SEMG for Identifying Hand Gestures using ICA

Ganesh R. Naik

, Dinesh K. Kumar

, Vijay Pal Singh

and M. Palaniswami

School of Electrical and Computer Engineering, RMIT University,

GPO Box 2476V. Melbourne, Victoria-3001, Australia.

Department of Electrical and Electronic Engineering, The University of Melbourne,

Parkville, Victoria-3010, Australia.

Abstract. There is an urgent need for establishing a simple yet robust system that

can be used to identify hand actions and gestures for machine and computer con-

trol. Researchers have reported the use of multi-channel electromyogram (EMG)

to determine the hand actions and gestures. The limitation of the earlier works is

that the systems are suitable for gross actions, and when there is one prime-mover

muscle involved. This paper reports overcoming the difﬁculty by using indepen-

dent component analysis to separate muscle activity from different muscles and

classiﬁed using backpropogation neural networks. The system is tested and found

to be effective in classifying EMG.

1 Introduction

Identiﬁcation of hand gesture has numerous applications, primarily related to control-

ling machines and computers. Some of the commonly employed modalities include

vision based systems [1,2], mechanical sensors [3], and the use of electromyogram, an

indicator of muscle activity [4,5]. Electromyogram has an advantage of being easy to

record, and is non-invasive. Electromyography (SEMG) is a result of the spatial and

temporal integration of the motor unit action potential (MUAP) originating from differ-

ent motor units. It can be recorded non-invasively and used for dynamic measurement

of muscular function. It is typically the only in vivo functional examination of muscle

activity used in the clinical environment. The analysis of EMG can be broadly cate-

gorised into two; (i) gross and global parameters and (ii) decomposition of EMG into

MUAP. Hand movement is a result of complex combination of multiple muscles. While

Dinesh et al. [6] have reported success in the use of multiple channels SEMG record-

ing for the purpose, the precise location of the electrodes and multi-channel recordings

make the system complex. A single channel system where the location of electrodes is

not critical to the results is highly desirable. But the difﬁculty in the use of single chan-

nel when there is a complex group of muscles that control the movement is the very

large variation in the magnitude and frequency content of the signal when the distance

between the recording electrodes and the muscle ﬁbres is changed. To determine the

hand action based on the muscle activity, it is important to identify the muscle activity

of the different muscles responsible for the action. When attempting to identify small

level of muscle activity such as required for hand gestures, and with simultaneous acti-

vation of number of closely spaced muscles, there is considerable cross talk. Similarity

R. Naik G., K. Kumar D., Pal Singh V. and Palaniswami M. (2006).

SEMG for Identifying Hand Gestures using ICA.

In Proceedings of the 2nd International Workshop on Biosignal Processing and Classiﬁcation, pages 61-67

DOI: 10.5220/0001223500610067

 SciTePress

in the spectrum and other properties of the activity from the different muscles makes

the separation of these difﬁcult. There is a need to separate the muscle activity origi-

nating from different muscles. With little or no prior information of the muscle activity

from the different muscles, this is a blind source separation (BSS) task. Blind separation

of independent sources is an important research area. Independent component analysis

(ICA) is an iterative BSS technique which has been found to be very successful and

has found applications in audio and biosignal applications. ICA has been proposed for

unsupervised cross talk removal from SEMG recordings of the muscles of the hand [7].

Research that isolates MUAP originating from different muscles and motor units has

been reported in 2004 [8]. A denoising method using ICA and high pass ﬁlter banks has

been used to suppress the interference of electrocardiogram (ECG) in EMG recorded

from trunk muscles [9]. Muscle activity originating from different muscles can be con-

sidered to be independent, and this gives an argument to the use of ICA for separation

of muscle activity originating from the different muscles. This paper proposes the use of

ICA for separation of muscle activity from the different muscles in the forearm to iden-

tify the hand action. ICA is an iterative technique where the only model of the signals

is the independence, and the distribution. The natural outcome of this is that the signals

are separated without there being any information of the order of the sources. While

this difﬁculty is generally not consequential for audio signals, this would be of concern

when working with muscle activity. The spatial location of the active muscle activity

is the determining factor of the hand action and gesture. To overcome this difﬁculty,

one approach that has been reported is the use of prior knowledge of the anatomy. The

advantage of this approach is the model based approach that provides a well deﬁned

muscle activity pattern. The difﬁculty with that approach is the need for well deﬁned

location of the electrodes.

2 Basic Principles of Independent Component Analysis (ICA)

ICA separates signals from different sources into distinct components. The fundamen-

tals of ICA rest on information theory. The technique is based unsupervised learning

rules where reduction of mutual information and increase in Gaussianity can be consid-

ered to be the cost function. Given a set of multidimensional observations, which are

a result of linear mixing of unknown independent sources through an unknown mixing

source, ICA can be employed to separate the signals from the different sources. The

independent sources may be sources for audio signals such as speech, voice, music, or

signals such as bioelectric signals. If the mixing process is assumed to be linear, it can

be expressed as

x = As (1)

where x = (x

, x

, ..., x

) is the recordings, s = (s

, s

, ..., s

) the original signals and

A is the n x n mixing matrix of real numbers. This mixing matrix and each of the

original signals are unknown. To separate the recordings to the original signals, an ICA

algorithm performs a search of the de mixing matrix W by which observations can be

linearly translated to form Independent output components so that

s = Wx (2)

For this purpose, ICA relies strongly on the statistical independence of the sources s.

This technique iteratively estimates the un-mixing matrix using the maximisation of

independence of the sources as the cost function [10].

2.1 ICA for SEMG Applications

Signals from different sources can get mixed during recording. Often it is required to

separate the original signals, and there is little information available of the original sig-

nals. An example is the cocktail party problem. Even if there is no (limited) information

available of the original signals or the mixing matrix, it is possible to separate the origi-

nal signals using independent component analysis (ICA) under certain conditions. ICA

is an iterative technique that estimates the statistically independent source signals from

a given set of their linear combinations. The process involves determining the mixing

matrix. The independent sources could be audio signals such as speech, voice, music,

or signals such as bioelectric signals.

A number of researchers have reported the use of ICA for separating the desired

SEMG from the artefacts and from SEMG from other muscles. While details differ,

the basic technique is that different channels of SEMG recordings are the input of ICA

algorithm.

The fundamental principle of ICA is to determine the un-mixing matrix and use that

to separate the mixture into the independent components. The independent components

are computed from the linear combination of the recorded data. The success of ICA to

separate the independent components from the mixture depends on the properties of the

recordings.

2.2 Statistical Properties of SEMG Recordings

Signals from Gaussian sources cannot be separated from their mixtures using ICA

[10], making such signals unsuitable for ICA applications. Mathematical manipulation

demonstrates that all matrices will transform this kind of mixtures to another Gaussian

data. However, a small deviation of density function from Gaussian may make it suit-

able as it will provide some possible maximization points on the ICA optimization

landscape, making Gaussianity based cost function suitable for iteration. If one of the

sources has density far from Gaussian, ICA will easily detect this source because it will

have a higher measure of non Gaussianity and the maxima point on the optimization

landscape will be higher. If more than one of the independent sources has non Gaussian

distribution, those with higher magnitude will have the highest maxima point in the op-

timization landscape. Given a few signals with distinctive density and signiﬁcant mag-

nitude difference, the densities of their linear combinations will tend to follow the ones

with higher amplitude. Since ICA uses density estimation of a signal, the Components

with dominant density will be found easier. Signals such as SEMG have probability

densities that are close to Gaussian while artefacts such as ECG and motion artefacts

have non Gaussian distributions. From the above, it can be suggested that ICA may suit-

ably isolate some of the above signals, while its efﬁcacy for separating the others maybe

questionable. It is difﬁcult to identify the quality of separation of EMG from one mus-

cle and the neighbouring muscles, or that of EEG from one channel to the neighbouring

recording sites, making it difﬁcult to conﬁrm or negate the above.

3 Methodology

3.1 Experimental Procedure

Ethics approval by the RMIT university committee of ethics of experiment on human

subject was granted to conduct the experiment. One healthy male subject was recruited

for the experiment. A written consent was signed by the subject for the above exper-

iment. A proprietary SEMG acquisition system by Delsys (USA) was used for data

collection. Four differential electrodes with inter-electrode distance of 10mm and gain

of 1000 were placed on the four muscles of the subject’s forearm as outlined in the table

(1) as shown below.

Table 1. Placement of the Elctrodes over the skin of the forearm.

Channel Muscle Function

1 Brachioradialis Flexion of forearm

2 Flexor Carpi Ulnaris (FCU) Adduction and ﬂexion of wrist

3 Flexor Carpi radialis (FCR) Abduction and ﬂexion of wrist

4 Flexor digitorum superﬁcialis (FDS) Finger ﬂexion while avoiding wrist ﬂexion

Subjects were asked to keep the forearm resting on the table with elbow at an angle

of 90 degree in a comfortable position. Three hand actions were performed and repeated

15 times. There was no external load. The actions are listed below.

(1) Wrist ﬂexion (without ﬂexing the ﬁngers).

(2) Finger ﬂexion (ring ﬁnger and the middle ﬁnger together without

any wrist ﬂexion).

(3) Finger and wrist ﬂexion together but normal along centre line.

While Brachioradialis is an elbow ﬂexor, a very little activity may be recorded in

this muscle while ﬁnger and/or wrist ﬂexion. FCU and FCR the two wrist ﬂexors that

are responsible for adduction and abduction of the wrist respectively and they perform

the ﬂexion in the normal direction along the centre line together. FDS performs the

ﬂexion of the middle ﬁnger and the ring ﬁnger.

The hand actions and gestures represented low level of muscle activity. The hand

actions were selected based on small variations between the muscle activities of the

different digitas muscles situated in the forearm. The recordings were separated using

ICA to separate activity originating from different muscles and used to classify against

the hand actions.

3.2 Analysis

The aim of this experiment was to test the use of ICA for separation of the EMG signals

for the purpose of identifying hand gestures and actions. For the ﬁrst set of experiments

recorded signals x were analysed using matlab software package. There were approx-

imately 30000 samples of the data which was the result of 15 times wrist movements.

Since there were four Chanel (x) electrodes it formed 4 x 4 mixing matrix. For each set

of experiments the EMG data from 15 repetitions were analysed using fast ICA matlab

package. The mixing matrix A was computed for the ﬁrst set of data. The computed

mixing matrix had been kept constant (Same mixing matrix had been used through out

the one main experiment) for the remaining set of experiments and in each case the

sources had been computed using the following formula

x = As (3)

Where x is the recorded data, A mixing matrix and s is the sources. The sources are

recovered using the following formula

s = Bx (4)

Where B is the inverse of mixing matrix A. This process was repeated for each of

the three experiments.Each experiment resulted in four sources. Root Mean Squares of

each experiment were computed using the following formula

RMS

∑

i=1

(5)

Where s is the source and N is the number of samples.The examples of one set of

results and RMS values were shown in the table (2) below.

Table 2. Example of one set of sub experiment results showing the RMS (Root Mean Square)

values.

Source RMS (Root Mean Square) value

Source1(s1) 0.0461

Source2(s2) 0.0366

Source3(s3) 0.0311

Source4(s4) 0.0209

The above set of process had been repeated for the three actions. The outcome of

this was a set of 12 set of examples, each example pertaining to three actions. These

12 sets of examples were used to train a backpropogation neural network with 3 inputs

and 4 outputs. After training, the network conditions were saved and used to test the

network data from experiments not used to train the network. The ability of the network

to correctly classify the inputs against known hand actions were used to determine the

efﬁcacy of the technique.

4 Results and Observations

A result of testing the network using ﬁve set of experiments are tabulated below. The

classiﬁcation accuracy was 100% for all the experiments. The results are outlined in

table (3).

Table 3. Neural network testing results.

Action Performed Action identiﬁed for each sub experiments

Wrist ﬂexion 100% 100% 100% 100% 100%

Finger Flexion 100% 100% 100% 100% 100%

Finger ﬂexion and wrist ﬂexion 100% 100% 100% 100% 100%

5 Discussions and Conclusion

A new approach that combines semi-blind ICA along with neural networks was used

to separate and identify hand gestures. The results demonstrate that the technique can

be effectively used to identify hand gestures based on surface EMG when the level of

activity is very small. The authors would like to mention that this is early stage of the

work, and work needs to be done to identify inter-day variations. It is also important to

test the technique for different actions, and for a large group of people. Further, there is

need to automate the semi-blind operation.

References

1. Schlenzig, J., Hunter, E., Jain, R. : Vision based hand gesture interpretation using recursive

estimation, Vol. 2. Twenty-Eighth Asilomar Conference on Signals, Systems and Computers,

(1994) 1267 - 1271

2. Rehg, J. M., Kanade, T. : Vision-based hand tracking for human-computer interaction, IEEE

Workshop on Motion of Non-Rigid and Articulated Objects, (1994) 16 - 22

3. Pavlovic, V. I., Sharma, R., Huang, T. S. : Visual interpretation of hand gestures for human-

computer interaction, Vol. 19. IEEE Transactions on Pattern Analysis and Machine Intelli-

gence, (1997) 677 - 695

4. Cheron, G., Draye, J., Bourgeios, M., Libert, G. : A Dynamic Neural Network Identiﬁcation

of Electromyography and Arm Trajectory Relationship During Complex Movements, Vol.

43. IEEE Trans. Biomed. Eng, (1996) 552 - 558

5. Koike, Y., Kawato, M. : Human Interface Using Surface Electromyography Signals, Vol. 79.

Electronics and Communications, Japan, (1996) 15 - 22

6. Djuwari, D., Kumar, D. K., Polus, B., Raghupathy, S. : Multi-step independent component

analysis for removing cardiac artefacts from back semg signals, 8th Australian and New

Zealand Intelligent Information Systems Conference, Australia, (2003)

7. Greco, A., Costantino, D., Morabito, F. C., Versaci, M. A. : A Morlet wavelet classiﬁcation

technique for ICA ﬁltered SEMG experimental data, Vol. 1. Neural Networks Proceedings

of the International Joint Conference, (2003) 66 - 71

8. Hideo, Nakamura., Masaki, Yoshida., Manabu, Kotani., Kenzo, Akazawa., Toshio, Mori-

tani. : The application of independent component analysis to the multi-channel surface elec-

tromyographic signals for separation of motor unit action potential trains, Vol. 14. Journal of

Electromyography and Kinesiology, (2004) 423 - 432

9. Yong, Hu., Li, X. H., Xie, X. B., Pang, L. Y., Yuzhen, Cao., Luk, K. D. K. : Applying Inde-

pendent Component Analysis on ECG Cancellation Technique for the Surface Recording of

Trunk Electromyography, IEEE Engineering in Medicine and Biology 27th Annual Confer-

ence, Shanghai (2005)

10. Hyvarinen, A.,Karhunen, J., Oja, E. :Independent Component Analysis, John Wiley, New

York (2001)