An eog-based smart helmet control system and method

Abstract

The application discloses an EOG-based intelligent helmet control system and method, which comprises a system software module and an external interaction device; the system software module comprises an EOG acquisition circuit, a communication module, a controller and an API interface, an EOG interaction module and a head-mounted display; the EOG acquisition circuit is used for acquiring multiple physiological signals; the controller and the API interface are used for the cooperative control and management of multiple modules, and realize the instruction conversion of the interface software of the EOG interaction module and the bottom terminal system control; the communication module is used for communication with the external interaction device; the EOG-based intelligent helmet control system and method can control the interaction interface in the EOG type bioelectricity mode under the condition that the hands of the staff are kept in the working state, and realizes the non-hand-operated human-computer interaction; the application is one of the core components of the intelligent construction site, and has a wide application prospect.

Description

Technical Field

[0001] This invention belongs to the field of smart helmet control technology, specifically relating to a smart helmet control system and method based on EOG. Background Technology

[0002] Safety helmets used in construction sites, fire stations, and other emergency situations need to ensure the efficiency and safety of workers. To address this, we propose an EOG-based smart helmet control method. Summary of the Invention

[0003] The purpose of this invention is to provide an EOG-based smart helmet control system and method to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an EOG-based intelligent helmet control system, comprising a system software module and an external interaction device; the system software module includes an EOG acquisition circuit, a communication module, a controller and API interface, an EOG interaction module, and a head-mounted display;

[0005] The EOG acquisition circuit is used to acquire multiple physiological signals;

[0006] The controller and API interface are used for the coordinated control and management of multiple modules in the device, and realize the conversion of interface software with the EOG interaction module and control instructions of the underlying terminal system.

[0007] The communication module is used to communicate with external interactive devices;

[0008] Head-mounted displays enable, but are not limited to, flat panel displays, virtual reality, augmented reality, mixed reality, holographic projection, or glasses-free 3D.

[0009] The external interaction device and the smart helmet exchange information via wired or wireless communication.

[0010] Preferably, the EOG acquisition circuit includes electrodes, a high-performance differential amplifier circuit, a filter circuit, a signal amplification and processing circuit, and a right leg drive circuit.

[0011] Preferably, the interface software for implementing the interaction with the EOG module and the instruction conversion between the underlying terminal system control include module control, specifically including device management and data transmission; device management includes device connection status detection, device access, and device deletion; data transmission is used for the controller to send control instructions to the EOG interaction module, which then controls the device to perform actions upon receiving the instructions.

[0012] Preferably, the controller completes specific control in three threads: Device connection status detection thread: First, a connection request is sent. If the EOG interaction module receives the request, it first determines whether it is already in a connected state. If so, it returns ACK; otherwise, it returns an EOG interaction module access request and includes device information in the request data packet. If the controller does not receive an ACK signal, it means that the EOG interaction module has lost connection, and the controller will perform an EOG interaction module disconnection operation.

[0013] The controller's EOG interaction module data receiving thread: When the EOG interaction module sends an access request, the controller will execute the relevant operations for the EOG interaction module to access; when the EOG interaction module completes its task and needs to disconnect, it can send a disconnect request to the controller, and the controller will execute the disconnect operation after receiving the request.

[0014] The controller's control command thread: Upon receiving a control command, the controller sends the command to the corresponding EOG interaction module based on the device type in the control command. The EOG interaction module has only one thread for receiving requests from the controller. When the EOG interaction module receives a status detection request, if the EOG interaction module is not yet connected to the controller, it sends a connection request; otherwise, it sends an ACK. When the EOG interaction module receives a control command, it executes the corresponding action.

[0015] Preferably, multiple electrodes are installed at the front end of the inner adjustment strap of the smart helmet via a locking component. The length of the inner adjustment strap can be adjusted to ensure that the forehead electrodes maintain complete contact with the human forehead and to ensure the quality of bioelectrical signal transmission.

[0016] A smart helmet method based on EOG, step 1: The user wears the smart helmet and turns on the EOG interaction module 4 to enter the interaction interface when the EOG acquisition circuit is working normally;

[0017] Step 2: In the interactive interface, parameters will be set for the interactive program. If multiple graphical user interfaces (GUIs) are required, the ID number of the currently displayed GUI will be sent back to the client program to provide a reference for the client's signal recognition. Different function options will be generated for the user to choose from in the GUI. In the current GUI, each option corresponds to a virtual object or button. The user completes the corresponding operation by selecting the desired virtual object or button. n virtual objects or buttons will be generated, each with a unique numerical label. These n virtual objects or buttons correspond one-to-one with the n signals from the acquisition end.

[0018] Step 3: Before the system starts running, a training operation is performed. The user focuses on the flashing target on the training interface and raises their eyebrows while the target is flashing. The signal recognition and processing process during training is as follows:

[0019] Step 31: The acquisition circuit synchronously acquires human bioelectric signals, performs bandpass filtering on the acquired data, extracts sampling points of a certain time length from the data, and downsamples the sampling points.

[0020] Step 32: Construct a feature vector from the downsampled data, and store the feature vector corresponding to each time the target flashes in the computer;

[0021] Step 33: Using the result obtained after averaging the feature vector as a threshold, establish a training model for the user to provide a personal model for subsequent recognition operations;

[0022] Step 4: Next, the user performs the GUI operation: In the current GUI, each virtual object or button flashes once in a certain order or randomly. While the virtual object or button that the user needs to select is flashing, the user selects the target by raising their eyebrows.

[0023] Step 5: The target to be selected in Step 4 is chosen by raising the eyebrows. The specific target recognition process is as follows:

[0024] Step 51: The EOG acquisition circuit takes a signal from the signal acquisition terminal corresponding to the digital mark as a feature vector. After all n buttons have flashed, n feature vectors are generated. These n feature vectors form a feature vector set Di.

[0025] Step 52: After preprocessing the feature vectors in the feature vector set Di, the feature vectors are then judged by a classifier; let L in this method represent the number of times the average of the current classification is calculated, that is, the number of rounds. min and L max Let θ and θ0 represent the minimum and maximum number of rounds required to identify a button, respectively; θ0 is the threshold controlling the output; then the specific judgment method is as follows:

[0026] (1) System initialization, set L = 0;

[0027] (2) Receive the new round of data and update L = L + 1;

[0028] (3) In the current round, corresponding to the number of targets in the GUI, feature extraction is performed to obtain n feature vectors;

[0029] (4) For each specific button, perform an average operation on the cumulative round feature vectors of the first L;

[0030] (5) Use a classifier to calculate the regression score for each button, denoted as RS. i RS i ={RS1, RS2, ... RS n}; Then, these RS i The scores are normalized, and the resulting n target regression scores are all in the interval [0, 1].

[0031] (6) If L ≥ L min And L < L max If yes, proceed to step (7); otherwise, proceed to step (9).

[0032] (7) Calculate RS max and RS sec The difference: Δθ = RS max -RS sec In the formula, RS max Represents RS i The maximum value among the nine regression scores, and S sec Represents the second largest value;

[0033] (8) If Δθ > θ0, proceed to step (9); otherwise, proceed to step (2).

[0034] (9) If L≥ L max Immediately jump to step (10);

[0035] (10) End the recognition of the current trial and output the recognition result (maximum score RS). max (The corresponding button); At the same time, jump to step 1 to prepare for the recognition of the next trial;

[0036] Step 53: Based on the method in step 52, determine whether the target feature vector exists in the feature vector set Di. If the target feature vector exists, proceed to the next step; otherwise, proceed to step 4.

[0037] Step 54: Repeat step 5 at least once to obtain feature vector set Di+1. Determine whether the feature vectors in feature vector set Di and feature vector set Di+1 are the same. If they are the same, determine the selection key according to the target feature vectors in feature vector set Di and feature vector set Di+1, and proceed to the next step. Otherwise, return to step 4.

[0038] Step 6: Determine if the current selection is the end output. If yes, proceed to step 10; otherwise, proceed to the next step.

[0039] Step 7: Determine if the input is complete. If the input is complete, proceed to step 53; otherwise, proceed to the next step.

[0040] Step 8: Proceed to the next work row and repeat steps 4 and 5;

[0041] Step 9: Then proceed according to the entered selections;

[0042] Step 10: End selection.

[0043] Preferably, the classification method adopted in step 52 is as follows:

[0044] Suppose that the target y is linearly correlated with the feature vector X, and Z is Gaussian noise, then the linear classifier takes the following form:

[0045]

[0046] set up The training set consists of feature vectors. and the corresponding binary state Composition. From (4-1), we can obtain the weights. Likelihood function:

[0047]

[0048] In the formula, Y represents a vector containing all training targets. This represents the matrix obtained by horizontally stacking the training feature vectors. Let L represent the inverse variance of the noise, and L represent the number of training samples.

[0049] In Bayesian linear discriminant analysis (BLDA), by extending... The dimension of, i.e. The last term is the bias term, which incorporates the Gaussian noise Z into the weights. In the middle, assuming weights The prior distribution is:

[0050]

[0051] To represent the precision of this distribution, where It is expanded to n+1 dimensions, and the last term is the bias term. It is an n+1 dimensional diagonal matrix, in the form of:

[0052]

[0053] Based on the likelihood function and the prior distribution, the posterior distribution can be calculated using Bayes' rule as follows:

[0054]

[0055] Since both the prior and likelihood distributions are Gaussian, the posterior distribution also has a Gaussian form. Its distribution can be determined by the mean (m) of w and the covariance (C), as shown below:

[0056]

[0057]

[0058] Ultimately, for To obtain its predicted regression value Distribution function:

[0059]

[0060] in This represents a normal distribution. Predict the mean of the distribution function. and variance for:

[0061]

[0062]

[0063] In which the mean This is the regression target value we need.

[0064] Compared with the prior art, the beneficial effects of the present invention are: the EOG-based intelligent helmet control system and method proposed in the present invention can control the interactive interface using EOG-type bioelectric methods while the worker's hands are in a working state, realizing non-manual human-computer interaction; the present invention is one of the core components of intelligent construction sites and has broad application prospects. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the system of the present invention;

[0066] Figure 2 This is a schematic diagram of the EOG acquisition circuit of the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Please see Figures 1 to 2The present invention provides a technical solution: an EOG-based intelligent helmet control system, comprising a system software module and an external interaction device; the system software module includes an EOG acquisition circuit, a communication module, a controller and API interface, an EOG interaction module, and a head-mounted display;

[0069] The EOG acquisition circuit is used to acquire multiple physiological signals;

[0070] The controller and API interface are used for the coordinated control and management of multiple modules in the device, and realize the conversion of interface software with the EOG interaction module and control instructions of the underlying terminal system.

[0071] The communication module is used to communicate with external interactive devices;

[0072] Head-mounted displays enable, but are not limited to, flat panel displays, virtual reality, augmented reality, mixed reality, holographic projection, or glasses-free 3D.

[0073] The external interaction device and the smart helmet exchange information via wired or wireless communication.

[0074] In this embodiment, preferably, the EOG acquisition circuit includes electrodes, a high-performance differential amplifier circuit, a filter circuit, a signal amplification and processing circuit, and a right leg drive circuit.

[0075] In this embodiment, preferably, the instruction conversion between the interface software for implementing the interaction with the EOG module and the control of the underlying terminal system includes module control, specifically including device management and data transmission; device management includes device connection status detection, device access, and device deletion; data transmission is used for the controller to send control instructions to the EOG interaction module, which then controls the device to perform actions upon receiving the instructions.

[0076] In this embodiment, preferably, the controller completes the specific control in three threads: Device connection status detection thread: First, a connection request is sent. If the EOG interaction module receives the request, it first determines whether it is already in a connected state. If so, it returns ACK; otherwise, it returns an EOG interaction module access request and includes device information in the request data packet. If the controller does not receive an ACK signal, it means that the EOG interaction module has lost connection, and the controller will perform an EOG interaction module disconnection operation.

[0077] The controller's EOG interaction module data receiving thread: When the EOG interaction module sends an access request, the controller will execute the relevant operations for the EOG interaction module to access; when the EOG interaction module completes its task and needs to disconnect, it can send a disconnect request to the controller, and the controller will execute the disconnect operation after receiving the request.

[0078] The controller's control command thread: Upon receiving a control command, the controller sends the command to the corresponding EOG interaction module based on the device type in the control command. The EOG interaction module has only one thread for receiving requests from the controller. When the EOG interaction module receives a status detection request, if the EOG interaction module is not yet connected to the controller, it sends a connection request; otherwise, it sends an ACK. When the EOG interaction module receives a control command, it executes the corresponding action.

[0079] In this embodiment, preferably, multiple electrodes are installed at the front end of the inner adjustment strap of the smart helmet by means of a fastening component. The length of the inner adjustment strap can be adjusted to ensure that the forehead electrodes maintain complete contact with the human forehead and to ensure the quality of bioelectrical signal transmission by the electrodes.

[0080] A smart helmet method based on EOG, step 1: The user wears the smart helmet and turns on the EOG interaction module 4 to enter the interaction interface when the EOG acquisition circuit is working normally;

[0081] Step 2: In the interactive interface, the parameters of the interactive program will be set: If multiple user graphical interfaces (GUIs) need to be displayed, the ID number of the currently displayed GUI will be sent back to the client program to provide a reference for the client's signal recognition; different function options will be generated for the user to choose from in the GUI; in the current GUI, each option corresponds to a virtual object or button, and the user can complete the corresponding operation by selecting the required virtual object or button; n virtual objects or buttons will be generated, each with a unique numerical label, and the n virtual objects or buttons will correspond one-to-one with the n signals from the acquisition end;

[0082] Step 3: Before the system starts running, a training operation is performed. The user focuses on the flashing target on the training interface and raises their eyebrows while the target is flashing. The signal recognition and processing process during training is as follows:

[0083] Step 31: The acquisition circuit synchronously acquires human bioelectric signals, performs bandpass filtering on the acquired data, extracts sampling points of a certain time length from the data, and downsamples the sampling points.

[0084] Step 32: Construct a feature vector from the downsampled data, and store the feature vector corresponding to each time the target flashes in the computer;

[0085] Step 33: Using the result obtained after averaging the feature vector as a threshold, establish a training model for the user to provide a personal model for subsequent recognition operations;

[0086] Step 4: Next, the user performs the GUI operation: In the current GUI, each virtual object or button flashes once in a certain order or randomly. While the virtual object or button that the user needs to select is flashing, the user selects the target by raising their eyebrows.

[0087] Step 5: The target to be selected in Step 4 is chosen by raising the eyebrows. The specific target recognition process is as follows:

[0088] Step 51: The EOG acquisition circuit takes a signal from the signal acquisition terminal corresponding to the digital mark as a feature vector. After all n buttons have flashed, n feature vectors are generated. These n feature vectors form a feature vector set Di.

[0089] Step 52: After preprocessing the feature vectors in the feature vector set Di, the feature vectors are then judged by a classifier; let L in this method represent the number of times the average of the current classification is calculated, that is, the number of rounds. min and L max Let θ and θ0 represent the minimum and maximum number of rounds required to identify a button, respectively; θ0 is the threshold controlling the output; then the specific judgment method is as follows:

[0090] (1) System initialization, set L = 0;

[0091] (2) Receive the new round of data and update L = L + 1;

[0092] (3) In the current round, corresponding to the number of targets in the GUI, feature extraction is performed to obtain n feature vectors;

[0093] (4) For each specific button, perform an average operation on the cumulative round feature vectors of the first L;

[0094] (5) Use a classifier to calculate the regression score for each button, denoted as RS. i RS i ={RS1, RS2, ... RS n}; Then, these RS i The scores are normalized, and the resulting n target regression scores are all in the interval [0, 1].

[0095] (6) If L ≥ L min And L < L max If yes, proceed to step (7); otherwise, proceed to step (9).

[0096] (7) Calculate RS max and RS sec The difference: Δθ = RSmax -RS sec In the formula, RS max Represents RS i The maximum value among the nine regression scores, and S sec Represents the second largest value;

[0097] (8) If Δθ > θ0, proceed to step (9); otherwise, proceed to step (2).

[0098] (9) If L≥ L max Immediately jump to step (10);

[0099] (10) End the recognition of the current trial and output the recognition result (maximum score RS). max (The corresponding button); At the same time, jump to step 1 to prepare for the recognition of the next trial;

[0100] Step 53: Based on the method in step 52, determine whether the target feature vector exists in the feature vector set Di. If the target feature vector exists, proceed to the next step; otherwise, proceed to step 4.

[0101] Step 54: Repeat step 5 at least once to obtain feature vector set Di+1. Determine whether the feature vectors in feature vector set Di and feature vector set Di+1 are the same. If they are the same, determine the selection key according to the target feature vectors in feature vector set Di and feature vector set Di+1, and proceed to the next step. Otherwise, return to step 4.

[0102] Step 6: Determine if the current selection is the end output. If yes, proceed to step 10; otherwise, proceed to the next step.

[0103] Step 7: Determine if the input is complete. If the input is complete, proceed to step 53; otherwise, proceed to the next step.

[0104] Step 8: Proceed to the next work row and repeat steps 4 and 5;

[0105] Step 9: Then proceed according to the entered selections;

[0106] Step 10: End selection.

[0107] In this embodiment, preferably, the classification method adopted in step 52 is:

[0108] Suppose that the target y is linearly correlated with the feature vector X, and Z is Gaussian noise, then the linear classifier takes the following form:

[0109]

[0110] set up The training set consists of feature vectors. and the corresponding binary state Composition. From (4-1), we can obtain the weights. Likelihood function:

[0111]

[0112] In the formula, Y represents a vector containing all training targets. This represents the matrix obtained by horizontally stacking the training feature vectors. Let L represent the inverse variance of the noise, and L represent the number of training samples.

[0113] In Bayesian linear discriminant analysis (BLDA), by extending... The dimension of, i.e. The last term is the bias term, which incorporates the Gaussian noise Z into the weights. In the middle, assuming weights The prior distribution is:

[0114]

[0115] To represent the precision of this distribution, where It is expanded to n+1 dimensions, and the last term is the bias term. It is an n+1 dimensional diagonal matrix, in the form of:

[0116]

[0117] Based on the likelihood function and the prior distribution, the posterior distribution can be calculated using Bayes' rule as follows:

[0118]

[0119] Since both the prior and likelihood distributions are Gaussian, the posterior distribution also has a Gaussian form. Its distribution can be determined by the mean (m) of w and the covariance (C), as shown below:

[0120]

[0121]

[0122] Ultimately, for To obtain its predicted regression value Distribution function:

[0123]

[0124] in This represents a normal distribution. Predict the mean of the distribution function. and variance for:

[0125]

[0126]

[0127] In which the mean This is the regression target value we need.

[0128] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An EOG-based intelligent helmet control system, characterized in that: It includes system software modules and external interaction devices; the system software modules include EOG acquisition circuit, communication module, controller and API interface, EOG interaction module, and head-mounted display; The EOG acquisition circuit is used to acquire multiple physiological signals; The controller and API interface are used for the coordinated control and management of multiple modules in the device, and realize the conversion of interface software with the EOG interaction module and control instructions of the underlying terminal system. The communication module is used to communicate with external interactive devices; Head-mounted displays enable, but are not limited to, flat panel displays, virtual reality, augmented reality, mixed reality, holographic projection, or glasses-free 3D. External interaction devices and smart helmets exchange information via wired or wireless communication. The EOG acquisition circuit includes electrodes, a high-performance differential amplifier circuit, a filter circuit, a signal amplification and processing circuit, and a right leg drive circuit. The interface software for interacting with the EOG interaction module and the instruction conversion between the underlying terminal system control include module control, specifically device management and data transmission. Device management includes device connection status detection, device access, and device deletion. Data transmission is used by the controller to send control commands to the EOG interaction module, which then controls the device to perform actions upon receiving the commands. The controller performs specific control in three threads: Device connection status detection thread: First, a connection request is sent. If the EOG interaction module receives the request, it first determines whether it is already in a connected state. If so, it returns ACK; otherwise, it returns an EOG interaction module access request and includes device information in the request data packet. If the controller does not receive an ACK signal, it means that the EOG interaction module has lost connection, and the controller will perform an EOG interaction module disconnection operation. The controller's EOG interaction module data receiving thread: When the EOG interaction module sends an access request, the controller will execute the relevant operations for the EOG interaction module to access; when the EOG interaction module completes its task and needs to disconnect, it can send a disconnect request to the controller, and the controller will execute the disconnect operation after receiving the request. The controller's control command thread: Upon receiving a control command, the controller sends the command to the corresponding EOG interaction module based on the device type in the control command. The EOG interaction module has only one thread for receiving requests from the controller. When the EOG interaction module receives a status detection request, if the EOG interaction module is not yet connected to the controller, it sends a connection request; otherwise, it sends an ACK. When the EOG interaction module receives a control command, it executes the corresponding action. Multiple electrodes are installed at the front end of the inner adjustment strap of the smart helmet via a locking component. The length of the inner adjustment strap can be adjusted to ensure that the forehead electrodes maintain complete contact with the human forehead and to ensure the quality of bioelectrical signal transmission.

2. A smart helmet control method based on EOG, characterized in that: Using the smart helmet control system described in claim 1, step 1: The user wears the smart helmet and, under the normal working condition of the EOG acquisition circuit, opens the EOG interaction module 4 to enter the interaction interface. Step 2: In the interactive interface, the parameters of the interactive program will be set: If multiple user graphical interfaces (GUIs) need to be displayed, the ID number of the currently displayed GUI will be sent back to the client program to provide a reference for the client's signal recognition; different function options will be generated for the user to choose from in the GUI; in the current GUI, each option corresponds to a virtual object or button, and the user can complete the corresponding operation by selecting the required virtual object or button; n virtual objects or buttons will be generated, each with a unique numerical label, and the n virtual objects or buttons will correspond one-to-one with the n signals from the acquisition end; Step 3: Before the system starts running, a training operation is performed. The user focuses on the flashing target on the training interface and raises their eyebrows while the target is flashing. The signal recognition and processing process during training is as follows: Step 31: The acquisition circuit synchronously acquires human bioelectric signals, performs bandpass filtering on the acquired data, extracts sampling points of a certain time length from the data, and downsamples the sampling points. Step 32: Construct a feature vector from the downsampled data, and store the feature vector corresponding to each time the target flashes in the computer; Step 33: Using the result obtained after averaging the feature vector as a threshold, establish a training model for the user to provide a personal model for subsequent recognition operations; Step 4: Next, the user performs the GUI operation: In the current GUI, each virtual object or button flashes once randomly. While the virtual object or button that the user needs to select is flashing, the user selects the target by raising their eyebrows. Step 5: The target to be selected in Step 4 is chosen by raising the eyebrows. The specific target recognition process is as follows: Step 51: The EOG acquisition circuit takes a signal from the signal acquisition terminal corresponding to the digital mark as a feature vector. After all n buttons have flashed, n feature vectors are generated. These n feature vectors form a feature vector set Di. Step 52: After preprocessing the feature vectors in the feature vector set Di, the feature vectors are then judged by a classifier; let L in this method represent the number of times the average of the current classification is calculated, that is, the number of rounds. min and L max Let θ and θ0 represent the minimum and maximum number of rounds required to identify a button, respectively; θ0 is the threshold controlling the output; then the specific judgment method is as follows: (1) System initialization, set L = 0; (2) Receive the new round of data and update L = L + 1; (3) In the current round, corresponding to the number of targets in the GUI, feature extraction is performed to obtain n feature vectors; (4) For each specific button, perform an average operation on the cumulative round feature vectors of the first L; (5) Use a classifier to calculate the regression score for each button, denoted as RS. i RS i ={RS1, RS2, ... RS n }; Then, these RS i The scores are normalized, and the resulting n target regression scores are all in the interval [0, 1]. (6) If L ≥ L min And L < L max If the result is positive, proceed to step (7); otherwise, proceed to step (9). (7) Calculate RS max and RS sec The difference: Δθ = RS max -RS sec In the formula, RS max Represents RS i The maximum value among the nine regression scores, and S sec Represents the second largest value; (8) If Δθ > θ0, proceed to step (9); otherwise, proceed to step (2). (9) If L≥ L max Immediately jump to step (10); (10) End the current trial recognition and output the maximum recognition score RS. max The corresponding button; simultaneously jump to step 1, preparing for the recognition of the next trial; Step 53: Based on the method in step 52, determine whether the target feature vector exists in the feature vector set Di. If the target feature vector exists, proceed to the next step; otherwise, proceed to step 4. Step 54: Repeat step 5 at least once to obtain feature vector set Di+1. Determine whether the feature vectors in feature vector set Di and feature vector set Di+1 are the same. If they are the same, determine the selection key according to the target feature vectors in feature vector set Di and feature vector set Di+1, and proceed to the next step. Otherwise, return to step 4. Step 6: Determine if the current selection is the end output. If yes, proceed to step 10; otherwise, proceed to the next step. Step 7: Determine if the input is complete. If the input is complete, proceed to step 53; otherwise, proceed to the next step. Step 8: Proceed to the next work row and repeat steps 4 and 5; Step 9: Then proceed according to the entered selections; Step 10: End selection.

3. The EOG-based smart helmet control method according to claim 2, characterized in that: The classification method adopted in step 52 is as follows: Suppose that the target y is linearly correlated with the feature vector X, and Z is Gaussian noise, then the linear classifier takes the following form: ； set up The training set consists of feature vectors. and the corresponding binary state Composition; weights can be obtained Likelihood function: ； In the formula, Y represents a vector containing all training targets. This represents the matrix obtained by horizontally stacking the training feature vectors. The inverse variance of the noise is represented by L, and the number of training samples is represented by L. In Bayesian linear discriminant analysis (BLDA), by extending... The dimension of, i.e. The last term is the bias term, which incorporates the Gaussian noise Z into the weights. In the middle, assuming weights The prior distribution is: ； The accuracy of this distribution, where It is expanded to n+1 dimensions, with the last term being the bias term. It is an n+1 dimensional diagonal matrix, in the form of: ； Based on the likelihood function and the prior distribution, the posterior distribution can be calculated using Bayes' rule as follows: ； Since both the prior and likelihood distributions are Gaussian, the posterior distribution also has a Gaussian form. Its distribution can be determined by the mean m of w and the covariance C, as shown below: ； ； Ultimately, for We obtain its predicted regression value. Distribution function: ； in Represents a normal distribution; predicts the mean of the distribution function. and variance for: ； ； In which the mean This is the regression target value.

Images