[ieee 2010 third international workshop on advanced computational intelligence (iwaci) - suzhou,...

6
Abstract—The paper presents an on-line brain-computer interface (BCI) based on visual evoked potential (VEP) P300. The BCI is applied to control a multi-DOF manipulator. This BCI system includes five modules which are visual stimulator, signal acquisition, data processing, communication and motion control of manipulator. Stimulation program and experimental scheme are designed based on LabVIEW platform and virtual instrument technology. The manipulator with the ability of six-direction-free moving and two-direction operation including grasping and relieving is self-designed and developed. In the experiment, the subject chooses the right oddball on a CRT/LCD displayer with eight blocks, those are forward, backward, up, down, left, right, grasp and release, and gazes at it. The electroencephalography (EEG) is sampled to extract the P300 characteristic. The algorithms of peak extraction, correlation analysis and wavelet transform are used to analyse EEG. Based on the comparing of the result of the algorithms, wavelet transform is select to extract the feature of EEG. The manipulator is controlled to move or operate by the subject’s EEG with wire or wireless communication. The experiments show that the subject with little training can control the manipulator. The improvement for the future research is also available in the paper. I. INTRODUCTION rain-computer interface (BCI) is applied in more and more fields such as patient’s recovery, special communication of disabled person. BCI is a direct communication channel between the human’s brain and the external world [1]. BCI is a radically new human-computer interface system. The intrinsic feature of a BCI is that it does not depend on the brain’s normal output pathways of peripheral nerves and muscles [2]. The system acquires and analyses the bioelectricity signals and the goal is to create a communication channel directly between the brain and Manuscript received April 10, 2010. This work is supported by fund of Chinese National Programs for High Technology Research and Development Project (2007AA04Z254), Tianjin, China Science & Technology Research and Development Project (08ZCKFSF03400), Chinese Academy of Sciences (CAS) Supporting Tianjin’s Science & Technology Research and Development Project (TJZX2-YW-06), China Postdoctoral Science Foundation (20090460501) and Science & Technology fund of Tianjin University of Technology and Education (YJS10-05). Shigang Cui is with the Tianjin Key Laboratory of Information Sensing and Intelligent Control, Tianjin, 300222, China (phone: 86-22-88181116; fax: 86-22-88181115; e-mail: cuisg@ 163.com). Xiong Su is with the Tianjin Key Laboratory of Information Sensing and Intelligent Control, Tianjin, 300222, China (e-mail: [email protected]). Genghuang Yang is with the Tianjin Key Laboratory of Information Sensing and Intelligent Control, Tianjin, 300222, China (e-mail: [email protected]). Li Zhao is with the Tianjin Key Laboratory of Information Sensing and Intelligent Control, Tianjin, 300222, China (e-mail: [email protected]). external device. The great development of computer or microprocessor improves complex data processing ability. The technology in bioelectricity signal processing, especially the algorithm of EEG feature extraction, makes real-time BCI system come true [3][4]. This paper presents an on-line brain-computer interface based on VEP P300 [5][6]. The system can control multi-DOF manipulator’s movements and two simple operations. The system includes visual stimulator, EEG acquisition, data processing, communication between the manipulator and BCI, and motion control of multi-DOF manipulator. The manipulator’s motion is controlled by EEG VEP. VEP stimulator based on classic paradigm P300 is designed to stimulate subject for P300. In the part of signal acquisition, evoked EEG is pre-processed, then the data is analysed by three methods, those are peak value, wavelet transform, correlation analysis. It is difficult to extract P300 in a short time as the typical algorithm requires a long data for split and addition. Based on many experiments and the comparison of the result, wavelet transform is select to extract feature of EEG for its higher accuracy rate than the other two methods. Therefore, the result based on wavelet transform is applied to control manipulator. There are eight control commands, six of which are the directions such as forward, backward, up, down, left, right, two of which are operations such as grasping and relieving. In the communications module, command is delivered through the wireless radio or RS-485 to control the movement of manipulator real-timely. This system is built on LabVIEW platform by virtual instrument technology. The block diagram is shown in figure 1. Fig. 1 Block diagram of manipulator II. SIGNAL PROCESSING A. Signal Acquisition 1) Hardware For EEG Aacquisition Design and Implementation of a Virtual Instrumentation Based Brain-Computer Interface Shigang Cui, Xiong Su, Genghuang Yang and Li Zhao B 116 Third International Workshop on Advanced Computational Intelligence August 25-27, 2010 - Suzhou, Jiangsu, China 978-1-4244-6337-4/10/$26.00 @2010 IEEE

Upload: lecong

Post on 05-Feb-2017

215 views

Category:

Documents


2 download

TRANSCRIPT

Abstract—The paper presents an on-line brain-computer interface (BCI) based on visual evoked potential (VEP) P300. The BCI is applied to control a multi-DOF manipulator. This BCI system includes five modules which are visual stimulator, signal acquisition, data processing, communication and motion control of manipulator. Stimulation program and experimental scheme are designed based on LabVIEW platform and virtual instrument technology. The manipulator with the ability of six-direction-free moving and two-direction operation including grasping and relieving is self-designed and developed. In the experiment, the subject chooses the right oddball on a CRT/LCD displayer with eight blocks, those are forward, backward, up, down, left, right, grasp and release, and gazes at it. The electroencephalography (EEG) is sampled to extract the P300 characteristic. The algorithms of peak extraction, correlation analysis and wavelet transform are used to analyse EEG. Based on the comparing of the result of the algorithms, wavelet transform is select to extract the feature of EEG. The manipulator is controlled to move or operate by the subject’s EEG with wire or wireless communication. The experiments show that the subject with little training can control the manipulator. The improvement for the future research is also available in the paper.

I. INTRODUCTION rain-computer interface (BCI) is applied in more and more fields such as patient’s recovery, special communication of disabled person. BCI is a direct

communication channel between the human’s brain and the external world [1]. BCI is a radically new human-computer interface system. The intrinsic feature of a BCI is that it does not depend on the brain’s normal output pathways of peripheral nerves and muscles [2]. The system acquires and analyses the bioelectricity signals and the goal is to create a communication channel directly between the brain and

Manuscript received April 10, 2010. This work is supported by fund of

Chinese National Programs for High Technology Research and Development Project (2007AA04Z254), Tianjin, China Science & Technology Research and Development Project (08ZCKFSF03400), Chinese Academy of Sciences (CAS) Supporting Tianjin’s Science & Technology Research and Development Project (TJZX2-YW-06), China Postdoctoral Science Foundation (20090460501) and Science & Technology fund of Tianjin University of Technology and Education (YJS10-05).

Shigang Cui is with the Tianjin Key Laboratory of Information Sensing and Intelligent Control, Tianjin, 300222, China (phone: 86-22-88181116; fax: 86-22-88181115; e-mail: cuisg@ 163.com).

Xiong Su is with the Tianjin Key Laboratory of Information Sensing and Intelligent Control, Tianjin, 300222, China (e-mail: [email protected]).

Genghuang Yang is with the Tianjin Key Laboratory of Information Sensing and Intelligent Control, Tianjin, 300222, China (e-mail: [email protected]).

Li Zhao is with the Tianjin Key Laboratory of Information Sensing and Intelligent Control, Tianjin, 300222, China (e-mail: [email protected]).

external device. The great development of computer or microprocessor improves complex data processing ability. The technology in bioelectricity signal processing, especially the algorithm of EEG feature extraction, makes real-time BCI system come true [3][4].

This paper presents an on-line brain-computer interface based on VEP P300 [5][6]. The system can control multi-DOF manipulator’s movements and two simple operations. The system includes visual stimulator, EEG acquisition, data processing, communication between the manipulator and BCI, and motion control of multi-DOF manipulator. The manipulator’s motion is controlled by EEG VEP. VEP stimulator based on classic paradigm P300 is designed to stimulate subject for P300. In the part of signal acquisition, evoked EEG is pre-processed, then the data is analysed by three methods, those are peak value, wavelet transform, correlation analysis. It is difficult to extract P300 in a short time as the typical algorithm requires a long data for split and addition. Based on many experiments and the comparison of the result, wavelet transform is select to extract feature of EEG for its higher accuracy rate than the other two methods. Therefore, the result based on wavelet transform is applied to control manipulator. There are eight control commands, six of which are the directions such as forward, backward, up, down, left, right, two of which are operations such as grasping and relieving. In the communications module, command is delivered through the wireless radio or RS-485 to control the movement of manipulator real-timely. This system is built on LabVIEW platform by virtual instrument technology. The block diagram is shown in figure 1.

Fig. 1 Block diagram of manipulator

II. SIGNAL PROCESSING

A. Signal Acquisition

1) Hardware For EEG Aacquisition

Design and Implementation of a Virtual Instrumentation Based Brain-Computer Interface

Shigang Cui, Xiong Su, Genghuang Yang and Li Zhao

B

116

Third International Workshop on Advanced Computational Intelligence August 25-27, 2010 - Suzhou, Jiangsu, China

978-1-4244-6337-4/10/$26.00 @2010 IEEE

Preamplifier and signal acquisition card are included in the hardware of EEG sampling. As the voltage of EEG ranges from -100uV to 100uV, it is necessary to amplify. Preamplifier amplifies EEG signal by 20,000 times [7] [8].

PXI card of National Instrument (NI) is adopted to sample the EEG after preamplifier. The card works in the synchronous acquisition mode. Industrial Personal Computer (IPC) PXI 1042 is selected to be the hardware platform. PXI-6070 is a type of high precision-acquisition card which includes 16 channels of analog input, two analog outputs, eight digital I/O ports, and a timer counter. Also PXI-6070 provides high-performance for data acquisition capability, with 12 bit resolution for input and output, maximum sampling rate of 1.25 MS / s, output speed up to 1 MS / s. The range of signal input is ± 0.05 to ± 10V which is fit for EEG sampling. The acquisition card can be programmed by the tools which are supported by configuration acquisition card item [9][10] of Measurement & Automation Explorer. Only several setting is required before the card begin to work.

2) Software For EEG Acquisition Work of software programming based on data acquisition

card is important. Figure 2 shows the programming of data acquisition processing. Procedure is implemented as the following order. First of all, hardware and software are initialized (AI Configure .vi). Secondly analog input starts(AI Start.vi). Thirdly reading data from channels are enabled (AI Read.vi); Finally data form, which is two-dimensional array, is transmitted to data processing module (Data Process). After processing is completed, it is required to end the analog input operations (AI Clear.vi).

In addition, circulatory function should be enabled before continuous acquisition is available. In order to analyse sampling data in off-line way, it is necessary to add data storage procedures (Data Storage). Upon completion of acquisition, data will be read from the buffer, and be deposited into document text [11].

Fig. 2 Flowchart of acquisition program

B. EEG Signal Processing Signal processing is the crucial part of this system, which

completes EEG signal processing and command output. The system's signal processing is completed on LabVIEW platform. EEG processing usually contains 2 major steps: signal pre-processing and feature extraction. The former is to improve signal to noise ratio (SNR), while the latter is to extract useful signal from EEG.

1) Signal Pre-processing Pre-processing includes four steps as figure 3 shows:

signal division blocks average, low-pass filtering, remove trend & baseline drift [12][13] and signal de-noising.

Fig.3 Signal Pre-processing Flowchart

Section of “signal division blocks average” is used to make the waveform smooth. Data window is moved along the sampling points. The sampling data in the window is adding up and then divided by the number of sampling points in the window. Section of “low-pass filter” is used to eliminate the signal of higher frequency. Section of “remove trend & baseline drift” is used to regulate the whole waveform.

In the part of pre-processing, superposition is the most important method. Because of the interference of spontaneous EEG, it is difficult to extract P300 feature. Multiple stacking is a good way to improve SNR. Noise in EEG is significantly weakened by superposition.

Figure 4 show the raw EEG waveform during a period of 3.6 seconds after superposition.

Fig. 4 Raw EEG waveform after superposition

Classic filter is used in processing of non-stationary signal to weaken the noise. The processing broadens the waveform and makes the signal peak smooth. Excessive using results in losing important information especially on the points of waveform peak. Wavelet transform is a method of time-frequency analysis which observes the waveform in the time and frequency scales. It can overcome the deficiencies above all [13]. Wavelet transform is used to eliminate low-frequency trends, baseline drift [9].

There is a function ( ) ( )2t L Rψ ∈ , if its Fourier transform

( )ψ ω satisfy conditions shown as formula (1),

( ) 2

C dR

ψ ωωψ

ω= < ∞∫ (1)

( )tψ is called basic wavelet or mother wavelet function. 1 *( , ) ( ) ta f t dtf a aR

WT ττ ψ −⎛ ⎞= ∫ ⎜ ⎟⎝ ⎠

(2)

117

Formula (2) is the algorithm for continuous wavelet transform. The coefficients ( )ψ ω based on the algorithm

are the result of wavelet transform of ( )tψ . Inverse wavelet transform formula is shown as the following.

1( , ) ( )0 ,2( ) da WT a t df aC a

f t τ ψ ττψ+∞ +∞∫ ∫−∞= (3)

Wavelet soft-threshold de-noising [14] is used in the section of “de-noising”. Several steps are included.

(1) decompose the raw EEG with wavelet transform (mother wavelet is Db4, and number of scales is 3;

(2) quantize the threshold of every sequence of coefficients;

(3) signal reconstruction using coefficients that has been quantized.

(a)

(b)

(c)

(d)

Fig. 5 Decomposition coefficient of three-scale Db wavelet

Figure 5 shows four sequences of wavelet coefficients

obtained by using wavelet transform to decompose EEG. These figures are noted by a ~ d. EEG from low frequency to high frequency in the three scales form sequence of wavelet coefficient. Each sequence of wavelet coefficients corresponds to the scope of information in different frequency bands. The detail of four EEG’s basic frequency is as the following: D1: 16 ~ 32Hz; D2: 8 ~ 16Hz; D3: 4 ~ 8Hz; A3: 0.5 ~ 4Hz.

High-frequency (frequency higher than 25 Hz), mainly in the cd1, high coefficient is eliminated near to zero by interference. In cd3 and cd2, eye clap and electromyography (EMG) interference still exists because frequency domain of the eyes moving and electromyography interference is overlapped with that of useful signal. Db4 wavelet is used as the mother wavelet because of its good support for orthogonal and compactness. Db4 wavelet is applied in many fields and has got great result in engineering application. Figure 6 shows the waveform of signal pre-processed.

Fig. 6 Signal after pre-processing

2) P300 Feature Extraction In this paper, three methods are tested to extract P300

feature of EEG. (1) Peak extraction Peak extraction is a conventional method for feature

extraction of P300. Waveform peak detection VI in signal processing templates is used to detect peak and trough. Quadratic fit is adopted. The input signal is processed by setting the threshold and data width to locate the peak or trough and to calculate the amplitude. Width is set based on the least-squares fitting for the number of data points. Width can not be greater than the half of width of the peak or trough. The sampling rate is 250Hz, and there are 900 samples. The threshold value is set as the half of maximum EEG data array, and width is 3. Figure 7 shows the EEG peaks map after peak detection. The total number of peaks is 75. Fitting algorithm is used for the secondary peak detection. Peak position does not match to the actual location of peaks. Peak position of the threshold can be used in the conversion formula as the following shows.

0[ ] * [ ]TimeLocation i t dt Locations i= + (4)

Fig. 7 EEG peak map

(2) Correlation analysis

118

Correlation analysis is widely used in random signal analysis. It can be used to analyse the linear relationship between two signals. There are two sequences of sampling signal ( )X t and ( )Y t with linear correlation and waveform relevance.

1 2, 3( ) { , ..... }nX t x x x x= (5)

1 2, 3( ) { , ..... }nY t y y y y= (6)

,

,2 2

[( )( )]

[( )] [( )]

x y x y

x y

x y x y

E x y

E x E y

σ μ μρ

σ σ μ μ

− −= =

− − (7)

The correlation between variables X(t) and Y(t), ,x yρ is the

degree of correlation coefficient. ,x yσ is the covariance of random variable X and Y;

xμ , yμ are the mean of X and Y.

According to Cauchy - Schwarz inequality: 2 2 2[( )( )] [( ) ] [( ) ]x y x yE x y E x E yμ μ μ μ− − ≤ − − (8)

,0 1x yσ≤ ≤ .

,x yρ =1 indicates that the linear correlation between two

random variables exists;

,x yρ =0 indicates that two random variables are unrelated.

The correlation function is adopted to calculate the similarity between two signals. Superposed average is the standard P300 signal for correlation analysis, the goal of experiment is to calculate the similarity between standard signal and VEP signal. The higher the similarity is, the more possibility the VEP is. In the experiment of brain-computer interface, the EEG data is collected by superposition. The realization is as the following.

First of all, use soft threshold of discrete wavelet analysis. Get the signal which is much closer to standard ERP signal.

Secondly, adjust the amplitude of the standard sample, the scope of its amplitude is set with the measured signal in the same order of magnitude, and then each group is the calculated signal with the standard of mutual relations between the numbers of samples.

Thirdly, compare the relationship between each other and obtain the greatest number of rows of data and disaggregated data which is used to determine the subjects in the experiment. Monitor the goals and objectives of line out, and then find the target subjects’ characteristic.

(3) Wavelet analysis Signal with noise is decomposed on multi-resolution by

wavelet transform, dispersion coefficient details and discrete approximation coefficient can be acquired. It has been proved that amplitude of discrete detail noise signal will decrease with the increase of wavelet scale, but the useful signal will not. On the other hand, researches indicate that frequency band of P300 is mainly concentrated in the low-frequency. In the low-frequency scale, positive wavelet coefficients that are representative of the P300 will exist. The basic idea to extract

P300 feature by wavelet transform is to detect positive wavelet coefficients.

In order to demonstrate the correctness of conclusion above, we will apply the wavelet analysis to extract the P300 feature. Figure 8 shows this processing. In left of figure 8, there are five non-response P300 target waveforms that 3-scale DB4 wavelet is used in wavelet decomposition. In the right of figure, there are five response P300 target waveforms that 3-scale DB4 wavelet is used in wavelet decomposition. Comparing waveforms in the left and right, in the discrete approximation coefficient and in the specific period 300~ 600ms, it is observed that, in 400ms nearby, there are obviously positive peaks in wavelet decomposition coefficients of P300 corresponding samples.

Fig. 8 P300 response and P300 non-response coefficient

III. MULTI-DOF MANIPULATOR MOTION CONTROL The manipulator contains 1 rotary module and 2 mobile

modules, as figure 9 shows. For rotary module, DC motor is adopted as driving device, while stepper motor for mobile module. In addition, incremental rotary encoder with resolution of 500p/n is also equipped for rotary module, to provide feedback signal required in semi-closed loop control. At the end of 2 mobile modules, trip switches are installed to limit the range within 0~256 mm. For rotary module, there are 2 electromagnetic proximity switches to make the whole pillar moving within the scope of ±180 degree. The manipulator implementation terminal is a fixture controlled by a rudder. ATmegaL16 microprocessor is applied to control the manipulator. The servo control system is shown in figure 9 and figure 10.

119

Fig.9 Structure of Manipulator

Fig.10 Servo control system of the manipulator

A. Communications Between Manipulator And BCI The communication protocol between manipulator and

BCI system is serial communication, and the command dispatched through D-link wireless networking, which adopt multiple Input Multiple Output (MIMO) technology. MIMO technology which is considered as multiple inputs and multiple outputs technology was raised by Bell Labs at the end of last century. It is the multi-antenna communication systems, with multi-antenna (or antenna array) and multi-channel used at both transmitter and receiver. It is the diversity spatial multiplexing and beam forming technology that is the key to upgrade networking transmission speed and transceiver capacity, with the basic tenet of searching for signal paths that can provide data and high transmission serials at the same time, and avoid ones which can possibly lead to packet errors, so that transfer rate can be upgraded.

Characteristic signal generated by brain-computer processing module cannot control multi-DOF manipulator directly, and re-encoding is needed for serials of switch quantity and corresponding portfolio, then the acquired motion control command can express the way of motion. There is a communication protocol format between remote PC and manipulator. During communication, “one question, one answer” protocol is strictly complied between host computer and manipulator. If the date feedback, feedback what is supposed to; if not, then “NULL Orders” will return. Serial port baud rate is 19200bps, 8 data bit, 1 stop bit, and no parity bit.

B. Software Of The Control System Figure 11 shows the P300-based multi-DOF manipulator

control system interface.

Fig.11 P300-based multi-DOF manipulator control system interface

There are eight blocks which represent eight basic order for the manipulator, due to the module size limit, some use the word abbreviation. Forw. represents forward movement, and back, the abbreviation of backward, represents backward movement. Up represents upward movement, down represents downward movement, right represents moving to right, grasp represents turning on a key, rele., abbreviation of the releasing, represents camera turning off. When experiment starts, eight blocks are flashing randomly, and induce a potential ERP. The results of running will be displayed below the flash module, As previously mentioned, three methods are used, in the experiment, it can be set which method(s) are used to control the Multi-DOF Manipulator

In the experiment, three different methods are applied for pattern recognition. The subjects receive the results of pattern recognition by feedback result in the dialog box. The cycle time of the flashing blocks can be set as "light" time and "dimmed" time.

As the above describes, the system can control the manipulator to move in six directions or operation in two simple ways. The movement of forward and backward is based on the gear driver which is absent in figure 9. The movement of the manipulator is based on positive x-axis, negative x-axis, positive y-axis, negative y-axis, positive rotation angle, negative rotation angle. The length and angle of the movement based on one order is set in the robot microcontroller. For example, if the length of x-axis is set 5cm, the manipulator will move 5 cm when it gets the order of left or right.

IV. EXPERIMENT BASED ON CONTROL SYSTEM

A. Experimental Processing This section is about the experiment of the P300-based

multi-DOF manipulator real-time motion control. To test the stability of the system, the commands are random. In the experiments, the guider gives an order and turn on the system. The subjects sit at front of the CRT (or LCD) monitor with the 50cm distance and gaze at the selected block noted on the dialog box. Before each command starts, each subject have 10 seconds for ready, During the experiment, the subject

120

should be highly concentrated, not to blink any eyes, the subjects have to maintain the status of 2 seconds after the flashing blocks stop. After about 3 seconds, the result of pattern recognition comes out in the monitor. At the same time, the order is sent to the robot microcontroller to move or operate the camera carrier. Each subject redoes the experiment for 3-4 times. Also the consecutive orders can be input as a row in the LabVIEW platform. The subjects under fatigue lead to poor test results, so the number of commands is not more than 4.

B. Comparison Of Three Methods And Result Three extraction methods are adopted for pattern

recognition. Many experiments indicate that method of wavelet transform is superior to the other two methods. Being different from the other two methods of linear analysis mentioned above, wavelet transform is a method of non-linear analysis. The advantages is its multi-resolution, it can detect feature of signal in different scales.

This manipulator motion adopts the step-by-step approach control as described in the former chapter. When a command transmission, the robot will implement 5 cm or 10 degree angle displacement in the corresponding direction.

Table I shows the parameter of two subjects in the test. The column of number means the times of the whole test, objective response means real times of orders, P300 response means the number of right order obtained, and accuracy rate is the ratio of P300 response. The statistics also shows that the subjects will get more satisfied result after more experiments.

TABLE.I THE ACCURACY OF STATISTICS OF TWO SUBJECTS

V. DISCUSSION AND CONCLUSION Two factors are used to evaluate a BCI. One is accuracy

rate, the other is speed. The key issue of BCI research remains how to improve accuracy rate of EEG pattern recognition. Although real-time control have been achieved, the two factors are not satisfied in the test especially in the application of patience’s recovery and assistance for disable person. Communication with higher speed will improve the system’s performance. The accuracy rate is unstable at the condition of fewer overlap. After about 20 overlaps, the accuracy rate is gradually stabilized. Increase of overlap times increases running time of the system procedure, and reduce the speed. For example, in the case of 20 overlaps and 8 command options, an average of 60 seconds is needed to start the motion.

The BCI system can be improved by the following three methods:

1) Speed up the system. One way is to reduce the single run-time. More complex stimulus mode is in badly need to shorten the time of “light" and "dimmed". The other way is to reduce the overlap times of waveform.

2) To increase the system's adaptive capacity and the level of automation. As an effective BCI system, it should take differences of individual subjects into account. Adjustment of algorithm leads to enhance system effectiveness.

3) Apply to real engineering. BCI is a good way to better the life of special person. More application in real world is the direction of BCI.

4) Change the platform for more convenience. The device based on embedded system such as DSP is developed in our lab. The complete test has not been finished now. It is the coming future work.

REFERENCES [1] E. Dochin, K.M.Spencer, and R.Wijensinghe, “The mental prosthesis:

Assessing the speed of a P300-based brain-computer interface,” IEEE Trans. Rehab. Eng., vol. 8, no. 2, pp.174-179, 2000.

[2] J. R.Wolpaw, N. Birbaumer, W.J. Heetderks, D.J. McFarland, P.H. Peckham and G. Schalk, “Brain-computer interface technology: A reviewof the first international meeting,” IEEE Trans. Rehab. Eng., vol. 8, no. 2, pp.164–173, 2000.

[3] Satoru Goto, Masatoshi Nakamura and Takenao Sugi, “Development of meal assistance orthosis for disabled persons using EOG signal and dish image,” International Journal of Advanced Mechatronic Systems, vol. 1, no. 2, pp.107-115, 2008.

[4] Birbaumer N., “Brain-computer-interface research: coming of age,” Clinical Neurophysiology, vol. 117, no. 3, pp. 479-483, 2006.

[5] J.R.Wolpaw, N. Birbaumer, D.J. McFarland, “Brain-computer interfaces for communication and control,” Clinical Neurophysiol. vol. 113, no.6, pp.767-791, 2002.

[6] H.Serby, E.Yom-Tov and G.F.Inbar, “An improved P300-based brain-computer interface,” IEEE Transactions on Neural System and Rehabilitation Engineering, vol.13, no.1, pp. 89-98. 2005.

[7] A. Schlögl, C. Keinrath, D. Zimmermann, R. Scherer, R. Leeb and G. Pfurtscheller, “A fully automated correction method of EOG artifacts in EEG recordings,” Clinical Neurophysiology, vol. 118, no. 1, pp. 98-104, 2007.

[8] National Instruments Corporation, The Measurement and Automation User Manual. Available: www.ni.com, 2005.

[9] National Instruments Corporation, LabVIEW for ECG signal processing. Available: www.ni.com.

[10] National Instruments Corporation, LabVIEW User Manual. Available: www.ni.com.

[11] National Instruments Corporation, The Measurement and Automation Catalog 2006. Available: www.ni.com.

[12] Y. J Wang and R. P. Wang, “A practical VEP-based brain-computer interface,” IEEE Trans. on Neural System and Rehabilitation Engineering, vol. 14, no. 2, pp. 234-239, 2006.

[13] P.J. Durka, “From wavelets to adaptive approximations: time-frequency parametrization of EEG,” Biomed Eng Online, vol. 2, no. 1, 2003.

[14] N. Hazarika, J.Z. Chen, A.C. Tsoi, A. Sergejew, “Classification of EEG Signals using the wavelet transform,” Signal Processing, vol. 59, no. 1, pp. 61-72,1997.

Number Objective Response

P300 Response

Accuracy Rate

Objective Response

P300 Response

Accuracy Rate

3 20 8 40% 20 12 60%

10 20 13 65% 20 14 70%

15 20 14 70% 20 12 60%

20 20 15 75% 20 16 80%

25 20 16 80% 20 17 85%

35 20 16 80% 20 16 80%

121