KR101416616B1 - Word speller based on multi-bio signals - Google Patents

Abstract

A word speller based on multi-bio signals disclosed. The word speller based on multi-bio signals includes an electroencephalogram signal processing unit determining target candidate characters in the manner of selecting one character from each of multiple character matrices based on the electroencephalogram signal of a human being corresponding to the multiple character matrix on which different characters are arranged; a safety signal processing unit detecting a character matrix gazed by the human being among the multiple character matrices based on the safety signal of the human being; and a target character determining unit determining, as a target character, a character belonging to the character matrix detected by the safety signal processing unit among the target candidate characters.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a word-

The present invention relates to a brain-computer interface, and more particularly, to a multi-bio-signal-based word speller.

Yong

Brain computer interface (BCI) technology is a technology that converts brain-conduction signals generated by brain activity into control commands for controlling external devices such as computers, wheelchairs, or virtual spaces. BCI can be used not only to detect user 's intention but also to extract emotional and cognitive status information from EEG and apply it to game to increase immersion.

There are various fields where BCI is used as a communication and control means such as video game, robot arm control, and wheelchair control, but research on word speller, which is the most basic application field, is still actively carried out.

Among the various brain signal measurement methods, the BCI technique using the non-invasive electroencephalogram (EEG) signal can be used as an auxiliary function to the other interface device because the utilization of the BCI technique is relatively higher than that of the other equipment. However, It is not easy to use it as hardware equipment and cumbersome measurement method. However, inexpensive EEG devices such as Emotiv and Neuro Sky are constantly on the market. Depending on the situation, the BCI technique based on P300 or SSVEP (steady state visual evoked potential) It can be used for a limited purpose.

The P300 technique refers to a technique for detecting a user's intention by generating a positive signal in a specific brain region 300 ms after the stimulus is given when the brain is given visual, auditory, or other stimuli . The P300 has the advantage of being able to generate a signal without training in comparison with other techniques and, in some cases, to detect a signal with only a small number of channels.

To detect the P300 signal induced by brain stimulation, it takes a certain number of repetitions and it takes time. As the number of commands to be input increases, the total number of stimuli to be generated increases, and the speed of execution accordingly decreases.

Event related potential (ERP) detection is one of the methods of constructing BCI, and P300 based word speller is a typical application method. The row / column paradigm (RCP) matrix configuration shown in FIG. 1, which is mainly used in the conventional P300-based word speller, causes the stimulus to be generated in units of row and column, The character at the intersection of the row is finally selected.

A number of studies have been carried out on support vector machines, neural networks, Bayesian linear discriminant analysis and multiple classification systems as classification algorithms to improve the performance of the P300 speller. Matrix size, frequency of stimulation, There have been many studies on performance evaluation. Although many performance improvements have been made with these studies, there is a lack of research on solving the performance degradation caused by the increase in the time required for stimulation as the number of characters applied to the speller increases. In other words, conventional word spurs have been difficult to improve detection speed and accuracy.

On the other hand, in the EOG, vertical and horizontal EOG signals are generated depending on the position of a specific object, and studies are being conducted to measure eye movement and motion using the characteristics of the EOG. In previous studies, there was research that the visual evoked potential (VEP) signal becomes clearer when the target character of the P300 speller is looked at. However, when the paralysis of the limb paralysis is impossible and the target character can not be gazed, only the experimental results show that the effect of the P300 speller using VEP is reduced, and no example of the word speller using the EOG signal can be found .

Published patent application No. 10-2012-100320 (Publication date: 2012. 09.12)

A. Nijholt, D. Tan, B. Alison, J. Del R. Milan, B. Graimann, "BrainGain: BCI for HCI and games," Proceedings of the AISB Symposium on Brain Computer Interfaces and Human Computer Interaction: A Convergence of Ideas, Aberdeen, pp. 32-35, 2008.

Embodiments of the present invention relate to a multi-bio-signal based word speller to solve the above problems.

The present embodiments provide a word speller based on multiple bio-signals that can improve the detection speed.

In addition, the present embodiments provide a multi-bio-signal based word speller that can improve accuracy.

The multi-bio-signal-based word speller according to an embodiment of the present invention may be configured to select one character from each of multiple-character matrices based on a human's electroencephalogram signal corresponding to a multi-character matrix in which different characters are arranged, An electroencephalogram signal processing unit for determining candidate characters; A safety signal processor for detecting a character matrix of the multiple character matrix based on the safety signal of the person; And a target character determiner for determining, as a target character, a character belonging to the character matrix detected by the safety degree signal processor among the target candidate characters.

According to embodiments of the present invention, the word speller detects the target character by causing the visual stimuli for multiple matrices to be generated simultaneously, and classifying the P300 signal generated by the electroencephalogram when the visual stimulus for the target character is given. In addition, by using the safety diagram (EOG) used as an auxiliary bio-signal for eliminating the existing noise, the overall stimulation occurrence time (i.e., sensing speed) can be shortened by utilizing the matrix to identify the matrix including the target character have. In other words, the accuracy of the word speller can be improved because it uses an electroencephalogram signal to sense characters in the matrix and a safety signal to identify the matrix.

1 is a diagram showing a character matrix used in a conventional P300-based word speller.
2 is a block diagram illustrating a multi-bio-signal-based word speller in accordance with an embodiment of the present invention.
3 is a diagram illustrating multiple character matrices used in a multi-bio signal based word speller in accordance with an embodiment of the present invention.
Figs. 4A and 4B are diagrams illustrating rows and columns selected in each of the multiple character matrices of Fig. 3; Fig.
FIG. 5 is a diagram illustrating a process in which a word speller based on multiple bio-signals according to an embodiment of the present invention determines a reference matrix based on an EOG signal.
6 is a photograph showing an experimental state of a word speller based on multiple bio-signals according to an embodiment of the present invention for a subject.
7 is a data sheet illustrating a P300 signal generated in a word speller based on multiple bio-signals according to an embodiment of the present invention.
FIG. 8 is a data sheet showing vertical and horizontal EOG signals for a target point on a matrix generated in a word speller based on multiple bio-signals according to an embodiment of the present invention.
FIGS. 9A to 9C are data sheets illustrating the accuracy characteristics of a word speller based on multiple bio-signals and a conventional P300-based word speller according to an embodiment of the present invention.
FIGS. 10A to 10C are data sheets illustrating the transmission speeds of a multi-bio signal based word speller and a conventional P300 based word speller according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will now be described more fully hereinafter with reference to the accompanying drawings, in which exemplary embodiments of the invention are shown. . In the detailed description of the embodiments of the present invention, if it is determined that a detailed description of known configurations or functions related to the present invention can not be applied to the present invention, a detailed description thereof may be omitted.

FIG. 2 is a block diagram illustrating a configuration of a multi-bio-signal-based word speller according to an embodiment of the present invention. Referring to FIG. 3 is a diagram illustrating a multi-character matrix used in a multi-bio-signal-based word speller in accordance with an embodiment of the present invention.

2, the multi-bio-signal based word speller includes an electroencephalogram (EEG) signal processor 10, an electrocardiogram (EOG) signal processor 20, and a target character determiner 30 .

The EEG signal processing unit 10 acquires EEG signals (e.g., P300 signals) from a user (or a subject) and detects target characters of a multi-character matrix from the EEG signals. The EEG signal processing unit 10 includes an EEG signal acquisition unit 12, a preprocessing and filtering unit 14, a target row / column classification unit 16, and a target candidate character selection unit 18.

The multi-character matrix includes at least two character matrices. At least two character matrices are categorized into at least two groups of different characters and numbers and each contain the categorized characters. For example, the multi-character matrix may include first through fourth character matrices MX1 through MX4, as shown in FIG. The first through fourth character matrices MX1 through MX4 include different character groups classified into four groups of alphanumeric characters. The first character matrix MX1 includes alphabetic characters "A to I" arranged in a matrix of 3X3. The second character matrix MX2 includes alphabetic characters "J to R" arranged in a matrix of 3X3. The third character matrix MX3 includes alphabetic characters "S to Z " and a number" 0 " arranged in a matrix of 3X3. The fourth character matrix MX4 includes "1 to 9" numbers arranged in a matrix of 3X3. The first through fourth character matrices are arranged in a matrix of 2x2. In addition, the first to fourth character matrices are spaced apart from each other such that an angular difference of at least 5 [deg.] Is generated at the observation position of the user (or the subject).

The EEG signal acquisition unit 12 detects EEG signals of a plurality of channels from a user (or a subject). For example, the EEG signal acquisition unit 12 includes 14 channels (T7, T8, P7, P8, CP5, CP6, P3, P4, PO3, Emotiv EEG head sets of PO4, F7, F8, FP1, FP2 may be used, but are not limited thereto. In this case, the EEG signals may be P300 signals.

The pre-processing and filtering unit 14 extracts EEG signals of a frequency band of approximately 1 to 20 Hz using a bandpass filter. The measurement timing of the EEG signal and the position of the signal where the visual evoked potential is generated for each user are slightly different. As such, in order to effectively process the EEG signals, the preprocessing and filtering unit 14 additionally uses a spatial filter to increase the signal-to-noise ratio. The spatial filter used in the preprocessing and filtering unit 14 sets a weight for each of the 14 channels based on training data for 14 channels and adjusts 14 channels of EEG signals according to a weight value set for each channel do.

The target row / column classifier 16 determines a target column and a row on each character matrix based on the EEG signals supplied from the preprocessing and filtering unit 14. In order to determine a target column and a row, the target row / column classifier 16 may use a FLDA (Fisher Linear Discriminant Analysis) technique, but the present invention is not limited thereto. The target row / column classifying unit 16 employing the FLDA technique calculates a variance value between classes and a variance value between classes for the target class and the non-target class based on the training data obtained in the training step, Values are used to calculate a linear transformation vector that maximizes the variance ratio between classes for variance within each class. In order to obtain the training data, the target row / column sorting unit 16 samples each of 14 channels of EEG signals from the preprocessing and filtering unit 14 in units of 128 times per second and generates a stimulus The 78 amplitude values extracted for the following 600 ms period are provided as training data as basic feature values. Thereafter, when 14 channels of EEG signals are input from the pre-processing and filtering unit 14, the target row / column sorting unit 16 reduces the dimension of each of the EEG signals using the linear transformation vector Target rows and columns on each character matrix MX by comparing the target and non-target classes for each of the rows / columns of each character matrix. In order to increase the reliability, the target row / column classifier 16 may repeat the dimension reduction and comparison process several times (for example, 12 times). In this case, the target row / column classifier 16 determines the rows and columns classified as the target class as the target row and column most. In this manner, the target row / column classifier 16 determines a pair of target columns and rows for each of the first through fourth character matrices MX1 through MX4. In other words, by the target row / column classifier 16, four pairs of target rows and columns corresponding to the first through fourth character matrices are determined. For example, as shown in FIG. 4A, the third columns in the first through fourth character matrices MX1 through MX3 may be selected by the target row / column sorting unit 16. In addition, as shown in FIG. 4B, the third columns in the first through fourth character matrices MX1 through MX4 can be selected by the target row / column sorting unit 16.

The target candidate character selector 18 selects target candidate characters of the first through fourth character matrices MX1 through MX4 based on the four pairs of target rows and columns supplied from the target row / . In other words, the target candidate character selector 18 selects an alphabet letter or a number located at a point where the target row and the target column of each pair intersect with each other as a target candidate character. For example, one of the alphabetical characters "A to I" may be selected as the target candidate character by the target row and target column for the first character matrix MX1, and the second character matrix MX2 One of the alphabetical characters of "J to R" may be selected as the target candidate character by the target row and the target column for the third character matrix MX3, and the target row and the target column for the third character matrix MX3 may be & "And the number of" 0 "may be selected as the target candidate character, and the target row and target column for the fourth character matrix MX4 may be any one of" 1 to 9 " It can be selected as a target candidate character. In this way, the target candidate character selector 18 can determine a target candidate character group including four target candidate characters based on four pairs of target rows and columns from the target row / column classifier 16. [ For example, based on FIGS. 4A and 4B, the target candidate character selector 28 selects the target letter character string "I" of the first character matrix MX1, "R" of the second character matrix MX2, Quot; 0 "of the fourth character matrix MX3 and" 9 "of the fourth character matrix MX4 as target candidate characters.

The EOG signal processing unit 20 obtains the horizontal and vertical EOG signals from the user (or the subject), and generates the EOG signals based on the EOG signals of the multiple character matrices (i.e., the first to fourth character matrices MX1 to MX4) The test character matrix is detected. The EOG signal processing unit 20 includes an EOG signal acquisition unit 22, an EOG signal correction unit 24, an eye position detection unit 26, and a gaze matrix detection unit 28.

The EOG signal acquisition unit 22 acquires horizontal and vertical EOG signals from a user (or a subject). To this end, the EOG signal acquisition unit 22 may include an EOG measurement device manufactured by BioPac, but is not limited thereto. The horizontal and vertical EOG signals obtained by the EOG signal acquisition unit 22 are supplied to the EOG signal correction unit 24. [

The EOG signal correcting unit 24 corrects the horizontal and vertical EOG signals from the EOG signal acquiring unit 22. The horizontal and vertical EOG signals corrected by the EOG signal correcting unit 24 are supplied to the eye position detecting unit 26.

The eye position detecting unit 26 detects the eye position of the user (or the subject) based on the corrected horizontal and vertical EOG signals from the EOG signal correcting unit 24. In detail, the eye position detecting unit 26 may detect where the eyeball stays on the right and left by comparing the corrected horizontal EOG signal with the horizontal EOG reference value Vhor_ref. In addition, the eyeball position detecting unit 26 may compare the corrected vertical EOG signal with the vertical EOG reference value Vver_ref to detect where the eyeball stays on the upper and lower sides. The result of the comparison between the corrected horizontal EOG signal and the horizontal EOG reference value Vhor_ref and the result of the comparison between the corrected vertical EOG signal and the vertical EOG reference value Vver_ref are input to the gaze matrix detector 28 as eye position detection signals, .

Alternatively, the eye-position detecting unit 26 may calculate an average value of the corrected horizontal and vertical EOG signals during a predetermined period, and calculate the calculated horizontal and vertical EOG average values with the horizontal and vertical EOG reference values Vhor_ref and Vver_ref It is possible to detect the position of the eyeball more accurately. The average value of the corrected horizontal EOG signal may be obtained by sampling the corrected horizontal EOG signal several times at predetermined time intervals and averaging the sampled horizontal EOG values. The average value of the corrected vertical EOG signal may be obtained by sampling the corrected vertical EOG signal several times at predetermined time intervals and averaging the sampled vertical EOG values. The average values of the horizontal and vertical EOG signals can be calculated as shown in Equations (1) and (2).

[Equation 1]

)/TVhor_t = (

) / T

In Equation 1, "Vhor_t" is the average value of the horizontal EOG signal, "Vhor_tk" is the kth sampling value of the horizontal EOG signal for the target character, and "T" is the sampling number.

&Quot; (2) "

)/TVver_t = (

) / T

In Equation 2, "Vver_t" is the average value of the vertical EOG signal, "Vver_tk" is the kth sampling value of the vertical EOG signal for the target character, and "T" is the sampling number.

As shown in Fig. 5, the gaze matrix detecting unit 28 detects the gaze position of the first to fourth character matrices MX1 (MX1) based on the eye position detection signal (i.e., comparison results) from the eye position detector 26 To MX4 of the character matrix MX that the user is looking at. More specifically, if the average value Vver_t of the corrected vertical EOG signal is smaller than the vertical EOG reference value Vver_ref (first condition Cond1) and the average value Vhor_t of the corrected horizontal EOG signal is greater than the horizontal EOG reference value (The third condition (Cond3)), the entry matrix detection unit 28 detects that the character matrix being looked at by the user is the first character matrix MX1. Wherein the corrected vertical EOG signal Vver_t is smaller than the vertical EOG reference value Vver_ref (first condition Cond1) and the average value Vhor_t of the corrected horizontal EOG signal is greater than the horizontal EOG reference value Vhor_ref (Fourth condition (Cond4)), the test matrix detection unit 28 detects that the character matrix being looked at by the user is the second character matrix MX2. Wherein the corrected vertical EOG signal Vver_t is greater than the vertical EOG reference value Vver_ref and the corrected horizontal EOG signal Vhor_t is less than the horizontal EOG reference value Vhor_ref, (Third condition (Cond3)), the gaze matrix detection unit 28 detects that the character matrix that the user is looking at is the third character matrix MX3. Finally, when the average value Vver_t of the corrected vertical EOG signal is greater than the vertical EOG reference value Vver_ref (the second condition Cond2) and the average value Vhor_t of the corrected horizontal EOG signal is greater than the horizontal EOG reference value Vhor_ref (The fourth condition (Cond4)), the test matrix detection unit 28 detects that the character matrix being looked at by the user is the fourth character matrix MX4. Further, the gazette matrix detector 28 supplies the target character determiner 30 with a gazette matrix detection signal indicating the detected character matrix.

The target character determination unit 30 determines whether or not the candidate character group (i.e., four candidate characters) from the target candidate character selector 18 based on the candidate matrix detection signal detected by the candidate matrix detection unit 28, As a target character. For example, when the target matrix detection signal indicates the first character matrix MX1, the target character determination unit 30 determines whether or not the four target candidate characters (e.g., "I", "R" (For example, "I") belonging to the first character matrix MX1 among the character strings "0" As another example, when the target matrix detection signal indicates the second character matrix MX2, the target character determination unit 30 determines the target candidate character (e.g., "I", "R" (E.g., "R") belonging to the second character matrix MX2 out of the character strings "0" As another example, when the target matrix detection signal indicates the third character matrix MX3, the target character determination unit 30 determines the four target candidate characters (for example, "I", "R" (For example, "0") belonging to the third character matrix MX3 among the first character matrix MX3, "0" As another example, when the target matrix detection signal indicates the fourth character matrix MX4, the target character determination unit 30 determines the four target candidate characters (e.g., "I", "R" (E.g., "9") belonging to the fourth character matrix MX4 among the character strings MX4, 0, and 9 is determined as the target character. The target character determined by the target character determining unit 30 is transmitted to the computer system as a character (or number) desired to be input by the user (or the subject).

Experiments have been performed on a multiple-body based word speller according to an embodiment of the present invention. In the experiment, if we assume that one block has occurred once for the whole row and column, the number of iterations of the block, that is, the number of repetitions, is changed to 6, 9, and 12, respectively. The flash generation is maintained for 0.1 second and the flash interval is set to 0.1 second. The repeat interval was 4 seconds apart. For each experiment, more than 10 runs were performed and the average of the results was obtained and the performance was measured. Whenever a block is executed, two P300 signals are generated when the target character belongs to row and column, and the non-p300 signal is processed for the remaining stimuli. In each experiment, a total of 10 target characters were presented for each run and the accuracy was compared with the classification result. In addition, the subjects were restrained by their body movements to minimize EMG noise, and the tests were limited to twice daily to eliminate noise caused by long-time measurement.

FIG. 6 is an experimental photograph with a 2-channel EOG electrode (horizontal EOG, vertical EOG) attached. FIG. 7 shows an average value of P300 signals measured 12 times for a signal generated in the P3 channel. FIG. 8 shows the horizontal EOG and vertical EOG signal results generated for each of the four gazing points of each character matrix.

The difference between the intensity of the vertical EOG signals of the first and fourth character matrices MX1 and MX4 and the second and third character matrices MX2 and MX3 is about 0.5 volt and the difference between the first and second character matrices MX2 and MX3 MX1 and MX2 and the third and fourth character matrices MX3 and MX4 are almost the same. The difference between the horizontal EOG values of the first and second character matrices MX1 and MX2 and the third and fourth character matrices MX3 and MX4 is about 0.5 volt, and the first and fourth character matrices MX1, MX4) and the horizontal EOG values of the second and third character matrices (MX2, MX3) hardly occurs. Since the character matrices MX are sufficiently spaced to generate a gaze angle difference of at least 5 degrees so that the difference between the EOG signals is sufficiently generated, the character matrix that the subject has stood at can be determined with an accuracy of 93% or more.

FIGS. 9A to 9C illustrate accuracy test results of a conventional P300-based word speller and a multi-bio signal based word speller according to an embodiment of the present invention. As can be seen from Figs. 9A to 9C, it is clear that the accuracy of the word speller according to the embodiment of the present invention is superior to that of the conventional word speller for all the number of repetitions of stimulation. The difference in performance between the conventional P300-based word speller and the word speller of the present embodiment can be verified by a variance analysis of the results of 15 subjects. The experiment was repeated 6 times (F (1,28) = 44.8, p <0.0002), 9 times (F (1,28) = 17.531, p = 0.00025) 28) = 24.15, p < 0.0002), the word spellar of this embodiment shows significant performance differences. In the case of six repetition times, the accuracy of the word speller of this embodiment was improved by an average of 17% for 15 subjects and up to 33% for the subjects.

FIGS. 10A to 10C are experimental results in which the transmission rate of the conventional P300-based word speller is compared with the transmission rate of the word speller of the present embodiment, and the transmission rate at the time of one transmission attempt can be expressed by Equation (3).

&Quot; (3) &quot;

TR = log ₂ N + p log ₂ p + (1-p) log 2 ((1-p) / (N-1))

In Equation (3), "N" is the number of classes and "p" is the accuracy.

The transmission rate may be calculated in units of bits / min by inputting the number of transmission attempts per minute in Equation (3). As shown in Figs. 10A to 10C, it is apparent that the transfer rate of the word speller of this embodiment is superior to that of the conventional word speller. Especially, when the number of stimulus repetition is 6, the difference of transmission rate is the largest. This indicates that the accuracy of the conventional word speller decreases sharply as the number of iterations decreases, and thus the overall transmission speed decreases even if the execution time is shortened. On the other hand, according to the embodiment of the present invention, even if the number of repetitions is decreased, the accuracy of the word speller based on the multi-bio signal can be reduced so that the execution time can be shortened and the overall transmission speed can be increased.

As described above, according to the embodiment of the present invention, the multi-bio-signal-based word speller divides the matrix size into a plurality of matrix groups so that visual stimuli are simultaneously generated in the matrix group. In order to distinguish the visual stimulation signals generated simultaneously in a plurality of matrices, a specific matrix is determined by measuring an EOG signal generated according to the eyeball position of the user. As a result, the overall stimulation time was reduced and the transmission rate increased 2 to 5 times. Also, in the experimental results, the accuracy of the word speller according to the embodiment of the present invention was improved by 17% on average when the stimulus repetition was performed six times.

The embodiments described above are merely examples for implementing a word speller based on multiple bio-signals according to the present invention. It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit and scope of the invention as defined in the appended claims. And should be protected to include to the extent practicable.

10; An EEG signal processing unit 12; EEG signal acquisition unit
14; A preprocessing and filtering unit 16; Target row / column classifier
18; A target candidate character selector 20; EOG signal processing section
22; An EOG signal acquisition unit 24; EOG signal correction unit
26; An eye position detecting unit 28; [0033]
30; The target-

Claims (6)

An electroencephalogram signal processing unit for determining target candidate characters by selecting one character from each of the multiple character matrices based on a human's electroencephalogram signal corresponding to a multi-character matrix in which different characters are arranged;
A safety signal processor for detecting a character matrix of the multiple character matrix based on the safety signals of the person; And
And a target character determination unit for determining, as a target character, a character belonging to the character matrix detected by the safety degree signal processing unit among the target candidate characters,
The electroencephalogram signal processing unit
An electroencephalogram signal acquisition unit for acquiring an electroencephalogram signal of the person;
A preprocessing and filtering unit for preprocessing and filtering an electroencephalogram signal from the electroencephalogram signal acquisition unit;
A target row / column classifier for classifying a target row and a target column for each of the multiple character matrices based on the electroencephalogram signal from the preprocessing and filtering unit;
Selecting the target candidate characters by selecting one character from each of the multiple character matrices based on the target rows and target columns classified by the target row / column classifier, and supplying the selected target candidate characters to the target character determiner Wherein the target candidate character selector comprises a target candidate character selector. 2. The multiple-body based word spell-checker of claim 1, wherein the multi-character matrix comprises at least two character matrices spaced apart such that an angular difference of at least 5 degrees can be generated at a human gaze point. delete 2. The apparatus of claim 1, wherein the preprocessing and filtering unit
A bandpass filter for extracting only a component of a specific frequency band from the electroencephalogram signal from the electroencephalogram signal acquisition unit; And
And a spatial filter for increasing a signal-to-noise ratio of the electroencephalogram signal from the band-pass filter. The apparatus as claimed in claim 1, wherein the safety signal processor
A safety degree signal acquisition unit for acquiring the safety degree signals from the person;
An eye position detecting unit for detecting the eye position of the person by comparing the safety degree signals from the safety degree signal acquisition unit with safety degree reference values; And
And a stray matrix detector for detecting a character matrix being examined among the multiple character matrices based on the eye position detected by the eye position detecting unit and for providing information on the detected character matrix to the target character determining unit A plurality of bio-signal based word spellers. [Claim 6] The system according to claim 5, wherein the safety signal processor includes a safety signal corrector for correcting the safety signals supplied from the safety signal acquisition unit to the safety position detector. Speller.

Images