BIOMEDICAL ENGINEERING VOLUME: 15 | NUMBER: 3 | 2017 | SEPTEMBER

An Investigation Into Time Domain Features of

Surface Electromyography to Estimate
the Elbow Joint Angle

TRIWIYANTO 1,3 , Oyas WAHYUNGGORO 1 , Hanung Adi NUGROHO 1 , HERIANTO 2

1
Department of Electrical Engineering & Information Technology, Faculty of Engineering,
Universitas Gadjah Mada, Jl. Grafika 2, 55281 Yogyakarta, Indonesia
2
Department of Mechanical & Industrial Engineering, Faculty of Engineering,
Universitas Gadjah Mada, Jl. Grafika 2, 55281 Yogyakarta, Indonesia
3
Department of Electromedical Engineering, Politeknik Kesehatan Surabaya,
Ministry of Health Indonesia Pucang Jajar Timur 10, 60282 Surabaya, Indonesia

triwiyanto123@gmail.com, oyas@ugm.ac.id, adinugroho@ugm.ac.id, herianto@ugm.ac.id

DOI: 10.15598/aeee.v15i3.2177

Abstract. In literature, it is well established that fea- 1. Introduction

ture extraction and pattern classification algorithms
play essential roles in accurate estimation of the el-
bow joint angle. The problem with these algorithms, Surface ElectroMyoGraphy (EMG) is often used to
however, is that they require a learning stage to recog- control an assist device such as the upper and lower
nize the pattern as well as capture the variability as- limb exoskeletons with the function to support human
sociated with every subject when estimating the elbow life [1]. It is obvious that the EMG signal can be related
joint angle. As EMG signals can be used to represent to the human limb motion. Several efforts on EMG
motion, we developed a non-pattern recognition method signal detection have been made to investigate the re-
to estimate the elbow joint angle based on twelve time- lationships between muscle groups and limb movement
domain features extracted from EMG signals recorded [2] and [3]. In the EMG detection stage, Tang et al.
from bicep muscles alone. The extracted features were [4] collected EMG signal from four muscle groups lo-
smoothed using a second order Butterworth low pass cated at biceps brachii, brachioradialis, triceps, and
filter to produce the estimation. The accuracy of the anconeus to estimate the elbow joint angle. Benitez
estimated angles was evaluated by using the Pearson’s et al. [5] recorded the EMG signals from two mus-
Correlation Coefficient (PCC) and Root Mean Square cle groups located at biceps and triceps to develop an
Error (RMSE).The regression parameters (Euclidean orthotic system. The methods that utilize more mus-
distance, R2 and slope) were then calculated to observe cle groups in estimating the elbow joint angle, however,
the effect of the features on elbow joint angle estima- would require more computational complexities in data
tion. In this investigation, we found that for a 10- processing.
second long recording period, the MyoPulse Percentage In order to get information related to elbow joint mo-
(MYOP) Rate produced the best accuracy: with PCC tion, the recorded EMG signal should be processed by
of 0.97 ± 0.02 (Mean±SD) and RMSE of 11.37 ± 3.04◦ using time, frequency, or time-frequency domain meth-
(Mean±SD), respectively. The MYOP feature also ods to produce informative features. After the feature
showed the highest R2 and slope value of 0.986±0.0083 extraction stage, the EMG features can represent use-
(Mean’s) and 0.746 ± 0.17 (Mean’s), respectively for ful information related to the joint angle, force, and
flexion and extension motions during all recorded peri- torque. Choosing an appropriate feature is essential
ods. because it determines the accuracy of the estimation.
Some previous studies have preferred to use time do-
main features over those extracted from frequency and
Keywords time-frequency domains to predict joint angle [6] and
[7] and torque [8]. This preference is due to reduced
EMG, feature extraction, non-pattern recogni- complexity in data processing and the application of
tion, time domain features. a simple algorithm to be implemented in the real

c 2017 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 448
BIOMEDICAL ENGINEERING VOLUME: 15 | NUMBER: 3 | 2017 | SEPTEMBER

time control. Generally, after the feature extraction (RMS), Integrated EMG (IEMG), Variance (VAR),
process, the joint angle or torque is estimated using and Mean Absolute Value (MAV). The complexity
a machine learning algorithm or a classifier to improve of the EMG signal could be quantified by using the
the accuracy. The methods used in human-machine Waveform Length (WL), Average Amplitude Change
interaction based on EMG control, are divided into (AAC), and Difference Absolute Standard Deviation
two categories: pattern recognition and non-pattern Value (DASDV) features. The calculated frequency-
recognition [1] methods. In the pattern recognition based informative features were as follows: Zero Cross-
methods, some previous studies used artificial neural ing (ZC), Sign Slope Change (SSC), Wilson Amplitude
networks [4], fuzzy controllers [1], and support vector (WAMP), and MYOPulse Percentage (MYOP) Rate.
machines [10] as their classifiers. The limitation in the
pattern recognition methods, however, is that the sys-
tem needs to be trained for each different subject due 1) RMS
to the variability in the EMG signal. Therefore, in
some cases, this method is not practically applicable. The Root Mean Square (RMS) value represents the
In the non-pattern recognition methods, some previ- mean power of a signal over a window length of EMG
ous studies used onset analysis, proportional control, samples. The mathematical equation to describe this
and threshold control [11] and [12]. These methods are feature is as follows [14]:
simple to be implemented but their accuracies tend to
v
be low. There is also limited literature on elbow-joint u N
u1 X
angle estimation using non-pattern recognition meth- RMS = t x2 , (1)
ods. N i=1 i

Although some efforts have been dedicated to pat-

tern recognition and non-pattern recognition methods where xi indicates the ith EMG signal and N indicates
for elbow-joint angle estimation, there are still some the length of the EMG signal.
limitations that should be addressed in furthering this
research. Therefore, the purpose of this study is to
develop a non-pattern recognition method for estimat- 2) IEMG
ing the elbow joint angle using a single muscle group
(biceps). To implement the proposed method, twelve The Integrated EMG (IEMG) value is an absolute
time-domain features were investigated and a second- summation of the EMG signal over a window length
order Butterworth low pass filter was applied to filter of EMG samples. The mathematical equation is de-
the features. The specific objectives of the study are scribed as follows [14]:
to:
N
X
IEMG = |xi | . (2)
• evaluate the accuracy of EMG features in esti- i=1
mating the elbow joint angle using the Pearson’s
Correlation Coefficient (PCC) and the Root Mean
Square Error (RMSE), 3) VAR
• evaluate the regression parameters (Euclidean dis-
The Variance of the EMG signal, EMG (VAR), is the
tance, R-squared, and slope) that relate to the el-
average value of the power of the EMG signal. VAR is
bow joint angle.
formulated as follows [14]:

N
1 X  2 
2. Theoretical Background VAR = x .
N − 1 i=1 i
(3)

2.1. Time Domain Features

4) MAV
The recorded EMG signal was extracted to get the
features that related to the human elbow-joint angle The Mean Absolute Value (MAV) is the average of
during flexion and extension motions. In this study, the absolute value of the EMG signal for a window
twelve Time-Domain (TD) features were extracted to length N . The MAV is formulated as [14]:
estimate the elbow joint angle. These features were
classified into three categories (based on energy, com- 1 X
N
plexities, and frequency information) [13]. The energy- MAV = |xi | . (4)
N i=1
based features were as follows: the Root Mean Square

c 2017 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 449
BIOMEDICAL ENGINEERING VOLUME: 15 | NUMBER: 3 | 2017 | SEPTEMBER

5) LOG 10) SSC

The Logarithm (LOG) parameter is a measure of the The Sign Slope Change (SSC) is the number of times
non-linear characteristic of the EMG signal. The LOG the slope of the signal changes its sign within a window
value is calculated based on the average of the loga- of length N . It is formulated as follows [14]:
rithm of the EMG signal. The LOG value is defined as
N −1
follows [14]: X
SSC = [f [(xi − xi+1 ) · (xi − xi+1 )]] ,
N
! i=1
 (10)
1 X 1, if → x ≥ threshold,
LOG = exp log(|xi |) . (5) f (x) =
N 0, otherwise.
i=1

11) WAMP
6) WL
The Wilson Amplitude (WAMP) is the number of
The Waveform Length (WL) is used to measure the times that the absolute value of the difference be-
length of the signal between two consecutive samples tween two consecutive samples (x
i+1 and xi ) exceeds
xi+1 and xi . WL is formulated as follows [14]: a threshold value. It is defined as follows [14]:
N
X −1 N −1
X
WL = |xi+1 − xi | . (6) WAMP = [f (|xi − xi+1 |)] ,
i=1  i=1 (11)
1, if → x ≥ threshold,
f (x) =
0, otherwise.
7) AAC

The Average Amplitude Change (AAC) is an the mean 12) MYOP

value of the waveform length within a window of length
N . AAC is written as follows [14]: The MyoPulse Percentage (MYOP) Rate is the average
of the number of times that the EMG signal exceeds
N −1 a predefined threshold. MYOP can be expressed as
1 X
ACC = |xi+1 − xi | . (7) [15]:
N i=1
N
1 X
MYOP = [f (xi )] ,
N i=1 (12)
8) DASDV 
1, if → x ≥ threshold,
f (x) =
0, otherwise.
The Difference Absolute Standard Deviation Value
(DASDV) is calculated based on the standard devia-
tion between xi+1 and xi . DASDV is defined as follows 2.2. Infinite Impulse Response
[14]:
v It is obvious that the EMG signal has random and
u N −1 stochastic characteristics in nature [16]. Therefore, in
u 1 X 2
DASDV = t (xi+1 − xi ) . (8) order to smooth and reduce the noise contaminating
N − 1 i=1
this signal, filtering is required. Commonly, the filter-
ing stage, as it has been performed in previous studies
[12] and [17], is conducted by applying a digital Low-
9) ZC Pass Filtered (LPF) to process the EMG signal. In this
study, an Infinite Impulse Response (IIR) LPF was de-
The Zero Crossing (ZC) value is the number of time signed and implemented. The LPF was constructed
that the signal crosses a certain threshold value. ZC is using a 2nd order Butterworth filter with cutoff fre-
calculated as [14]: quencies set between 80 Hz and 100 Hz, respectively.
The IIR filter was implemented using a cascade bi quad
N
X −1 filter. This digital filter was then implemented by using
ZC = [f (xi · xi+1 ) ∩ |xi − xi+1 |] ≥ threshold, the following difference equation [18]:
i=1
 (9)
1, if → x ≥ threshold, y[n] = b0 x[n] + b1 x[n − 1] + · · · + bP x[n − P ]−
f (x) = (13)
0, otherwise. +a1 y[n − 1] − a2 y[n − 2] − · · · − aQ y[n − Q],

c 2017 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 450
 value. ZC is LPF was constructed using a 2nd order Butterworth filter
with cutoff frequencies set between 80 Hz and 100 Hz,
respectively. The IIR filter was implemented using a
 threshold cascade bi quad filter. This digital filter was then
implemented
BIOMEDICAL by using the following difference equation
ENGINEERING VOLUME: 15 | NUMBER: 3 | 2017 | SEPTEMBER
[18]:

]  b0indicates
y[nx[n]
where x[n]  b1x[the
n nth
1]  ..input
 bP x[sample,
n  P] y[n] indi- 3.3. Data Collection
 athe
(9) cates n  1output
1 y[nth ]  a2 y[nsample,
 2]  ..b0 ,abQ1y, [bnP, aQ1], (13)
a2 , and aQ
are the filter coefficients, and P = Q is the filter order. Before the data collection process, the participants
ber of times the th
Where x[n] indicates the n input sample, y[n] indicates were instructed to follow some specific steps. EMG
in a window of the nth output sample,b0, b1, bP, a1, a2, and aQ are the filter signals were recorded while the subject’s arm held the
coefficients, and P=Q is the filter order. exoskeleton and moved it in flexion and extension mo-
3. Materials and Method tions within the range of 0 to 140◦ . As mentioned
i 1 )]] above, the exoskeleton was loaded with a 1 kg load
(10) 3.
3.1. Materials
Participants and Method (see Fig. 1). The motion periods were guided using
a metronome program so that the flexion and extension
3.1. Participants
To implement the proposed method, four healthy male movements could be regulated for 2 seconds, 4 seconds,
participants with no history of muscular disorder (age: 8 seconds and 10 seconds periods. EMG signals were
22.4To± implement
3.2 years the
old,proposed 65.45 ±
weight:method, four healthy
5.67 male re- recorded using a sampling frequency of 1000 Hz. For
kg) were
number of times participants
cruited for with
this no
study history
after of muscular
giving disorder
informed (age:
consent. each period of motion, the participants performed flex-
e between two Before 22.4±3.2
the years
dataold, weight: 65.45±5.67
collection process,kg)thewereparticipants
recruited ion and extension motions for eight cycles (designated
threshold value. werefor this study after giving informed consent. Before the
recommended not to do any hard work especially by C1, C2, C3, C4, C5, C6, C7, and C8) so that the to-
data collection process, the participants were
anything that could
recommended potentially
not to do harm
any hard work the elbow
especially anythingjoint. tal dataset comprised of 128 data points (4 participants
Thethat could potentially harm the elbow joint. The the × 4 periods × 8 cycles).
participants were instructed on how to perform
participants
flexion were instructed
and extension on how to and
movements perform the flexion
were informed
and any
about extension movements
potential and were
risk that couldinformed about in
be involved anycar-
potential
(11) rying risk that
out these could be involved in carrying out these
motions. 3.4. Data Processing
motions.
Figure 2 shows the processing of EMG signals to es-
timate the elbow joint angle. The collected EMG
signals from biceps were processed to extract twelve
Time-Domain (TD) features with a length of window
s the average of of 200 milliseconds. The feature extraction process was
gnal exceeds a
ssed as[15]: conducted for each cycle of motion with the total of
eight cycles. All of the extracted TD features such as:
EMGF (RMS, IEMG, VAR, MAV, LOG, WL, AAC,
DASDV, ZC, SSC, WAMP, and MYOP) were calcu-
old (12) lated for each cycle and motion period. In order to
obtain the estimated angle, the second order Butter-
worth low pass filter was then applied to smooth the
features. As mentioned before, this IIR low pass fil-
Fig.Fig.
1: 1:The
TheExoskeleton
Exoskeleton frame to synchronize
frame the elbow-joint
to synchronize motion
the elbow-joint
ter was designed using the cut-off frequencies specified
motion. above to smooth out the EMG signals. The filtered fea-
ture Melbas then assumed as the estimated elbow joint
angle. To evaluate the performance of the proposed
method, the estimated elbow joint angle was analyzed
3.2.
CTRONIC ENGINEERING Equipment 3 using the Pearson’s Correlation Coefficient (PCC) and
Root Mean Squared Error (RMSE). The PCC was used
A one-channel EMG system comprised of: a pre- to evaluate the relationship between the extracted TD
amplifier, a band pass filter (with cut-off frequencies features and the elbow joint angle. The RMSE value
of 20 to 500 Hz, respectively), a summing amplifier, was used to evaluate the deviation between the esti-
and an adjustable gain amplifier, was built. EMG sig- mated angle and the measured angle. The linearity
nals were collected using three disposable surface (pre- of the estimated angle was also evaluated using linear
2
gelled Ag/AgCl) bioelectrodes. Two bioelectrodes were regression parameters namely R , Slope and the Eu-
positioned on the biceps muscle with the third one clidian Distance.
placed on the hand as a common ground electrode.
The participants held an exoskeleton frame which was
used to synchronize the elbow joint motion (see Fig. 1). 3.5. Statistical Analysis
The elbow joint angle of the exoskeleton was collected
using a linear potentiometer which was located at the The statistical Analysis of Variance (ANOVA) was per-
joint between the arm and forearm of the exoskeleton. formed to observe if there was any statistical difference
A one kilogram (1 kg) load was placed on the forearm in performance and the regression parameters between
of the exoskeleton. the periods of motion (10 seconds, 8 seconds, 4 seconds,

c 2017 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 451
BIOMEDICAL ENGINEERING VOLUME: 15 | NUMBER: 3 | 2017 | SEPTEMBER

Time Domain Feature

EMG RMS, IEMG, VAR, MAV, LOG, WL,
AAC, DASDV, ZC, SSC, WAMP, MYOP
EMGF
Measured angle
Digital Filtering
2nd order Butterworth LPF

EMGL
Estimated angle
Evaluation
Pearson’s correlation coefficient, RMSE,
R2, slope and Euclidean distance
Fig. 2: The processing of EMG signas for flexion and extension movements to estimate the elbow joint angle. EMG signals were
collected from biceps; time domain features were extracted and smoothed using a second order Butterworth low pass filter.

and 2 seconds). The significance test was established (Mean±SD) value. In the motion period of 8 seconds,
with confidence level of 95 % (alpha = 0.05). as shown in Fig. 3(c) and Fig. 3(d), the estimated
angle based on the MYOP feature shows the high-
est correlation coefficient (0.97 ± 0.01) and the lowest
4. Results and Discussion RMSE (11.25 ± 2.44◦ ) value. Figure 3(e) and Fig. 3(f)
show that the estimated angle from the MYOP fea-
ture has the highest accuracy (correlation coefficient
The recorded EMG signals and the measured angles ac-
= 0.91 ± 0.04 and RMSE = 17.58 ± 3.08◦ ). The highest
quired from four participants were processed offline for
accuracies of the estimated angle are also found from
feature extraction and evaluation. A predefined thresh-
the estimated angle based on the MYOP feature in the
old was required for ZC, SSC, WAMP, and MYOP fea-
motion period of 2 seconds (0.88±0.05 and 20.13±2.69◦
tures. The cut-off frequency of the LPF was also essen-
for correlation coefficient and RMSE, respectively).
tial which determined the smoothness of the estimated
Over all periods of motion, there is a minimum of
angle. In this work, the threshold and cut-off frequen-
RMSE of 6.07◦ and a maximum correlation of 0.99 that
cies were chosen such that elbow joint angle estimation
occurred in the 10 second period of motion. Among
could be made at the maximum performance. The de-
the other features, the correlation coefficient of the es-
tailed results of this study are explained and discussed
timated angle from Zero Crossing (ZC) feature showed
in the following subsection.
the widest variance (Fig. 3(a), Fig. 3(b), Fig. 3(c),
Fig. 3(d), Fig. 3(e) and Fig. 3(f)). The estimated an-
gle based on the VAR feature showed wider variance of
4.1. Accuracy of the Elbow Joint RMSE compared to the other features. The ANOVA
Angle Estimation tests showed that there was a significant difference (p-
value < 0.05) in accuracy between groups of periods
In this work, a relationship between the estimated an- (10 seconds, 8 seconds, 4 seconds and 2 seconds) for all
gle and the measured angle was indicated by the PCC. features except for the MYOP feature. In the period of
A coefficient score approaching 1 indicates that there is motion of 8 seconds and 10 seconds, the MYOP feature
a strong relationship and a score approaching 0 shows showed that there was no significant difference in the
that there is a weak relationship. In the motion period RMSE value (p-value > 0.05). This indicates that the
of 10 seconds Fig. 3(a) and Fig. 3(b), the results show estimated angle using the MYOP feature is more con-
that the estimated angles based on the MYOP feature sistent and produces higher accuracy to estimate the
have the highest correlation coefficient (0.97 ± 0.02) elbow joint angle for different motion periods compared
(Mean±SD) and the lowest RMSE (11.37 ± 3.04◦ ) to the other features.

c 2017 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 452
SECTION POLICIES
SECTION POLICIES VOLUME: XX |XX
VOLUME: NUMBER: X | 2015
| NUMBER: | MONTH
X | 2015 | MONTH
SECTION POLICIES
SECTION POLICIES VOLUME: XXXX
VOLUME: | NUMBER: X |X2015
| NUMBER: | MONTH
| 2015 | MONTH
BIOMEDICAL ENGINEERING VOLUME: 15 | NUMBER: 3 | 2017 | SEPTEMBER
SECTION POLICIES VOLUME: XX | NUMBER: X | 2015 | MONTH
SECTION
SECTION POLICIES
POLICIES VOLUME:
VOLUME: XX | XX | NUMBER:
NUMBER: X | 2015
X | 2015 | MONTH
| MONTH
SECTION POLICIES VOLUME: XX | NUMBER: X | 2015 | MONTH
1.001.00 40 40
0.950.95 35
1.001.00 40 40

(degrees)
0.900.90 30
0.950.95 35 35
0.851.00 25 40
Correlation
Correlation

0.85

(degrees)
0.901.00
1.00 30
40 35 40

(degrees)
0.95
0.800.90
0.80 20 30
Correlation
Correlation

0.850.95 35
25

(degrees)
0.95
0.85
0.90
1.00 15 4025 3035
0.750.75

RMSE
(degrees)
(degrees)
30
Correlation

0.800.90
0.90
0.80
0.85 20 35 30
20
25
0.700.95
0.70 10
Correlation
Correlation

0.750.85
0.85 25
15 25

(degrees)
0.75
0.80 5 15 20

RMSE
0.90 30

RMSE
0.650.65
0.80 20
Correlation

0.70 0.80 10 20
0.70
0.85
0.75 0 10
25 15

RMSE
0.600.60 15
0.650.75
0.75 520 10 515

RMSE
0.80
0.65
0.70

RMSE

RMS RMSRMS

ZC ZC ZC

MYOP
MAVMAV
LOGLOG

WL WL WL
IEMG

VAR VARVAR

AACAAC

ZCZCSSC SSC

WAMP
DASDV
ZC ZC
LOG

MYOP
RMSRMS

WL
IEMG
VAR
MAVMAV

AAC

ZCZCSSC SSC
WAMP
DASDV

ZC
LOG WLLOG

MYOP
RMS

WL
IEMG
VAR
LOGMAV

AAC

SSC
WAMP
DASDVDASDV
0.600.70
0.70
0.75
0.60 10015 010 5
(a) 0.65 0.65

RMSE
5 05

MYOP
IEMG

WAMP
DASDV
MYOP
IEMG
VAR

AAC

WAMP
DASDV
0.65

RMS

ZC

MYOP
MAV
LOG
WL
IEMG
VAR

AAC

SSC
WAMP
DASDV
10

ZC
LOG

MYOP
RMS

WL
0.70
IEMG
VAR
MAV

AAC

SSC
WAMP
DASDV
0.60 (b)(b)0 0
(a) (a)0.60 0.60
0.65 5

RMS

MYOP
MAV
LOG
WL
IEMG
VAR

AAC

SSC
WAMP
DASDV
LOG

MYOP
RMS

WL
IEMG
VAR
MAV

AAC

SSC
WAMP
(b)0 40

MYOP
MAV
LOG
IEMG

AAC

SSCSSC
WAMP
DASDV
ZC ZC
LOG

MYOP
RMS

WLWL
IEMG
VAR
MAV

AAC

SSCSSC
WAMP
1.000.60 (a) 10 s. DASDV (b) 10 s.
(a) 1.00 (b)40

RMS

MYOP
MAV
LOG
WL
IEMG
VAR

AAC

SSC
WAMP
DASDV
MYOP
RMS

WL
IEMG
VAR
MAV

AAC

SSC
WAMP
(a)(a) 0.95 DASDV (b)

RMS

ZC

MYOP
MAV
LOG
WL
IEMG
VAR

AAC

WAMP
DASDV
0.95 35 35
LOG

MYOP
RMS
IEMG
VAR
MAV

AAC

WAMP
DASDV

(a) 1.00 (b)40 40

0.900.90 30 30

(degrees)
RMSE (degrees)
0.95
1.00 35 40
Correlation
Correlation

0.85
1.000.85
1.00 25 40 25 40
30 35

(degrees)
0.90
0.95
0.80
0.950.80
1.00 20 35 20
2540 30
Correlation

0.95
0.85
0.90 35

(degrees)
0.75
0.900.75 15 30 15

(degrees)
RMSE
0.95
2035 25
Correlation

0.90 30

RMSE (degrees)
0.80
0.85
Correlation
Correlation

0.85
0.70 0.70
0.90 10 25
30 10

(degrees)
0.85
0.75
0.80 15 20 25
RMSE
Correlation

0.80
0.65 0.65
0.85 20
5 20
25 5
0.80
0.70
0.75 10 15
RMSE
0.75
0.60 0.60
0.80 15
0 15
20 0
RMSE
0.75
0.65
0.70 5 10
RMS

ZC

MYOP MYOP
LOG
WL
RMS

ZC

MYOP MYOP
LOG
WL

SSC
IEMG IEMG
VAR

AAC

SSC

WAMP WAMP
RMS IEMG

MAV
VAR

AAC

DASDV
VAR MAV

SSC WAMP
DASDV

RMS

ZC

MYOP MYOP
WL WLLOG
WL
RMSRMSRMS

ZC ZC

MYOP
IEMG

LOG
WL

SSC
VAR

AAC
IEMG

ZCSSC SSC

WAMP
LOGMAV
VAR

AACAAC

ZC ZC DASDV

WAMP
MAVMAV
0.70

DASDV
0.75 10
15 10
RMSE

(c) (c)0.650.70
0.60
0.65 (d) (d) 0 5
0.70 105
0.65 5

MYOP
IEMG

VARVAR

WAMP
DASDV
RMS

ZC

MYOP
LOG
WL

SSC
IEMG
VAR

AAC
MAV

WAMP
DASDV

0.60
0.60
0.65 05 0
0.60 0
ZCZC
LOG
WL
AAC
MAV

DASDV

RMS

LOG
WL
IEMG

SSC
VAR

AAC

WAMP
MAV

DASDV
RMS

ZC ZC
LOG

WLWL

SSCSSC
IEMG
VAR

AAC
MAV

WAMP
DASDV

MYOP
LOG
IEMG

SSC
AAC

WAMP
MAV

DASDV
(c)0.60 0

W
(d)
MYOP
RMS

LOG
WL

SSC
IEMG
VAR

AAC
MAV

WAMP
DASDV

RMS

ZC

MYOP
LOG
WL
IEMG

SSC
VAR

AAC

WAMP
MAV

DASDV
(c) (d)40 40

D
1.00
(c) 1.00
MYOP
RMS

LOG
IEMG
VAR

AAC
MAV

WAMP
DASDV

MYOP
RMS

LOG
WL
IEMG

SSC
VAR

AAC

WAMP
MAV

DASDV
(d)
(c)
0.950.95 (c) 8 s. (d)35 35 (d) 8 s.
RMSE (degrees)
(degrees)

1.00 40 40
0.901.00
0.90 30 30
Correalation
Correalation

0.95 0.95
1.00
1.000.85
354035
25 40
25
(degrees)

0.85
(degrees)

0.90 0.90
1.00
0.95 30 3040
Correalation

35
Correalation

0.95
0.80 0.80
1.00 40
20 20
(degrees)

0.85 0.85
0.95 25 2535
RMSE

0.90
0.900.75 3035 15
RMSE (degrees)

0.75 15
Correalation

0.95
Correalation

(degrees)

0.80 0.80
0.90
0.85 20 2030
Correalation

0.85
0.70 0.70
0.90 25
10 30 10
Correalation

RMSE
RMSE

0.75 0.75
0.85
0.80
0.80 152025 1525
0.65
0.650.85 5 5
0.70 0.70
0.80 1020
RMSE

0.60 0.75
0.750.60
0.80 101520
0 0
0.65 0.65
0.75 515 515
RMSE

0.70
0.70 10
LOG
RMS

ZC

MYOP
WL
IEMG
VAR
MAV

AAC

SSC
WAMP
DASDV DASDV
LOG
RMSRMS

ZC ZC

MYOP
WL
IEMG
VAR
VAR MAVMAV

WLAAC

ZC SSC SSC
SSC WAMP
DASDV

0.75
ZC ZC

MYOP
LOGLOG

WL WL
IEMG

SSC SSC
VARVAR

AACAAC

WAMP
MAVMAV

ZC ZCDASDV
ZC ZC

MYOP
LOGLOG

WLWLWL WL
IEMG

SSCSSC SSC
VAR VAR

AACAAC

WAMP
MAVMAV

DASDV

(e) 0.60 0.60

0.70
(e)0.65
0.65 0105 010
0.70 (f) (f)
LOG
RMS

ZC

MYOP
IEMG
VAR
MAV

AAC

SSC
WAMP
DASDV
LOG

MYOP
WL
IEMG
VAR

AAC

WAMP
DASDV

0.65

MYOP
IEMG

WAMP
DASDV

05 5

MYOP
IEMG
IEMG VAR

WAMP
DASDV

0.60
0.60
0.65
(e) (e) 0.60
0.60 (f) (f)0 0
LOG
RMS

ZC ZC

MYOP
WLWL
IEMG
VAR
MAV

AAC

SSCSSC
WAMP
DASDV
LOG
RMS

ZC

MYOP
WL
IEMG
VAR
MAV

AAC

SSC
WAMP

MYOP
LOG
IEMG

AAC

WAMP
MAV

DASDV

(e)
LOG

MYOP
RMS

WL
IEMG

MAV

AAC

WAMP
DASDV

ZC

MYOP
LOG

WL

SSC
VAR

AAC

WAMP
MAV

DASDV
LOG

MYOP
RMS
IEMG
VAR
MAV

AAC

WAMP
DASDV

MYOP
LOG
IEMG

SSC
VAR

AAC

WAMP
MAV

DASDV

(e) 1.00
1.00 40 40
(e) (f) (f)
35 35
0.950.95 40 40
RMSE (degrees)
(degrees)

1.001.00 (e) 4 s. 30 30 (f) 4 s.

0.900.90 4035
0.95 0.95 35 40
CorelationCorelation

1.00
Corelation

0.851.00 25 25
(degrees)

0.85
(degrees)

0.90 0.90
1.00
0.95
1.00 4030
35
30 40
35
0.800.95
0.80 20 20
(degrees)
Corelation

RMSE (degrees)
Corelation

0.85
0.95
0.90
0.85 0.90 30
35 25
25 30
RMSE

0.750.75 15 2015
(degrees)
Corelation
Corelation

0.80 0.80
0.90
0.85 30 25
25
20
0.700.85
0.70 10 10
RMSE
RMSE

0.75
0.85
0.80
0.75 2515
20
15
0.650.80
0.65 5
205
RMSE

0.70 0.70
0.75
0.80
0.75 2010
15
10 15
0.600.60 0 50
RMSE

0.65 0.65
0.70
0.75
0.70 15
10
5 10
ZC

MYOP
RMS

LOG WLLOG
WL
VAR

AAC

SSC
VARIEMG

LOGMAV

WAMP
DASDV DASDV
ZC ZC

MYOP
RMSRMS

LOG
WL
VAR

AAC

ZCSSC
IEMG

MAVMAV

WAMP
DASDV

MYOP MYOP
RMSRMS

ZC ZC
LOGLOG
WL WL

WAMP
VARIEMG

SSC SSC
VARVAR

AACAAC
MAVMAV

ZCZCDASDV

MYOP
RMSRMS

ZC ZC
LOGLOG
WL WL

WAMP
IEMG

SSC SSC
VAR

AACAAC
MAVMAV

DASDV

0.60 0.60
0.65
0.70 10
05 05
(g) (g) 0.65 (h) (h)
MYOP
RMS RMS

LOG
WL
VAR

AAC

SSC
IEMG

MAV

WAMP
DASDV

0.60
0.65
MYOP
AAC

ZCSSC
IEMG

WAMP
DASDV

MYOP
WAMP
IEMG

DASDV

05 0
MYOP
WAMP
IEMG

0.60
DASDV

(g) (g) 0.60

ZC

MYOP
RMS

LOG
WL
VAR

AAC

ZCSSC
IEMG

MAV

WAMP
DASDV

(h)(h)0
MYOP
RMS

LOG
WL

WAMP
IEMG

SSC
VAR

AAC
MAV

DASDV
MYOP
RMS

LOG
WL
VAR

AAC

SSC
IEMG

MAV

WAMP

RMS

ZC
LOG
WL

WAMP
IEMG

SSC
VAR

AAC
MAV

DASDV

(g)
MYOP
WL
VAR

AAC

SSC
IEMG

MAV

WAMP
DASDV

(g) (h)(h)
MYOP
RMS

LOG
WL

WAMP
IEMG

SSC
VAR

AAC
MAV

DASDV

Fig. Fig.
3: 3:Effect Effect of features
of features EMGEMG to theto accuracy
the accuracy of elbow-joint
of the the elbow-jointangleangle estimation.
estimation. The The boxplot
boxplot of Pearson’s
of Pearson’s correlation
correlation coefficient
coefficient for period
for period of of
motion
motion (a) 10s,
(a) 10s, (c)(e)
(c) 8s, 8s,4s(e)and
4s (g)
and (g)
2s. 2s. The
The boxplot
boxplot of RMSE
of RMSE for period
for period of motion
of motion (b) 10s,
(b) 10s, (d)(f)
(d) 8s, 8s,4s(f)2and
4s and2s.
(h) (h) 2s.
Fig.Fig.
3: 3: Effect Effect of features
of features EMG EMG
(g) 2 to the
to the s. accuracy
accuracy of the
of the elbow-joint
elbow-joint angle
angle estimation.
estimation. TheThe boxplot
boxplot (h) of Pearson’s
s.
of Pearson’s correlation
correlation coefficient
coefficient for for period
period of of
Fig.
motion 3:
motion
(a) (a)Effect
10s, 10s, of (e)
(c)
(c) 8s, features
8s, 4s 4sEMG
(e)and and
(g) to The
(g)
2s. the accuracy
2s. The of RMSE
boxplot
boxplot of the elbow-joint
of RMSE angle
for period
for period estimation.
of
ofanglemotion
motion (b) (b)The
10s, 10s,
(d)boxplot
(d)
8s, (f)of
8s, (f)Pearson’s
4s 4s and
and 2s.correlation
(h) (h) 2s. coefficient for period of
Fig.
motion 3:(a) Effect
10s, (c)ofof features
8s,features
(e) EMG
4s features
and (g)to to
2s.to the
The accuracy
boxplot of
ofthe the
RMSE elbow-joint
forelbow
period of estimation.
motion (b) 10s, The boxplot of Pearson’s correlation coefficient for period of
Fig.
Fig.3:3: Effect
Effect EMG the accuracy of elbow-joint angle estimation. The(d) 8s,boxplot
boxplot (f) 4sPearson’s
of and (h) 2s.correlation coefficient for period of of
Fig. 3:
motion The effect
(a) 10s, (c)of8s,features
TD(e) 4s EMG
and (g)on 2s.the
the
Theaccuracy
boxplot of
accuracy the
ofof elbow-joint
the
RMSE angle
joint
for period estimation.
ofangle
motion The
estimation.
(b) 10s, The
(d) 8s, (f)boxof
4s Pearson’s
plot
and correlation
(h)of2s.Pearson’s coefficient
Correlation for period
motion
motion (a)(a)10s,
10s,
Coefficient (c)(c)
8s,
were (e)(e)
8s, 4s4sandand
calculated(g)(g)
2s.forThe
2s. boxplot
The
the of RMSE
boxplot
following ofperiods
RMSE for for
period
of of motion
period
motion: of(a) (b)s,10s,
motion
10 (b) (d)
(c)10s, 8s,
8 s, (f)
8s,4s
(d)(e) 4(f)sand4s (h)
and and 2s.
(g) (h)
2 2s.
s. The box plot of the
RMSE value for periods of motion: ( b) 10 s, (d) 8 s, (f) 4 s and (h) 2 s.

c 2017 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 453

© 2015
© 2015 ADVANCES
ADVANCES IN ELECTRICAL
IN ELECTRICAL ANDAND ELECTRONIC
ELECTRONIC ENGINEERING
ENGINEERING 1 1
 SECTION POLICIES VOLU
VOLUME: XX | NUMBER: X | 2015 | MONTH
SECTION POLICIES
BIOMEDICAL ENGINEERING VOLUME: VOLUME: XX | NUMBER:
15 | NUMBER: X | 2015 | MONTH
3 | 2017 | SEPTEMBER

RMS
RMS IEMG
IEMG WL
WL AAC
AAC ZC SSC
VAR
VAR MAVMAV DASDV Mea.Angle WAMP MYOP
RMS IEMG DASDV
WL Mea.Angle
AAC ZC SSC
LOG
LOG Mea.Angle
Mea.Angle Mea.Angle
VAR MAV DASDV Mea.Angle WAMP MYOP
1.01.0 LOG Mea.Angle Mea.Angle 1.0
(a) Flexion (c) Flexion (e) F
Estimated Angle [NORM]
[NORM]

[NORM]
1.0 1.0
0.80.8 (a) Flexion (c) Flexion (e) Flexion 0.8
[NORM]

[NORM]
0.8 0.8
0.60.6 0.6
Angle

Angle
0.6 0.6
Angle

Angle
0.40.4 0.4
Estimated

Measured
0.4 0.4
Estimated

0.2

Measured
0.20.2
0.2 0.2
0.00.0 0.0
00 11 22 33 44 00 1 2 3 4
0.00 0 1 1 2 2 33 44 0.0
0 Time
1Time[seconds]
[seconds]
2 3 4 0 Time [seconds]
1 Time2[seconds] 3 4 0 2 [seconds]
1 Time 3 4
(a)Time [seconds]
Flexion. Time [seconds]
(b) Flexion. [seconds]
Time(c) Flexion.
RMS IEMG WLWL AAC
AAC ZC SSC
VAR MAV DASDV Mea.Angle WAMP MYOP
RMS
RMS IEMGIEMG WLWL AAC
AAC ZC
Mea.Angle SSC
LOG
VAR
VAR MAV Mea.Angle
MAV DASDV
DASDV Mea.Angle
Mea.Angle WAMP MYOP 1.0
1.0 LOG
LOG Mea.Angle
Mea.Angle Mea.Angle
(d) Extension (f) Extension 1.0
[NORM]

(b) Extension

[NORM]
1.01.0
0.8
Estimated Angle [NORM]

0.8
[NORM]

(b) Extension

[NORM]
0.80.8 0.8
0.6 0.6
Angle

Angle
0.60.6 0.6
Angle

Angle
0.4 0.4
Estimated

Measured
0.40.4 0.4
Estimated

Measured
0.2 0.2
0.20.2 0.2
0.0 0.0
0.00.0 4 6 8 10 4 6 8 10 4 6 8 100.0
4 4 6 Time
6 8 8
[seconds] 1010 44 6
6Time 88
[seconds] 10
10 4 6
Time 8
[seconds] 10
Time
Time[seconds]
[seconds] Time [seconds]
Time[seconds] Time [seconds]
Fig. 4: Sample of time respons of the normalized EMG features for period of motion of 10 seconds. EMG features of RMS, IEMG, VAR, MAV,
(d) Extension. (e) Extension. (f) Extension.
and 4:
Fig. LOG for (a) flexion
Sample of timeand (b) extension
respons motion. EMG
of the normalized EMG features
featuresoffor
WL, AAC,ofand
period DASDV
motion of 10for (c) flexion
seconds. EMG and (d) extension
features of RMS,motion.
IEMG,EMG
VAR,features
MAV,
of ZC, SSC, WAMP and MYOP for (e) flexion and (f) extension motion.
and LOG for (a) flexion and (b) extension motion. EMG features of WL, AAC, and DASDV for (c) flexion and (d) extension motion. EMG features
of ZC,
Fig. 4: SSC, WAMP
Typical timeand MYOP for
response for(e)
theflexion and (f) extension
normalized estimatedmotion.
angle for a motion period of 10 seconds. The estimated angle based
on RMS, IEMG, VAR, MAV, and LOG features for (a) flexion and (d) extension motions. The estimated angle from WL,
AAC, and DASDV features for (b) flexion and (e) extension motions. The estimated angle from ZC, SSC, WAMP and
MYOP features for (c) flexion and (f) extension motions.

The results of our proposed method are compara- 4.2. Response of Estimated Angle to
ble with those presented in several previous studies [3] Time
and [19]. Pau et al. developed a model to estimate
the elbow joint angle using the Hill-based method and Figure 4(a), Fig. 4(b), Fig. 4(c), Fig. 4(d), Fig. 4(e) and
a genetic algorithm in two muscle groups (biceps and Fig. 4(f) show a typical response of the estimated angle
triceps) [3]. In their study, they achieved an RMSE to time for a motion period of 10 seconds (red line indi-
value of 18.6 ± 6.5◦ for five continuous cycles. cate the measured angle). Ideally, the estimated angle
Tang et al. studied the elbow joint angle estimation should be comparable to the measured angle. To test
problem using artificial neural networks as a classifier. this proximity, the Euclidean Distance (ED) was cal-
In their research, they utilized four muscle groups (bi- culated to present the closeness between the pattern of
ceps brachii, brachioradialis, triceps brachii and an- the estimated angle and the measured angle as shown
coneus). The RMSE values of their model were 10.7◦ , in Eq. (14):
9.67◦ , 12.42◦ for motion period of 2 seconds, 4 seconds, v
uN
and 8 seconds, respectively [19]. uX 2
ED = t (EMGL − Anglei ) , (14)
i=1

c 2017 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 454

©©© 2015
2015
2015 ADVANCES
ADVANCES
ADVANCES INININELECTRICAL
ELECTRICAL
ELECTRICAL AND
AND
AND ELECTRONIC
ELECTRONIC
ELECTRONIC ENGINEERING
ENGINEERING
ENGINEERING 222
©2015
2015
©©2015 ADVANCES
ADVANCES
ADVANCES INININ ELECTRICAL
ELECTRICAL
ELECTRICAL AND
AND
AND ELECTRONIC
ELECTRONIC
ELECTRONIC ENGINEERING
ENGINEERING
ENGINEERING 222
BIOMEDICAL ENGINEERING VOLUME: 15 | NUMBER: 3 | 2017 | SEPTEMBER

Tab. 1: The Euclidean Distance between elbow joint angle and features for flexion and extension motions (period of motion:
10 seconds, 8 seconds, 4 seconds, and 2 seconds). The bold text indicates the lowest value of the Euclidean Distance.

Flexion motion (NORM) Extension motion (NORM)

Features
T = 10∼s T = 8∼s T = 4∼s T = 2∼s T = 10∼s T = 8∼s T = 6∼s T = 2∼s
RMS 0.671 0.921 0.615 0.729 0.741 0.606 0.52 0.547
IEMG 0.625 0.898 0.61 0.761 0.682 0.597 0.519 0.563
VAR 1.34 1.598 1.023 1.088 1.493 1.202 0.686 0.742
MAV 0.625 0.898 0.61 0.761 0.682 0.597 0.519 0.563
LOG 0.654 0.884 0.622 0.786 0.66 0.602 0.545 0.587
WL 0.76 1.091 0.645 0.715 0.952 0.699 0.548 0.595
AAC 0.76 1.091 0.645 0.715 0.952 0.699 0.548 0.595
DASDV 0.696 1.039 0.661 0.718 0.827 0.632 0.551 0.575
ZC 0.445 0.313 0.289 0.244 0.805 0.766 0.602 0.642
SSC 1.449 1.547 1.08 0.974 1.709 1.003 0.763 0.618
WAMP 0.312 0.325 0.242 0.341 0.452 0.445 0.549 0.565
MYOP 0.581 0.624 0.532 0.678 0.329 0.191 0.43 0.496

Tab. 2: The Linear Regression R2 values between the estimated and measured angles. The R2 values were calculated for all
periods of motion (10 second, 8 seconds, 4 seconds, and 2 seconds) for the flexion and extension movements.

Flexion Extension
Features
T = 10∼s T = 8∼s T = 4∼s T = 2∼s T = 10∼s T = 8∼s T = 6∼s T = 2∼s
RMS 0.952 0.949 0.965 0.969 0.974 0.995 0.993 0.981
IEMG 0.956 0.947 0.967 0.968 0.981 0.993 0.995 0.985
VAR 0.824 0.812 0.86 0.813 0.916 0.949 0.993 0.997
MAV 0.956 0.947 0.967 0.968 0.981 0.993 0.995 0.985
LOG 0.957 0.942 0.973 0.967 0.988 0.99 0.994 0.988
WL 0.962 0.939 0.971 0.978 0.977 0.993 0.995 0.988
AAC 0.962 0.939 0.971 0.978 0.977 0.993 0.995 0.988
DASDV 0.955 0.943 0.966 0.974 0.974 0.996 0.993 0.986
ZC 0.946 0.99 0.977 0.964 0.856 0.889 0.914 0.91
SSC 0.852 0.831 0.847 0.863 0.968 0.964 0.987 0.993
WAMP 0.995 0.998 0.995 0.993 0.976 0.996 0.942 0.93
8 MYOP 0.979 0.987 0.997 0.993 0.989 0.991 0.983 0.971

Tab. 3: The Linear Regression Slope values between the estimated and the measured angle. The slopes were calculated for all
periods of motion (10 seconds, 8 seconds, 4 seconds, and 2 seconds) for the flexion and extension movements.

Flexion Extension
Features
T = 10∼s T = 8∼s T = 4∼s T = 2∼s T = 10∼s T = 8∼s T = 6∼s T = 2∼s
RMS 0.752 0.693 0.608 0.466 −0.735 −0.619 −0.535 −0.434
IEMG 0.761 0.703 0.609 0.446 −0.752 −0.622 −0.534 −0.418
VAR 0.68 0.569 0.499 0.299 −0.806 −0.51 −0.533 −0.394
MAV 0.761 0.703 0.609 0.446 −0.752 −0.622 −0.534 −0.418
LOG 0.733 0.712 0.597 0.432 −0.757 −0.616 −0.507 −0.399
WL 0.707 0.639 0.593 0.484 −0.652 −0.585 −0.505 −0.394
AAC 0.707 0.639 0.593 0.484 −0.652 −0.585 −0.505 −0.394
DASDV 0.743 0.656 0.583 0.483 −0.701 −0.613 −0.502 −0.406
ZC 0.766 0.852 0.804 0.825 −0.454 −0.811 −0.548 −0.452
SSC 0.637 0.624 0.488 0.426 −0.726 −0.627 −0.514 −0.47
WAMP 0.822 0.889 0.831 0.726 −0.702 −0.88 −0.666 −0.511
MYOP 0.892 0.851 0.727 0.554 −0.917 −0.909 −0.621 −0.497

where N indicates the number of samples, Anglei based on the WAMP feature tended to show smaller
stands for the-ith measured angle and the EMGL shows ED values in the elbow flexion trajectory (for all peri-
the filtered features (estimated angle). In general, ods of motion) compared to those based on the other
a small value of ED indicates a close relationship be-
features. For the elbow extension trajectory, the esti-
tween the estimated angle and the measured angle. ED mated angles based on the MYOP feature showed the
was measured for all periods of motion (10 seconds, lowest value (0.329, 0.191, 0.430, and 0.496 for motion
8 seconds, 4 seconds and 2 seconds) and for all of the
period of 10 seconds, 8 seconds, 4 seconds and 2 sec-
TD features. onds, respectively). The estimated angle based on the
VAR feature tended to have higher Euclidean Distance
Table 1 shows the summary of the ED values for values compared to other features for all periods of mo-
all motion periods and features. The estimated angle tion.

c 2017 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 455
SECTION
SECTION
SECTION
SECTION POLICIES
POLICIES
POLICIES
SECTION
BIOMEDICAL POLICIES
ENGINEERING VOLUME:
VOLUME:
VOLUME:VOLUME:
VOLUME:
15 XX
XX
XX
XX |NUMBER:
|NUMBER:
3NUMBER:
|| NUMBER:
| NUMBER: | 2017 |XX|XX || 2015
|20152015
2015
SEPTEMBER || MONTH
MONTH
| MONTH
| MONTH

RMS IEMG
IEMG VAR
VAR ZC
ZCZC SSC
SSC
SSC
RMS
RMSRMS IEMG
IEMGIEMG VAR
VAR
VAR WL
WL
WL
WL
WL AAC
AAC
AAC
AAC DASDV
DASDV
DASDV
DASDV
DASDV ZC SSC
WAMP
WAMP
WAMP MYOP
MYOP
MYOP
MAV
MAV
MAVMAV
MAV LOG
LOG
LOG
LOG
LOG WAMP MYOP
1.0 1.0
1.0
1.0
1.0
[NOMR]
Estimated Angle [NOMR]
Angle[NOMR]
Estimated Angle [NOMR]

0.8 0.8
0.8
0.8
0.6 0.6
0.6
0.6
EstimatedAngle

0.4 0.4
0.4
0.4
Estimated

0.2 0.2
0.2
0.2
0.0 0.0
0.0
0.0
0.0 0.0 0.2 0.2 0.40.4 0.60.6 0.80.8 1.01.0 0.0
0.0 0.2
0.0 0.2 0.4
0.2 0.4 0.60.6 0.80.8 1.01.0 0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.2
0.0 0.2Measured
0.4 Angle
0.4 0.6 [NORM]
0.6 0.8 1.01.0 0.00.0 0.2
0.8 0.2 0.40.4 0.60.6 0.80.8 1.01.0 0.00.0 Measured
[NORM] 0.20.2 0.40.4 0.6 0.80.8 1.01.0
Angle0.6
[NORM]
Measured Angle [NORM] Measured
Measured
Measured Angle
Angle
Angle [NORM]
Measured
Measured Angle
Angle [NORM]
[NORM] Measured
Measured Angle
Angle [NORM]
[NORM] Measured
Measured Angle
Angle [NORM]
[NORM]
(a) Flexion. (b) Flexion. (c) Flexion.
RMS IEMG VAR WL AAC DASDV ZC SSC
RMSMAV
RMS IEMG LOG
IEMG VAR
VAR WL
WL AAC
AAC DASDV
DASDV WAMP
ZC
ZC MYOP
SSC
SSC
WL AAC DASDV
MAV
MAV LOG
LOG WAMP
WAMP MYOP
MYOP1.0

Estimated Angle [NOMR]

1.01.0
0.8

Estimated Angle [NOMR]

Estimated Angle [NOMR]
0.80.8
0.6
0.60.6
0.4
0.40.4
0.2
0.20.2
0.0
1.0 0.8 0.6 0.4 0.2 1.0 0.8 0.6 0.4 0.2 1.0 0.8 0.6 0.4 0.2
0.00.0
Measured Angle [NORM] Measured Angle [NORM] Measured Angle [NORM]
1.0
1.0 0.8
0.8 0.6
0.6 0.4
0.4 0.20.2 1.01.0
1.0 0.8
0.8
0.8 0.6
0.6
0.6 0.4
0.4
0.4 0.2
0.2
0.2 1.01.0 0.80.8 0.60.6 0.40.4 0.20.2
Fig.Measured
5:
Measured Angle
Angle [NORM]
[NORM]
The relationship Measured
Measured
between the normalized elbow-joint
Measured Angle
Angle
angle
Angle [NORM]
to [NORM]
[NORM]
the normalized EMG featuresMeasured
during flexion
Measured (a-c)
Angle
Angle [NORM]
[NORM]
and extension (d-f) motion
for period of motion 10s.
(d) Extension. (e) Extension. (f) Extension
Fig.5:5: The
Fig. Therelationship
relationship betweenthethenormalized
between normalizedelbow-joint
elbow-jointangle
angle
toto
thethe normalized
normalized EMG
EMG features
features during
during flexion
flexion (a-c)
(a-c) and
and extension
extension (d-f)
(d-f) motion
motion
forforperiod
periodofofmotion
motion10s.
10s.
Fig. 5: The relationship between the normalized measured angle to the normalized EMG features during flexion (a), (b), and (c),
and extension (d), (e), and (f) for period of motion = 10 seconds.

4.3. The Effects of the Elbow Joint 0.889). In the extension trajectory, the estimated angle
Angle on EMG Signal Features based on the MYOP feature had the best Slope value
(ranged from −0.917 to −0.909 for periods of motion
10 seconds and 8 seconds, respectively).These values
Figure 5 shows a typical relationship between the esti- indicated that the estimated angle was almost linearly
mated and measured angles for both flexion and exten- related to the measured angle. The negative value in-
sion movements (period of motion = 10 seconds). The dicated a negative response between the measured an-
R2 and Slope values were calculated to evaluate the gle and the estimated angle. From the ANOVA test,
linear regression between the estimated and measured unfortunately, we found that there was a significant
angle as shown in Tab. 2 and Tab. 3. Table 2 shows difference (p-value < 0.05) between the Slopes during
that the R2 of the estimated angle from the MYOP extension and flexion movements. The Slope values
and WAMP features are higher and more consistent for decreased for the periods of motions from 10 seconds,
all periods of motion (ranged between 0.938 and 0.998) 8 seconds, 4 seconds and 2 seconds, respectively. Ide-
compared to those calculated from other features. This ally, the slope of the features should be constant so that
means that the estimated angles were fitted closely to the model can be used for any periods of motion. In
the measured angles. the future, a model that can compensate the decrement
of the slope is needed.
The R2 values of the model developed by Tang et
al. were 0.83, 0.87 and 0.79 for the periods of mo-
tion 2 seconds, 4 seconds and 8 seconds, respectively
[19]. Table 3 shows several of the Slope values for 5. Conclusion
various EMG features. In the flexion trajectory, the
estimated angle based on the WAMP and MYOP fea- This study presents an investigation of TD features
tures showed the best Slope value (ranged from 0.727 to to estimate the elbow joint angle using EMG features

c 2017
©©2015
2015 ADVANCES
ADVANCES IN IN ELECTRICAL
ELECTRICAL ANDELECTRONIC
AND ELECTRONICENGINEERING
ENGINEERING 456 3
ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 3

© 2015 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 3

BIOMEDICAL ENGINEERING VOLUME: 15 | NUMBER: 3 | 2017 | SEPTEMBER

based on a non-pattern recognition method. Some ternational Journal of Precision Engineering and
parameters such as the: Pearson’s Correlation Coef- Manufacturing. 2014, vol. 15, iss. 10, pp. 2185–
ficient, Root Mean Squared Error, Euclidean Distance, 2192. ISSN 2005-4602. DOI: 10.1007/s12541-014-
R2 , Linear Regression Slope were evaluated to obtain 0580-x.
the best EMG features in estimating the elbow joint
angle. The EMG signals for this study were collected [7] LI, Q., Y. SONG, Z. HOU and B. ZHU.
from biceps alone enabled us to estimate the elbow sEMG Based Joint Angle Estimation of Lower
joint angle. Our findings show that for a 10 sec- Limbs Using LS-SVM. In: 20th International
ond long recording period, the MyoPulse Percentage Conference on Neural Information Processing
(MYOP) Rate produced the best accuracy: with PCC (ICONIP). Daegu: Springer, 2013, pp. 292–300.
of 0.97 ± 0.02 (Mean±SD) and RMSE of 11.37 ± 3.04◦ ISBN 978-364242053-5. DOI: 10.1007/978-3-642-
(Mean±SD), respectively. The MYOP feature also 42054-2_37.
showed the highest R2 and Slope value 0.986 ± 0.0083
(Mean±SD) and 0.746±0.17 (Mean±SD), respectively [8] LOCONSOLE, C., S. DETTORI, A. FRISOLI,
for flexion and extension motions during all recorded C. A. AVIZZANO and M. BERGAMASCO. An
periods. EMG-based approach for on-line predicted torque
control in robotic-assisted rehabilitation. In: 2014
IEEE Haptics Symposium (HAPTICS). Houston:
References IEEE, 2014, pp. 181–186. ISBN 978-1-4799-3131-
6. DOI: 10.1109/HAPTICS.2014.6775452.
[1] OSKOEI, M. A. and H. HU. Myoelectric con- [9] REZA, S. M. T., N. AHMAD, I. A. CHOUD-
trol systems-A survey. Biomedical Signal Process- HURY and R. A. R. GHAZILLA. A Fuzzy
ing and Control. 2007, vol. 2, iss. 4, pp. 275–294. Controller for Lower Limb Exoskeletons
ISSN 1746-8094. DOI: 10.1016/j.bspc.2007.07.009. during Sit-to-Stand and Stand-to-Sit Move-
[2] TRIWIYANTO, O. WAHYUNGGORO, H. A. ment Using Wearable Sensors. Sensors. 2014,
NUGROHO and HERIANTO. Quantitative re- vol. 14, iss. 3, pp. 4342–4363. ISSN 1424-8220.
lationship between feature extraction of sEMG DOI: 10.3390/s140304342.
and upper limb elbow joint angle. In: Inter-
national Seminar on Application for Technol- [10] TSAI, A.-C., T.-H. HSIEH, J.-J. LUH and
ogy of Information and Communication (ISe- T.-T. LIN. A comparison of upper-limb mo-
mantic). Semarang: IEEE, 2016, pp. 44–50. tion pattern recognition using EMG signals dur-
ISBN 978-1-5090-2326-4. DOI: 10.1109/ISEMAN- ing dynamic and isometric muscle contractions.
TIC.2016.7873808. Biomedical Signal Processing and Control. 2014,
vol. 11, iss. 1, pp. 17–26. ISSN 1746-8094.
[3] PAU, J. W. L., S. S. Q. XIE and A. J. PUL- DOI: 10.1016/j.bspc.2014.02.005.
LAN. Neuromuscular Interfacing: Establishing an
EMG-Driven Model for the Human Elbow Joint. [11] LENZI, T., S. M. M. D. ROSSI, N. VI-
IEEE Transactions on Biomedical Engineering. TIELLO and M. C. CARROZZA. Intention-Based
2012, vol. 59, iss. 9, pp. 2586–2593. ISSN 1558- EMG Control for Powered Exoskeletons. IEEE
2531. DOI: 10.1109/TBME.2012.2206389. Transactions on Biomedical Engineering. 2012,
vol. 59, iss. 8, pp. 2180–2190. ISSN 1558-2531.
[4] TANG, Z., H. YU and S. CANG. Impact of Load DOI: 10.1109/TBME.2012.2198821.
Variation on Joint Angle Estimation From Sur-
face EMG Signals. IEEE Transactions on Neu- [12] LEE, S., H. KIM, H. JEONG and J. KIM. Anal-
ral Systems and Rehabilitation Engineering. 2016, ysis of musculoskeletal system of human during
vol. 24, iss. 12, pp. 1342–1350. ISSN 1558-0210. lifting task with arm using electromyography. In-
DOI: 10.1109/TNSRE.2015.2502663. ternational Journal of Precision Engineering and
[5] BENITEZ, L. M. V., M. TABIE, N. WILL, Manufacturing. 2015, vol. 16, iss. 2, pp. 393–398.
S. SCHMIDT, M. JORDAN and E. A. KIRCH- ISSN 2005-4602. DOI: 10.1007/s12541-015-0052-
NER. Exoskeleton Technology in Rehabilitation: y.
Towards an EMG-Based Orthosis System for Up-
per Limb Neuromotor Rehabilitation. Journal [13] PHINYOMARK, A., P. PHUKPATTARA-
of Robotics. 2013, vol. 2013, iss. 1, pp. 1–13. NONT and C. LIMSAKUL. Feature reduc-
ISSN 1687-9600. DOI: 10.1155/2013/610589. tion and selection for EMG signal classifica-
tion. Expert Systems with Applications. 2012,
[6] JANG, G., J. KIM, Y. CHOI and J. YIM. Human vol. 39, iss. 8, pp. 7420–7431. ISSN 0957-4174.
shoulder motion extraction using EMG signals. In- DOI: 10.1016/j.eswa.2012.01.102.

c 2017 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 457
BIOMEDICAL ENGINEERING VOLUME: 15 | NUMBER: 3 | 2017 | SEPTEMBER

[14] CHOWDHURY, R. H., M. B. I. REAZ, M. A. B. About Authors

M. ALI, A. A. A. BAKAR, K. CHELLAPPAN
and T. G. CHANG. Surface Electromyography TRIWIYANTO was born in Surabaya, Indonesia.
Signal Processing and Classification Techniques. He received his M.Sc. degree in Electronic Engineer-
Sensors. 2013, vol. 13, iss. 9, pp. 12431–12466. ing in Institute of Technology Sepuluh Nopember
ISSN 1424-8220. DOI: 10.3390/s130912431. in 2004, Surabaya, Indonesia. He is currently
a Ph.D. candidate in Electrical Engineering at Gadjah
[15] FOUGNER, A. L. Proportional MyoelectricCon-
Mada University, Yogyakarta, Indonesia. His research
trol of a Multifunction Upper-limb Prosthesis.
interests include biomedical signal analysis, embedded
Trondheim, 2007. Thesis. Norwegian University of
system, electronic instrumentation, assistive and
Science and Technology. Supervisor Tor Engebret
rehabilitation devices.
Onshus.

[16] LUCA, C. J. D. Surface Electromyography: De- Oyas WAHYUNGGORO was born in Jog-
tection and Recording. In: DelSys [online]. 2002. jakarta, Indonesia. He received his Ph.D. degree
Available at: http://www.ti.com/lit/an/ in Electrical and Electronic Engineering from the
slva372c/slva372c.pdf. Universiti Teknologi Petronas, Malaysia in 2011. His
research interests include biomedical signal processing,
[17] JANG, G., J. KIM, Y. CHOI and J. YIM. Human intelligent system, and control system.
shoulder motion extraction using EMG signals. In-
ternational Journal of Precision Engineering and Hanung Adi NUGROHO was born in Jog-
Manufacturing. 2014, vol. 15, iss. 10, pp. 2185– jakarta, Indonesia. He received his Ph.D. degree
2192. ISSN 2234-7593. DOI: 10.1007/s12541-014- in Electrical and Electronic Engineering from the
0580-x. Universiti Teknologi Petronas, Malaysia in 2012. His
research interests include biomedical signal processing
[18] TAN, L. Digital Signal Processing Fundamentals and image processing.
and Applications. 1st ed. San Diego: Elsevier,
2008. ISBN 978-0-0805-5057-2. HERIANTO was born in Jogjakarta, Indone-
sia. He received his D.Eng. degree in the Department
[19] TANG, Z., K. ZHANG, S. SUN, Z. GAO, of Mechanical and Control Engineering, Tokyo In-
L. ZHANG and Z. YANG. An upper-limb power- stitute of Technology, Japan, in 2009. His research
assist exoskeleton using proportional myoelectric interests include robotics and manufacture. His
control. Sensors. 2014, vol. 14, iss. 10, pp. 6677– current research is product design and development
6694. ISSN 1424-8220. DOI: 10.3390/s140406677. especially in rehabilitation robot.

c 2017 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 458