa
re
in
computer
s.
Gesture
r
ec
ogni
ti
on
is
on
e
of
t
h
e
essential
te
chn
ique
s
to
bu
i
ld
user
-
fri
endly
i
nte
rfa
ce
s.
Us
ual
l
y
g
e
stures
c
an
b
e
origi
na
te
d
from
an
y
bod
ily
m
oti
on
or
state
,
but
comm
onl
y
origi
na
te
from
the
fa
ce
o
r
hand.
Gesture
r
e
cogni
ti
on
ena
bl
e
s
users
to
in
te
r
a
ct
with
the
d
evice
s
wi
thout
ph
y
sic
al
l
y
tou
ch
ing
th
em.
Th
is
p
ape
r
d
esc
rib
es
h
ow
hand
gestur
e
s
are
traine
d
to
p
erf
orm
c
ert
a
i
n
a
ct
ions
li
k
e
sw
it
chi
ng
p
age
s,
sc
roll
ing
up
or
do
wn
in
a
pag
e.
Ke
yw
or
d
s
:
Gestu
re
recog
ni
ti
on
Hu
m
an
com
pute
r
interact
io
n
User
-
f
rien
dly i
nterf
ace
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
B
Y
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
Abhishe
k B
,
B.E, C
om
pu
te
r
Scie
nce a
nd E
ng
i
neer
i
ng,
BM
S I
nst
it
ute o
f
Tech
nolo
gy,
Ban
galor
e
, In
dia
.
Em
a
il
:
anu
pam
ahs@bm
sit
.in
1.
INTROD
U
CTION
Gestu
re
rec
ogni
ti
on
is
a
te
chn
i
qu
e
w
hich
is
use
d
to
unde
rstand
a
nd
analy
ze
the
hum
an
body
la
ngua
ge
and
i
nteract
with
the
us
er
acc
ordin
gly.
T
his
in
tu
rn
hel
ps
in
bu
il
di
ng
a
bri
dge
betwee
n
the
m
achine
an
d
t
he
us
e
r
to
com
m
un
ic
at
e
with
eac
h
othe
r.
Gestu
re
rec
ogniti
on
is
us
e
fu
l
in
proce
ssing
the
in
f
or
m
a
ti
on
wh
ic
h
can
no
t
be
conveyed
t
hro
ugh
s
peech
or
te
xt.
Gestu
re
s
are
the
sim
plest
m
eans
of
com
m
un
ic
at
i
ng
s
om
et
hin
g
that
is
m
eaningfu
l.
T
his
pa
per
in
volves
im
ple
m
entat
ion
of
t
he
sys
tem
that
ai
m
s
t
o
desi
gn
a
visi
on
-
based
hand
gestu
re
recog
niti
on
sys
tem
with
a
high
c
orrect
detect
ion
rate
al
ong
with
a
high
-
perform
ance
crit
er
ion
,
wh
ic
h
ca
n
work
in
a
real
ti
m
e
Hu
m
an
C
om
pu
te
r
I
nteracti
on
syst
e
m
with
ou
t
ha
vi
ng
a
ny
of
t
he
lim
it
a
ti
on
s
(
gloves,
un
i
for
m
backg
rou
nd
et
c.)
on
the
us
e
r
en
vi
r
on
m
ent.
The
syst
em
can
be
def
i
ned
usi
ng
a
fl
ow
c
ha
rt
that
c
on
ta
in
s
three
m
ai
n
ste
ps
, t
he
y are:
Lear
ning
, D
et
ect
io
n,
Re
cogniti
on
as s
how
n
i
n
Fi
gure
1
.
Learn
i
ng
:
It in
vo
l
ves
tw
o aspect
s
su
c
h
a
s
Trainin
g
datas
et
:
This
is
t
he
dataset
that
c
onsist
s o
f
dif
fere
nt
ty
pes
of h
a
nd
ge
sture
s
tha
t
are u
se
d
to
tr
ai
n
the syst
em
b
ased
on which
th
e syst
e
m
p
erfo
rm
s the acti
on
s
.
Feat
ur
e
E
xtrac
ti
on
:
It
in
vo
l
ve
s
dete
rm
ining
the
centr
oid
t
hat
div
ide
s
th
e
im
age
into
t
w
o
halves
at
it
s
geo
m
et
ric Centre.
Detect
ion
Ca
ptu
re
sce
ne: Ca
ptures t
he
i
m
ages th
r
ough
a w
e
b
cam
era,
which is
u
se
d as an
in
pu
t t
o
t
he
syst
em
.
Pr
e
processin
g:
Im
ages
that
ar
e
captu
red
t
hro
ugh
the
we
bca
m
are
com
par
ed
with
the d
at
aset
to
rec
ogniz
e
the v
al
id
han
d m
ov
e
m
ents that are
need
e
d
t
o per
form
the r
e
qu
i
red act
ions.
Evaluation Warning : The document was created with Spire.PDF for Python.
Com
pu
t. Sci.
I
nf. Tec
hnol.
Han
d gesture
r
ecognit
ion
us
in
g ma
c
hin
e
lear
ning
algorit
hm
s
… (
Abh
is
hek
B
)
117
Hand
Detect
io
n:
The
re
qu
i
r
e
m
ents
for
ha
nd
detect
io
n
i
nvolv
e
the
in
pu
t
im
age
fro
m
the
web
ca
m
.
The
im
age
s
hould
be
fetc
hed
with
a
s
peed
of
20
f
ram
es
pe
r
sec
ond.
Dista
nce
s
houl
d
al
s
o
be
m
ai
ntained
betwee
n
t
he
ha
nd
a
nd
the
cam
era.
A
ppr
ox
im
at
e
distance
tha
t
shou
l
d
be
bet
ween
hand
t
he
ca
m
era
is
ar
ou
nd
30 to 1
00 cm
. Th
e
vid
e
o
i
nput
is stor
e
d fr
am
e b
y
fr
a
m
e into
a m
at
rix
after
pr
e
processi
ng.
Re
cogniti
on
Gestu
re
Re
c
og
niti
on
:
The
nu
m
ber
of
fi
ng
e
r
s
prese
nt
in
the
ha
nd
gestu
re
i
s
dete
rm
ined
by
m
aking
us
e
of
def
ect
po
i
nts
pr
ese
nt
in
t
he
gestu
re.
T
he
resu
lt
ant
gestu
re
obta
ine
d
is
fed
t
hro
ugh
a
3D
im
ensional
Conv
olu
ti
onal
Neural
Netw
ork
c
onsecuti
vel
y t
o
rec
ognize
the curre
nt
ges
ture.
Perfo
rm
ing
act
ion
:
T
he
recog
nized
gest
ur
e
i
s
us
e
d
as
an
in
pu
t
t
o
perform
the
act
ions
re
quire
d
by
t
he
use
r.
Figure
1. Flo
w
char
t
hum
an
com
pu
te
r
inte
ra
ct
ion
2.
LIT
ERATUR
E SU
RV
E
Y
The
im
ple
m
entat
ion
is
div
i
de
d
int
o
fou
r
m
ai
n
ste
ps
:
1.
I
m
age
En
ha
nce
m
ent
an
d
Se
gm
entat
ion
2.
Or
ie
ntati
on
D
et
ect
ion
3.
F
e
at
ur
e
E
xtracti
on
4
[
1]
.
Cl
ass
ific
at
ion
.
T
his
w
ork
was
f
oc
us
se
d
on
ab
ove
f
our
cat
egories
but
m
ai
n
lim
it
a
ti
on
was
c
hange
of
c
olo
r
was
ha
pp
e
ni
ng
ve
ry
rap
i
dly
by
the
change
in
the
diff
e
re
nt
li
gh
ti
ng
co
ndit
ion,
wh
ic
h
m
ay
cause
e
rror
or
e
ven
fail
ures.
F
or
exam
ple,
du
e
to
in
suffici
en
t
li
gh
t
co
ndit
io
n,
t
he
existe
nce
of
ha
nd
a
rea
is
not
de
te
ct
ed bu
t
th
e
non
-
sk
i
n re
gions
are
m
ist
aken
f
or
t
he ha
nd
area
beca
us
e
of
sam
e
colo
r
[2]
.
I
nv
olv
es
th
ree
m
ai
n
ste
ps
f
or
hand
ge
sture
recog
niti
on
syst
e
m
:
1.
Segm
entat
ion
2.
Feat
ur
e
Re
pr
ese
ntati
on
3.
Re
c
ogniti
on
Tec
hn
i
qu
es
.
The
syst
em
is
base
d
on
H
an
d
gestu
re
rec
og
niti
on
by
m
od
el
ing
of
the
ha
nd
in
s
pa
ti
al
do
m
ai
n.
The
sy
ste
m
us
es
va
rio
us
2D
and
3D
ge
om
e
tric
an
d
non
-
ge
om
et
ric
m
od
el
s
for
m
od
el
ing
.
It
ha
s
use
d
F
uzzy
c
-
Me
ans
cl
us
te
ring
al
go
rithm
wh
ic
h
res
ulted
in
an
acc
ur
acy
of
85.
83%.
T
he
m
ai
n
dr
a
w
back
of
th
e
syst
em
is
it
does
not
c
onsid
er
gestu
re
rec
ogniti
on
of
te
m
poral
s
p
ace,
i.e
.
m
otion
of
ges
tures
and
it
is
unable
to
cl
assify
i
m
a
ges
with
c
om
plex
bac
kgr
ound
i.e.
w
her
e
the
re
are
oth
e
r
ob
j
ec
ts
in
the
scene
with
the
ha
nd
obj
e
ct
s
[3]
.
T
his
s
urvey
f
oc
us
es
on
the
hand
ge
sture
recog
niti
on
us
in
g
dif
f
eren
t
ste
ps
li
ke
data
acqu
isi
ti
on,
pr
e
-
proce
ssin
g,
s
egm
entat
ion
a
nd
s
o
on.
S
uit
able
in
put
de
vi
ce
shou
l
d
be
sel
ect
ed
f
or
th
e
data
acqu
isi
ti
on.
Th
ere
a
re
a
num
ber
of
in
pu
t
de
vi
ces
for
data
a
cqu
isi
ti
on.
Som
e
of
t
hem
are
data
gloves
,
m
ark
er
,
and
ha
nd
im
a
ges
(
from
webcam
/Kinect
3D
S
e
nsor
).
B
ut
the
lim
i
ta
ti
on
with
this
w
ork
was
c
hang
e
in
the
il
lu
m
inati
on
,
r
otati
on
a
nd
ori
entat
ion
,
scal
in
g
pro
blem
and
sp
eci
al
hard
w
are
w
hich
is
pret
ty
costli
er
[4]
.
The
syst
e
m
i
m
ple
m
entat
ion
is
di
vid
ed
int
o
t
hree
phases:
1.
Hand
gest
ur
e
recog
niti
on
us
i
ng
kin
et
ic
ca
m
era
2.
Algorithm
s
for
ha
nd
detect
io
n
recog
niti
on
3.
Ha
nd
ge
sture
rec
ogniti
on.
The
li
m
i
ta
ti
on
her
e
is
that
t
he
ed
ge
detect
ion
an
d
s
egm
entat
ion
al
gorithm
s
us
ed
her
e
are
not
very
eff
ic
ie
nt
w
he
n
c
om
par
ed
t
o
neural
netw
ork
s.
T
he
dataset
b
ei
ng c
on
si
d
ere
d he
re
is ver
y
sm
all an
d ca
n be
used
to dete
ct
v
e
ry
few sig
n gestu
r
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2722
-
3221
Com
pu
t. Sci.
I
nf. Tec
hnol.
,
V
ol.
1
, N
o.
3
,
N
ov
em
ber
20
20
:
117
–
12
0
118
The
Syst
em
arch
it
ect
ur
e
c
on
s
ist
s
of:
1.
Im
age
ac
qu
isi
ti
on
2.
Se
gm
entat
ion
of
hand
re
gion.
3.
Dista
nc
e
trans
form
s
m
e
thod
for
gestu
re
r
eco
gnit
ion
[
5]
.
T
he
li
m
i
t
at
ion
s
of
this
syst
e
m
i
nv
ol
ve
1.
T
he
num
ber
s
of
gestu
res
that
ar
e
rec
ognize
d
ar
e
le
ss
2.
The
ge
stures
r
eco
gniz
ed
we
re
not
use
d
to
co
ntr
ol
any
a
pp
li
cat
io
ns
[
6]
.
In
t
his im
ple
m
entat
ion
t
her
e
are th
ree m
ai
n
al
gorithm
s that
are use
d: 1.
Viola and
jon
e
s
Algorithm
. 2
. C
onve
x
Hu
ll
Al
go
rith
m
.
3.
The
Ad
a
Boo
st
base
d
le
arn
i
ng
Algo
rithm
.
The
w
ork
was
acc
om
plis
hed
by
trai
ning
a
set
of
featur
e
set
w
hi
ch
is
local
c
on
t
our
se
qu
e
nce
.
The
li
m
it
at
ion
s
of
t
his
syst
em
are
that
it
re
qu
i
res
tw
o
set
s
of
i
m
ages
for
cl
assifi
cat
ion.
On
e
is
t
he
po
sit
ive
set
th
at
co
ntains
t
he
re
qu
ir
ed
im
ag
es,
the
ot
her
is
the
ne
gative
s
et
that
con
ta
in
s
c
on
tra
dicti
ng
im
ages
[7]
.
The
syst
em
i
m
ple
m
entation
c
onsist
s
of
three
c
om
po
ne
nts:
1.
Ha
nd
de
te
ct
ion
2.
Gest
ur
e
rec
ogniti
on
3.
H
uma
n
-
C
om
pu
te
r
I
nteracti
on
(H
C
I)
.
It
has
im
ple
m
ente
d
the
f
ollow
i
ng
m
et
ho
dolo
gy:
1.
t
he
in
put
im
age
is
pr
e
proce
ssed
a
nd
the
ha
nd
detect
or
tri
es
to
filt
er
out
the
ha
nd
f
r
om
the
in
pu
t
im
age
2.A
CNN
cl
assifi
e
r
is
e
m
plo
ye
d
to
rec
ognize
ge
stures
from
the
processe
d
im
a
ge,
wh
il
e
a
Kal
m
an
Fil
te
r
is
us
ed
to
est
i
m
at
e
the
posit
ion
of
t
he
m
ou
s
e
c
ur
s
or.
3.
The
recog
niti
on
a
nd
est
im
a
ti
on
resu
lt
s
a
re
s
ubm
i
tt
ed
to
a
c
ontr
ol
Ce
ntre w
hich d
eci
des
th
e act
ion t
o be
take
n.
On
e
of the li
m
i
ta
ti
on
s
of this
s
yst
e
m
is that i
t reco
gniz
es
on
l
y t
he
sta
ti
c
i
m
ages
[8]
.
T
his
im
ple
m
entat
ion
f
oc
use
s
on
detect
io
n
of
ha
nd
gest
ur
es
us
i
ng
j
a
va
an
d
neural
net
works.
It
is
di
vid
e
d i
nto
t
wo
phases:
-
1.
Detect
io
n m
odule
us
i
ng
j
av
a
w
he
re
i
n
the
hand
is
detect
e
d
us
i
ng
backg
r
ound
su
bt
racti
on
a
nd
co
nversi
on
of
vid
e
o
feed
int
o
HS
B
vi
deo
fee
d
th
us
detect
in
g
s
kin
pi
xels.
2.
T
he
sec
ond
m
odul
e
is
the
pre
dicti
on
m
od
ule;
a
c
onvoluti
onal
ne
ur
al
netw
ork
is
us
e
d.
The
in
put
fee
d
im
age
is
gain
ed
from
Java.
The
in
pu
t
im
a
ge
is
fed
i
nto
t
he
ne
ur
al
netw
ork
an
d
is
anal
yz
ed
with
res
pe
ct
to
the
datas
et
i
m
ages.
On
e
of
the
li
m
it
at
ion
s of t
his syste
m
is that i
t requires
soc
ket pr
ogram
m
ing
i
n order
to c
onnect
j
a
va
a
nd p
yt
ho
n
m
odules.
3.
IMPLEME
N
TATION
A
ha
nd
ge
sture
rec
ogniti
on
s
yst
e
m
was
de
ve
lop
e
d
t
o
ca
pt
ur
e
the
ha
nd
ge
stures
bei
ng
pe
rfor
m
ed
by
the
us
er
a
nd
to
con
t
ro
l
a
c
om
pu
te
r
syst
em
based
on
t
he
i
ncom
ing
in
f
or
m
at
i
on.
Ma
ny
of
th
e
existi
ng
syst
em
s
in
li
te
ratur
e
ha
ve
im
ple
m
ented
gestu
re
rec
ogni
ti
on
us
in
g
onl
y
sp
at
ia
l
m
odel
li
ng
,
i.e.
rec
ogniti
on
of
a
sing
le
gestu
re
a
nd
not
te
m
po
ral
m
odel
li
ng
i.e
. r
ec
ogniti
on of
m
otion of g
est
ures.
Also
,
t
he
e
xisti
ng
syst
em
s
have
not
been
im
ple
m
ented
i
n
real
ti
m
e
,
t
hey
us
e
a
pre
ca
ptu
re
d
im
age
as
a
n
i
nput
f
or
gest
ur
e
rec
og
niti
on
.
T
o
ov
e
r
com
e
these
existi
ng
pro
blem
s
a
ne
w
arc
hitec
ture
has
bee
n
devel
op
e
d
w
hich
aim
s
to
design
a
vision
-
base
d
ha
nd
gestu
r
e
recog
ni
ti
on
syst
em
with
a
hi
gh
co
rrec
t
detect
ion
r
at
e
al
ong
with
a
high
-
pe
rfor
m
ance
c
rite
rio
n,
wh
ic
h
can
w
ork
in
a
real
tim
e
HCI
syst
e
m
witho
ut
hav
in
g
a
ny
of
the
m
entione
d
stric
t
lim
i
ta
tio
ns
(
gloves,
unif
or
m
backg
rou
nd
et
c.)
on
the
us
e
r
env
i
ronm
ent.
The
desig
n
is
com
po
sed
of
a
hum
an
com
pu
te
r
interact
io
n
syst
e
m
wh
ic
h uses
ha
nd
gestu
res
as
in
pu
t
f
or
c
omm
u
nicat
ion
as s
how
in
Fig
ure
2
.
Figure
2. Desi
gn of t
he pr
opose
d HCI
syst
em
Evaluation Warning : The document was created with Spire.PDF for Python.
Com
pu
t. Sci.
I
nf. Tec
hnol.
Han
d gesture
r
ecognit
ion
us
in
g ma
c
hin
e
lear
ning
algorit
hm
s
… (
Abh
is
hek
B
)
119
Inp
ut to
the syst
e
m
is fr
om
the w
eb
cam
era o
r
a
pr
e
recor
de
d
vi
deo
se
quen
ce. Late
r
it
d
et
ect
s the sk
i
n
colo
r
by
usi
ng
an
a
da
ptive
al
gorithm
in
the
be
ginnin
g
of
the
fr
am
es.
F
or
th
e
cu
rr
e
nt
us
er
sk
in
col
or
has
to
be
fixe
d
base
d
on
the
li
ghti
ng
an
d
cam
era
pa
ra
m
et
er
and
c
ondi
ti
on
.
O
nce
it
is
bee
n
fixe
d,
ha
nd
is
local
iz
e
d
with
a
histo
gr
am
cl
ust
ering
m
et
ho
d.
T
hen
a
m
achine
le
a
rn
i
ng
al
gorithm
is
been
use
d
to
detect
the
ha
nd
gest
ur
es
i
n
consecuti
ve
f
r
a
m
es
to
disti
nguis
h
t
he
c
urr
ent
ge
sture
.
T
hese
gestu
res
are
us
ed
a
s
a
n
input
for
a
c
om
pu
te
r
app
li
cat
io
n
as
sh
ow
n
in
Fi
gur
e 3
.
T
he
syst
e
m
is divid
e
d
i
nt
o
3 s
ub
syst
em
s:
3.1.
Ha
n
d
an
d
Mo
tion De
tecti
on
The
Web
-
cam
era
capt
ur
es
th
e
ha
nd
m
ov
em
ent
an
d
pro
vide
s
it
as
input
t
o
O
pe
nCV
an
d
Tens
orFlo
w
Object
detect
or.
E
dge
detect
ion
an
d
sk
i
n
de
te
ct
ion
are
perf
or
m
ed
to
obta
in
the
bounda
ry
of
the
hand.
T
his
is
then sent t
o
the
3D CN
N.
3.2.
Datase
t
Dataset
is
us
e
d
f
or
t
rainin
g
t
he
3D
C
NN.
T
wo
ty
pe
s
of
da
ta
set
s
are
bein
g
use
d
–
on
e
f
or
t
he
ha
nd
detect
ion
an
d
the
oth
er
for
t
he
m
otion
or
gestu
re
de
te
ct
ion.
Ha
nd
dete
ct
ion
us
es
E
G
O
dataset
,
M
ot
ion
or
Gestu
re Rec
ogniti
on
us
es
Jest
er
dataset
.
3.3.
3D C
NN
CNN
’s are
a
cl
ass
of
dee
p l
ea
rn
i
ng
ne
ur
al
ne
tworks
us
e
d f
or
analy
zi
ng
vi
de
os
a
nd
im
ages. It
co
ns
ist
s
of
seve
ral
la
ye
r
s
–
i
nput
la
ye
r,
hidden
la
ye
rs
a
nd
out
put
la
ye
r
.
It
perform
s
back
propagati
on
for
bette
r
acc
uracy
and
e
ff
ic
ie
ncy
.
It
pe
rfo
rm
s
trai
ning
a
nd
ver
ific
at
io
n
of
the
rec
ogniz
ed
gest
ur
es
a
nd
hu
m
an
c
om
pu
te
r
interact
ions
ta
ke
place
–
tu
rn
i
ng
of
the
pag
es
,
zoo
m
ing
in
a
nd
zo
om
ing
out.
The
interact
io
ns
wit
h
t
he
co
m
pu
te
r
ta
ke
place
w
it
h t
he help
of Py
Au
t
oGUI o
r S
yst
e
m
Calls.
Figure
3. Syst
em
r
ecog
nize
d hand
gest
ur
es
4.
CONCL
US
I
O
N
The
im
po
rtanc
e
of
ge
sture
r
ecognit
ion
li
es
in
bu
il
di
ng
e
f
fici
ent
hu
m
an
-
m
achine
i
ntera
ct
ion
.
T
his
pap
e
r
des
cribe
s
how
the
im
pl
e
m
entat
ion
of
t
he
syst
e
m
is
do
ne
base
d
up
on
the
im
ages
ca
pt
ur
e
d.
Ha
nd
det
ect
ion
is
done
us
in
g
O
pen
C
V
a
nd
Te
ns
orFl
ow
ob
j
e
ct
detect
or
.
A
nd
furthe
r
it
is
e
nh
a
nce
d
for
i
nterpreta
ti
on
of
ge
sture
s
by the c
om
pu
t
er to pe
rfor
m
acti
on
s li
ke
sw
it
chin
g
the
p
a
ge
s,
scr
olli
ng
up
or down t
he pa
ge.
ACKN
OWLE
DGE
MENT
This work
is
done
,
s
uper
vise
d
a
nd
s
upporte
d
by
the
st
ud
e
nt
s
an
d
fac
ulty
m
e
m
ber
s
of
t
h
e
De
par
tm
ent
of Com
pu
te
r S
ci
ence a
nd Eng
ineerin
g, BM
S
In
sti
tute
of Tec
hnology, Ba
ng
al
or
e.
REFERE
NCE
S
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2722
-
3221
Com
pu
t. Sci.
I
nf. Tec
hnol.
,
V
ol.
1
, N
o.
3
,
N
ov
em
ber
20
20
:
117
–
12
0
120
[1]
.
M.
Panwar
and
P.
Singh
Mehra
,
“
Hand
gesture
rec
ogni
ti
on
for
hum
an
compute
r
intera
c
ti
on
,
”
2
011
Inte
rnation
al
Confe
renc
e
on
I
mage
Informatio
n
Proce
ss
ing
,
Sh
imla,
pp
.
1
-
7
,
20
11.
[2]
.
Rafi
qul
Z
aman
Khan
and
Noor
Adnan
Ibra
h
eem
.
“
Com
par
it
iv
e
Stud
y
of
Han
d
Gesture
R
ec
o
gnit
ion
S
y
st
em
.”
Inte
rnational
Co
nfe
renc
e
of
Ad
va
nce
d
Comput
er
Sci
en
ce
&
Infor
mation
Techno
lo
gy
,
2012
.
[3]
.
Arpita
R
a
y
Sark
ar,
G
.
San
y
a
l,
S.
Majumder
.
“
Han
d
Gesture
Rec
og
nit
ion S
y
stems
:
A
Surve
y
.”
Inte
r
nati
onal
Journa
l
of
Computer
Ap
pli
cations
,
vo
l. 7
1,
n
o
.
15,
pp.
25
-
37,
Ma
y
2013.
[4]
.
Manjuna
th
AE,
Vijay
a
Kum
ar
B
P,
Ra
je
sh
H
.
“
Com
par
at
ive
S
tud
y
of
Hand
G
esture
R
ec
ogni
tion
Algorit
hm
s
.”
Inte
rnational
Jo
urnal
of
Re
searc
h
in
Comput
er
a
nd
Comm
unic
ation Tec
hnolog
y
,
v
ol
3,
no.
4,
Apr
i
l
2014
.
[5]
.
Dn
y
an
ada
R
Jad
hav,
L.
M.
R.
J
Lobo
,
Nav
iga
t
i
on
of
Pow
erPoi
nt
Us
ing
Hand
Gesture
s,
In
te
rn
at
ion
al
Journ
al
of
Scie
nc
e and
Res
ea
rch
(IJS
R)
201
5.
[6]
.
Ruchi
Man
ish
G
ura
v,
Prem
ana
nd
K.
Kadb
e,
Re
al
ti
m
e
fing
er
track
ing
and
con
tour
det
e
ct
ion
for
ges
ture
r
ec
ogn
it
ion
using Ope
nCV
,
I
EE
E
Confer
enc
e
Ma
y
2015
,
Pun
e
Indi
a.
[7]
.
Pei
Xu,
Depa
rt
m
ent
of
Elec
tr
ical
and
Com
pute
r
Engi
ne
eri
ng
,
U
nive
rsit
y
of
Min
nesota
,
A
Real
-
t
ime
Hand
Gestu
re
Rec
ognition
and
Hum
an
-
Com
pute
r
Int
eract
ion
S
y
stem
,
Rese
arc
h
Paper
April
201
7.
[8
]
.
P.
Sugan
y
a
,
R.
S
at
h
y
a
,
K.
Vijay
a
l
akshm
i
.
“
Detect
i
on
and
R
ec
ogni
tion
of
Gesture
s
to
Contro
l
the
S
y
s
t
em
Applicat
ions
b
y
Neur
al
Ne
tworks
.”
Int
ernati
on
al
Journal
of
Pure
and
Applied
M
athe
mati
cs
,
vol
.
118
,
no.
10
,
pp.
399
-
405
,
Janu
a
r
y
2018
.
Evaluation Warning : The document was created with Spire.PDF for Python.