Professional Documents
Culture Documents
1, MARCH 2009
69
1. Introduction
A human-computer interface (HCI) is a communication
system that uses bio-signals from brain or other parts of the
nervous system to control peripheral devices[1]. The arm
especially forearm is the most important tool for human.
Human forearm assists hand, wrist, and finger to achieve
specific actions[2]. An electric potential generated by
muscle contraction is called electromyogram (EMG)
signal[3]. Previous studies suggest that different compartments of the forearm muscle relate to hand, twist and finger
movement, and EMG signals can still be measured from the
forearm muscle even when the hand is lost[4]. EMG signals
can be used in various applications, such as computer
peripheral control[5], gesture discrimination[6], handwriting
recognition[7], artificial prosthesis control[8], and clinical
analysis and rehabilitation[9]. The key issue in a computer
peripheral control or artificial prosthesis control is the
control system[10].
Manuscript received December 10, 2008; revised January 5, 2009. This
work was supported by the National Natural Science Foundation of China
under Grant No. 60736029 and 30525030.
The authors are with the Key Laboratory for NeuroInformation of
Ministry of Education, School of Life, University of Electronic Science
and Technology of China, Chengdu, 610054, China (e-mail: dyao@
uestc.edu.cn).
JOURNAL OF ELECTRONIC SCIENCE AND TECHNOLOGY OF CHINA, VOL. 7, NO. 1, MARCH 2009
70
Analogy acquiring
Preamplifier
Electrode
Low-pass
filter
Notch
filter
Post
amplifier
(a)
ADC
Feature
extraction
Classify
Application
Digital
filter
Median
FCR nerve
BR
Ulnar
artery
APL
Ulnar
nerve
Ulnar
vein
ECRL
Superficial
radial nerve
ECRB
Posterior
interosseous
nerve
Antebrachial fascia
EDM
Interosseous membrane
In order to prevent polarization effect, the gain of preamplifier is not enough to amplify EMG signal. The post
amplifier boosts the signal amplitude and adjusts the offset.
Analog-to-digital converter (ADC) changes the analog
signal to digital signal for digital signal processing. The
high-precision ADC can provide good digital data for
classification algorithms.
We must put the electrode on the right forearm position
because different compartments of forearm muscle relate to
hand, finger, and wrist movements[12]. Fig. 2 shows the
cross section of mid forearms muscle compartments and
nerves[13].
The selected EEG measurement system has high
CMRR (110 dB) and high precision ADC (16 bits).
Although EEG signal and EMG signal belong to weak
physical signal, the raw EEG signal magnitude is on the
order of a few microvolts and the raw EMG signal
3. Classification Algorithm
As shown in Fig. 1 (b), the classification algorithm is
used to classify features, which are extracted from digital
signals filtered by digital filter, and mapped classification
results to corresponding control commands[14],[15].
3.1 Feature Extraction
EMG signal power spectral density is non-stationary
since forearm EMG is a complex nonlinear signal and may
rapidly change when muscle condition is rest or
contraction[16]. Based on this character, PSD values are
extracted from EMG signals.
There are two ways to estimate power spectrum: one is
to estimate the signal auto-correlation function and then to
solve its Fourier transform, known as the auto-correlation
method. It can be denoted as
1
R x ( m) =
N
Px (e j ) =
N | m | 1
xn xn + m , m = 0, 1, 2,"
(1a)
n=0
N 1
m =
m = ( N 1)
R x (m)e j m =
R x (m)e j m
(1b)
X (e j ) = DTFT [ x(n)]
(2a)
2
1
Px (e j ) =
X (e j )
N
(2b)
up
up
down
down
left
left
right
right
yes
yes
no
no
20
40
60
p(x | i ) =
1
(2 )
l/2
Ci
1/ 2
2360
matrix in l l dimension.
The discriminant function is denoted as
up
down
left
right
yes
no
2350
2340
2330
2320
2310
2300
10
20
30
40 50
60
Frequency (Hz)
70
(3)
p ( x | i ) P ( i )
(10)
80
P (i | x) =
(7)
2290
0
71
(4)
i =1
(5)
If P (1 | x) < P (2 | x) then x 2 .
(6)
exp (x i )T Ci1 (x i )
2
(11)
yi (x) = N
c
1/ 2
1
T
1
exp (x j ) C j (x j )
Cj
2
j =1
Ci
1/ 2
4. Experiment Results
We define six hand poses that represent different
control commands of control system: up, down, left, right,
yes, and no, as shown in Fig. 5. The rest state is that hand
parallels to forearm and hand does not flex or extend. There
are four states: up, down, left, and right compatible
with wrist flexion or extension in the same direction of
literal meaning of command. The yes is hand grasp and
the no is palm extension. All motions are completed in
the manner that forearm parallels to the ground. Forearm
positions in Fig. 5 are changed in order to be conveniently
understood. Every run is in the same order: rest, up, rest,
down, rest, left, rest, right, rest, yes, rest, and no. Each pose
holds five seconds. EMG data set has ten trials. Three EMG
data sets are used for Bayesian classifier.
The EMG signals are measured with four Ag/AgCl
surface electrodes. In order to collect the preferable EMG
on forearm, the electrodes are placed on the forearm
according to physiological mechanism of movement of
muscle. Four electrodes are separately tied on surface skin
four muscle groups: flexor caripi radialis, flexor digitorum
profundus, extensor carpi ulnaris, extensor digitorum.
JOURNAL OF ELECTRONIC SCIENCE AND TECHNOLOGY OF CHINA, VOL. 7, NO. 1, MARCH 2009
72
References
Yes
Up
Right
Rest
Left
Down
No
Channels 1
35%
65%
30%
Channels 2
65%
45%
50%
Channels 3
40%
70%
40%
Channels 4
40%
40%
55%
5. Conclusion
In this paper, we propose a new control system with the
commands of six pre-defined hand poses. The Bayesian
classifier is used to classify the power spectral density of
the signal. In order to examine the effectiveness of this
control system, an experiment is carried out to collect the
EMG signals with EEG amplifier. The experiment results
show that the control system has high accuracy in two
relative commands and single electrode measurement can
also perform well. The control system can be widely used
in computer peripheral control and artificial prosthesis
control. Our future works will focus on classification of
multi-class hand poses and on-line processing.
Acknowledgment
The authors would like to thank UESTC Youth Foundation
under Grant No. L08010901JX0772 for support.
[1] I. Moon, M. Lee, J. Chu, and M. Mun, Wearable EMGbased HCI for electric-powered wheelchair users with motor
disabilities, in Proc. 2005 IEEE International Conf. on
Robotics and Automation, Barcelona, 2005, pp. 2649-2654.
[2] L. J. Hargrove, K. EnglehartE, and B. Hudgins, A comparison of surface and intramuscular myoelectric signal
classification, IEEE Trans. Biomedical Engineering, vol. 54,
no. 5, pp. 847-853, May 2007.
[3] K. R. Wheeler, M. H. Chang, and K. H. Knuth, Gesturebased control and EMG decomposition, IEEE Trans.
Systems, Man, and Cybernetics-Part C: Applications And
Reviews, vol. 35, no. 4, pp. 503-514, 2006.
[4] Y. Su, M. H. Fisher, A. Wolczowski, G. D. Bell, D. J. Burn,
and R. X. Gao, Towards an EMG-controlled prosthetic hand
using a 3-D electromagnetic positioning system, IEEE Trans.
Instrumentation and Measurement., vol. 56, no. 1, pp.
187-186, Feb. 2007.
[5] L. J. Trejo, K. R. Wheeler, C. C. Jorgensen, R. Rosipal, S.
Clanton, and B. Matthews et al., Multimodal neuroelectric
interface development: a survey of research at NASA Ames
research center, IEEE Trans. Neural Systems and Rehabilitation Engineering, vol. 11, no. 2, pp. 199-203, 2003.
[6] K. K. Jung, J. W. Kim, and H. K. Lee, EMG pattern
classification using spectral estimation and neural network,
in Proc. SICE Annual Conf., Japan, 2007, pp. 1108-1111.
[7] A. Lansari, F. Bouslama, M. Khasawneh, and A. A. Rawi, A
novel electromyography (EMG) based classification
approach for Arabic handwriting, in Proc. International
Joint Conf. Neural Networks, Portland, 2003, pp. 2193-2196.
[8] C. Cipriani, F. Zaccone, S. Micera, and M. C. Carrozza, On
the shared control of an EMG-controlled prosthetic hand:
analysis of user-prosthesis interaction, IEEE Trans. Robotics,
vol. 24, no. 1, pp. 170-184, Feb. 2008.
[9] K. T. Ebersole and D. M. Malek, Fatigue and the electromechanical efficiency of the vastus medialis and vastus
lateralis muscles, Journal of Athletic Training, vol. 43, no. 2,
pp. 152-156, Apr. 2008.
[10] R. Kato, T. Fujita, H. Yokoi, and T. Arai, Adaptable EMG
prosthetic hand using on-line learning methodinvestigation
of mutual adaptation between human and adaptable
machine, in Proc. 15th IEEE International Symposium on
Robot and Human Interactive Communication, Roman, 2006,
pp. 599-604.
[11] K. Kiguchi, T. Tanaka, and T. Fukuda, Neuro-fuzzy control
of a robotic exoskeleton with EMG signals, IEEE Trans.
Fuzzy Systems, vol. 12, no. 4, pp. 481-490, Aug. 2004.
[12] Y. Su, A. Wolczowski, M. H. Fisher, G. D. Bell, D. Burn,
and R. X. Gao, Towards an EMG controlled prosthetic hand
using a 3D electromagnetic positioning system, in Proc.
IMTC 2005Instrumentation and Measurement Technology
Conf., Ottawa, 2005, pp. 261-266.
[13] C. A. Boles, S. Kannam, and A. B. Cardwell, The forearm
anatomy of muscle compartments and nerves, American
Journal of Roentgenology, vol. 174, pp. 151-159. Jan. 2000.
[14] M. Arvetti, G. Gini, and M. Folgheraiter, Classification of
EMG signals through wavelet analysis and neural networks
73