Eye movement interaction system based on magnetic contact lenses

Abstract

The invention discloses an eye movement interaction system based on magnetic contact lenses, belongs to the field of man-machine interaction, and particularly relates to a mode recognition and classification method in the field of eye movement detection and design and implementation of a real-time interaction system of the mode recognition and classification method. The magnetic contact lens abandons the dependence on illumination, is suitable for dark scenes, frequent blinking or eyelid sagging and other special situations, and widens the applicable scene boundary of eye movement detection. The design of the magnetic contact lenses based on the flexible material has high comfort, the Hall effect sensor and the lens bracket are integrally matched to improve the portability and the stability of the system, and the system is suitable for long-time wearing. High-robustness signal decoding is achieved through a machine learning model, simulation experiments and real tests show that the recognition accuracy can be stabilized at 85%, and the system response time is within 200 ms. The system supports various interaction modes such as English input, mouse control and eye movement tracking, and can be expanded to various scenes or medical rehabilitation auxiliary application in the future.

Description

Technical Field

[0001] The present invention belongs to the field of human-computer interaction, and in particular relates to a pattern recognition and classification method in the field of eye movement detection and the design and implementation of a real-time interaction system thereof. Background Art

[0002] With the rapid development of smart wearable technology and human-computer interaction systems, eye tracking, as an efficient and natural human-computer interaction method, has shown broad application prospects in medical rehabilitation, virtual reality, intelligent control, driver assistance, and accessible communications. By accurately sensing the direction, trajectory, and dwell time of eye gaze, the system can understand user intent and enable more convenient interaction. Currently, mainstream eye movement detection technologies rely primarily on camera-based optical methods, particularly infrared video eye trackers, which detect gaze direction by capturing the corneal reflection point and pupil center position, offering high detection accuracy and resolution. However, traditional optical eye tracking technology still has a number of significant limitations in practical applications. First, optical tracking is sensitive to external lighting conditions and is prone to misjudgment or signal loss in low-light or high-light environments. Furthermore, the technology is highly dependent on the state of the eyelids. If the user closes their eyes or exhibits symptoms such as ptosis, the system will be unable to effectively detect eye movement direction. Second, due to the requirement for high-frame-rate cameras and image processing algorithms, the overall system is bulky, power-hungry, and expensive, making it difficult to apply in mobile and portable scenarios, especially for long-term wear or privacy-conscious applications.

[0003] In recent years, smart contact lens technology, as an emerging wearable platform, has gradually attracted widespread attention from academia and industry. By integrating micro-sensors, energy harvesting and transmission modules into flexible lenses, these products possess multiple functions such as physiological parameter monitoring, drug release control, and augmented reality display, demonstrating the potential of "invisible wearable devices" in the next generation of interactive terminals. Eye tracking technology based on smart contact lenses has therefore become a research hotspot. However, most existing research focuses on monitoring physiological indicators such as intraocular pressure and glucose levels, with relatively little exploration of eye movement recognition and interactive applications. Related systems are still in the proof-of-concept stage and have not yet formed a practical interactive solution.

[0004] In this context, magnetic contact lenses, as a new invisible device with non-optical detection capabilities, can work with an external Hall-effect sensor array to sense and calculate the magnetic field disturbances caused by eye movement in real time, thereby accurately determining the direction and trajectory of eye movement. Compared to optical solutions, magnetic induction methods are less sensitive to lighting conditions and eyelid occlusion, and have greater environmental adaptability. Furthermore, the sensor is small, low-power, and well-integrated and portable, making it suitable for special application scenarios such as dark environments, eyelid disorders, and critically ill patients.

[0005] Magnetic contact lenses overcome the limitations of traditional optical methods and provide a technological foundation for a new, more environmentally adaptable, low-power, and highly private eye movement detection and interaction method. This invention, developed within this technological context, aims to construct an intelligent contact lens eye movement interaction system and promote its practical application in cutting-edge fields such as human-computer interaction, medical assistance, and virtual reality. Summary of the Invention

[0006] In response to the current stagnation in the development of invasive eye tracking technology and the immature application of smart contact lenses in the fields of human-computer interaction and eye tracking, this invention takes magnetic contact lenses as the research core, explores their application potential in eye tracking and interactive control, and further explores their unique advantages and applicable scenarios compared to traditional optical eye tracking technology.

[0007] The present invention proposes an eye movement interaction system based on magnetic contact lenses, which includes: a magnetic contact lens, a magnetic field signal acquisition module, a signal processing and eye movement recognition module, and an interactive output module;

[0008] The magnetic contact lens is worn on the eyes of the target person; the magnetic field signal acquisition module is a pair of frame glasses, which are matched with the target person. Each frame of the frame glasses is equipped with a circle of Hall effect sensors and a corresponding acquisition controller for detecting magnetic field changes; the sampling data of the Hall effect sensors is transmitted to the signal processing and eye movement recognition module for eye movement recognition, and the recognition results are sent to the interactive output module, which controls the interactive target based on the recognition results;

[0009] The magnetic field signal acquisition module adopts a multi-threaded architecture, and the data collected by each Hall effect sensor is set up as an independent sub-thread for transmission. The magnetic field signal acquisition module also includes a data buffer pool. The buffer pool caches in the following manner: after the acquisition controller completes a three-way handshake with all Hall effect sensors and establishes a connection, the acquisition controller controls the Hall effect sensors to continuously acquire magnetic field change signals and continuously writes newly received TCP data packets to the buffer pool until the number of stored data points exceeds N. The TCP data packets are data packaged by the Hall effect sensors using the TCP protocol, and the data packets include magnetic field strength information of eight channels at the same time. When the amount of data stored in the buffer pool exceeds N data points, the system automatically removes the earliest N milliseconds of data and unpacks the newly entered data for transmission. The buffer pool uses low-byte data to store low addresses and high-byte data to store high addresses. Therefore, the magnetic field strength of a single channel at a single time point requires 4 bytes to be fully represented and must be restored in order from low byte to high byte. The eight channels share one buffer pool.

[0010] The specific processing method in the signal processing and eye movement recognition module is:

[0011] Step 1: Preprocessing. The preprocessing method includes two stages: data prediction and update.

[0012] In the forecasting phase, it is estimated and covariance estimation Calculated by the following formulas:

[0013]

[0014] in, is a priori estimate of the magnetic field strength, represents the posterior magnetic field strength at time k-1, Prior error covariance matrix, P k-1 represents the posterior error covariance matrix at time k-1, A is the state transfer matrix, B is the control input matrix, u k-1 is the control input, Q is the process noise covariance matrix, and the superscript T indicates the transpose;

[0015] In the update phase, the predicted magnetic field strength is corrected using the observed data; the update equation is as follows:

[0016]

[0017] Among them, K k is the gain matrix, H is the measurement matrix, R is the observation noise covariance matrix, z k is the measured value, is the posterior state estimate, P k is the posterior error covariance matrix, I is the identity matrix

[0018] Step 2: Calculate the mean, maximum, and minimum values ​​of the magnetic field data within the sampling window as features for pattern recognition;

[0019] Step 3: Support vector machine builds an eye tracking classification model to identify eye movements.

[0020] Furthermore, the number of Hall effect sensors is 8, N = 100, and 3200 bytes are parsed within a 100ms sampling window to fully restore the eight-channel magnetic field strength data; the depacketization formula for the TCP data packet is as follows:

[0021]

[0022] Where V[i] represents the magnetic field strength measured by the sensor at a single channel and a single time point, and D[i] represents a single byte in the buffer pool;

[0023] Furthermore, when training the eye tracking classification model constructed by the support vector machine, a magnetic field signal acquisition module is used to collect sample signals, and the sample signals are labeled automatically or manually; the label is the direction of eye movement.

[0024] Furthermore, the interactive target is a mouse, and eye movements up, down, left, and right correspond to mouse movements up, down, left, and right, respectively, achieving basic cursor positioning functions. The remaining four directions are left-click, right-click, middle-click, and exit functions of the mouse.

[0025] Furthermore, the interactive target is the twenty-six English letters. The first layer of coding divides the 26 English letters into multiple subgroups and maps them to the four eye movement directions of up, down, left and right respectively; the remaining four directions correspond to the space, period, delete and exit functions respectively; the second layer of coding provides eight directions of detailed selection, each direction corresponds to a specific letter, and one is reserved for returning to the first layer function; the user selects the final target character through the second eye movement.

[0026] Furthermore, the eye movement interaction system is first calibrated before interaction. The calibration method is as follows:

[0027] A nine-point grid calibration method is used to establish the mapping relationship between the magnetic field space and the screen space; the calibration points are evenly distributed in a 3×3 matrix; the system displays nine balls in sequence at the upper left, upper, upper right, left, center, right, lower left, lower, and lower right positions on the screen, and requires the eyes to fixate in the corresponding directions; weighted center of mass positioning is performed during the fixation process.

[0028] The present invention has the following advantages:

[0029] This study built a complete eye-movement interaction system based on magnetic contact lenses, which has the following advantages:

[0030] 1. Magnetic contact lenses eliminate the dependence on light and are suitable for special situations such as dark scenes, frequent blinking or drooping eyelids, broadening the scope of applicable scenarios for eye movement detection.

[0031] 2. The magnetic contact lens design based on flexible materials is highly comfortable. The integrated adaptation of the Hall effect sensor and the frame improves the portability and stability of the system and is suitable for long-term wear.

[0032] 3. The machine learning model achieves highly robust signal decoding. Simulation experiments and real tests show that the recognition accuracy can be stabilized at 85%, and the system response time is within 200ms.

[0033] 4. The system supports multiple interaction methods such as English input, mouse control, eye tracking, etc., and can be expanded to various scenarios or medical rehabilitation assistance applications in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is the system connection and acquisition flow chart.

[0035] Figure 2 Store, annotate, and analyze graphs for data.

[0036] Figure 3 This is the signal processing flow chart.

[0037] Figure 4 This is a schematic diagram of the mouse control module.

[0038] Figure 5 This is a schematic diagram of the English input module.

[0039] Figure 6 Schematic diagram of the structure of a gallium arsenide-based Hall effect sensor.

[0040] Figure 7 Schematic diagram of 3D printed glasses frame.

[0041] Figure 8 This is a diagram showing a 3D eyeball.

[0042] Figure 9 This is the model test result diagram.

[0043] Figure 10 This is the test result diagram of the English input module. DETAILED DESCRIPTION

[0044] System connection and acquisition process Figure 1As shown. In addition to the main thread of the system, this study adopts a multi-threaded architecture and places the connection between the system background and the sensor data processing unit in an independent sub-thread. In order to avoid catastrophic consequences caused by multiple threads operating on shared data segments at the same time, the system uses critical section resources to protect common data blocks. In view of the data packet transmission characteristics of the TCP protocol, this study has built a set of efficient data caching and unpacking mechanisms in the system background. Since TCP is a streaming transmission protocol, the order in which its data arrives is not strictly arranged according to the sampling time points, so it is necessary to establish a buffer pool on the system side to receive data packets to ensure signal integrity. The present invention sets the buffer pool length to 100ms, that is, under the sampling rate of 1000Hz, the buffer pool can accommodate data from 100 time points. After the three-way handshake is completed to establish a connection, the system continues to write newly received TCP data packets to the buffer pool until the number of stored data points exceeds 100. When the amount of data stored in the buffer pool exceeds 100 data points, the system automatically removes the earliest 100ms data and unpacks the newly entered data to ensure the real-time and efficient utilization of the data stream. Because the system uses little-endian data storage, meaning the low byte is stored at a low address and the high byte is stored at a high address, the magnetic field intensity of a single channel at a single time point requires four bytes to fully represent, and must be restored in order from low byte to high byte. This study uses an eight-channel sensor for data acquisition, so within a 100ms sampling window (100 data points), 3200 bytes must be parsed to fully restore the eight-channel magnetic field intensity data. The unpacking formula is as follows:

[0045]

[0046] Where V[i] represents the magnetic field strength measured by the sensor at a single channel and a single time point, and D[i] represents a single byte in the buffer pool;

[0047] Furthermore, in response to real-time application and data analysis needs, this study designed the following Figure 2The data storage, labeling, and real-time visualization of magnetic field signals shown in the figure meet the system's subsequent offline analysis and online real-time analysis requirements for magnetic field signals. The system uses the ofstream function of the iostream library in the C++ standard library to open or create a text file, and appends the newly collected data to the last line of the file in sequence through stream output. The system provides two labeling methods to meet different experimental requirements: 1) In training mode or simulation experiments, the system automatically labels according to the timestamp and experimental status based on the preset experimental process. For example, when conducting an eye movement direction classification experiment, the system can synchronously mark the current eye movement direction in the magnetic field signal data. 2) During the data collection process, users can independently mark special events in the experiment. During the data collection period, users can start special labeling by pressing the space bar and end it by pressing the space bar again. This function facilitates users to analyze sudden events and helps users capture instantaneous changes in the magnetic field. In addition, the present invention integrates the real-time display function of the magnetic field signals of each channel into the system, so that users can intuitively observe the magnetic field changes generated by the magnetic contact lens during eye movement.

[0048] The main contents of the signal processing and eye movement recognition module are as follows: Figure 3 The three steps of signal preprocessing, feature extraction and pattern recognition are shown. The present invention introduces preprocessing of the original magnetic field signal. It includes two stages: prediction and state update. In the prediction stage, the system predicts the state (here, the magnetic field strength) and error covariance based on the system's state equation. State estimation and covariance estimation Calculated by the following formulas:

[0049]

[0050] in, is the prior estimate of the state, Prior error covariance matrix, A is the state transfer matrix, B is the control input matrix, u k-1 is the control input, and Q is the process noise covariance matrix.

[0051] In the update phase, the system uses the observed data to correct the prediction results. The update equation is as follows:

[0052]

[0053]

[0054] Among them, K k is the gain matrix, H is the measurement matrix, R is the observation noise covariance matrix, z k is the measured value, is the posterior state estimate, Pk is the posterior error covariance matrix, and I is the identity matrix.

[0055] After the signal preprocessing is completed, it is necessary to extract features that can effectively characterize the eye movement information in order to build an efficient classification model. The present invention divides the time series signal into 200 sampling points, i.e. 200ms, as a feature window to ensure that the change pattern of the magnetic field intensity during eye saccade and gaze can be fully covered. The amplitude change of the periocular magnetic field can effectively characterize the movement characteristics of the eyeball. Therefore, in the feature extraction process, the present invention selects the mean, maximum and minimum values ​​as model features to capture the main change trends of the magnetic field signal. The present invention uses a support vector machine to construct an eye tracking classification model. Support vector machine is a supervised learning algorithm widely used in pattern recognition. It can realize effective classification of different categories in feature space by constructing an optimal hyperplane. Its core idea is to find the maximum margin hyperplane to improve the generalization ability of classification.

[0056] Step 1: Collect the magnetic field signal around the eye. Figure 6 Customized miniature GaAs-based Hall effect sensors and Figure 7 As shown in the 3D-printed eyeglass frame, eight Hall-effect sensors are first embedded in a ring shape within the custom frame. The sensor processing unit is then connected to the backend system. Finally, the user puts on the magnetic contact lenses and then the 3D-printed frame. After completing these steps, the eight sensors begin sensing magnetic field changes around the eye and transmit signals to the backend system at a sampling rate of 1000Hz for signal acquisition.

[0057] Step 2: Start training and collect data. To reduce external magnetic field interference and enhance data reliability, training is conducted in a relatively stable magnetic field environment. First, the user sits quietly in front of the computer with their eyes fixed squarely to collect resting-state data. A standardized visual guidance process is then used. During the visual prompt phase, the system randomly displays a hollow red arrow pointing in a specific direction for 2 seconds. This phase allows the user to prepare for eye movements and ensures stable data collection. During the eye movement phase, the system immediately displays a solid red arrow in the same direction for 2 seconds, prompting the user to officially perform an eye movement. At this point, the system simultaneously activates the magnetic field data acquisition device to record changes in magnetic field intensity sensed by the sensor. After each trial, the system pauses for 1 second to prevent data aliasing caused by continuous eye movements. The experiment then proceeds to the data collection phase for the next direction. The user is required to complete eight eye movements in different directions (up, down, left, right, upper left, upper right, lower left, and lower right). The order of trials within each direction is randomized to avoid learning effects or directional bias during data collection. The entire experimental process was repeated 10 times to ensure sufficient data for subsequent support vector machine model training and testing.

[0058] Step 3: Signal processing and data training. Process the data during user training, smooth the raw data, and filter out signal noise. Divide the data into trials of 200 points (200ms) each and extract its labels, and perform feature extraction on each trial. The extracted training features are the maximum, minimum, and mean of the trial. Use five-fold cross validation to input the spliced ​​features and corresponding labels into the support vector machine model for model training. Finally, save the trained model in the form of a .model file. The system background loads the .model file and reads the model parameters. This facilitates the subsequent testing process.

[0059] Step 4: Mouse control module. On the basis of stable mouse control, individuals can use eye movements to complete a variety of common computer operations. The user can trigger the right mouse button operation by turning the eyeballs toward the upper right, thereby realizing functions such as popping up the document menu; then, by controlling the movement of the mouse pointer and combining it with a left click, the user can complete common document operations such as opening a file and selecting content. In the web browsing scenario, the user can simulate a left mouse click by moving the eyeballs toward the upper left to access the web page link; and when the user moves the eyeballs toward the lower left, the system can recognize it as a middle mouse button click, and with subsequent up and down eye movements, the web page scroll bar can be controlled, thereby completing up and down browsing of the page.

[0060] Step 5: Test the mouse control module. At the beginning of the experiment, a solid yellow circle with a radius of 75 pixels appears on the screen. The user is required to control the mouse cursor by eye movement, starting from the center of the screen, to move it to the target circle. Once the cursor successfully contacts the circle, the system generates a new target location and asks the subject to continue moving the cursor for the next round of the task. This entire process is repeated five times. The system records the time required for the subject to complete each target location task and compares it with the calculated theoretical time to evaluate the performance of the mouse control module. The formula for calculating the theoretical time is shown below.

[0061]

[0062] Among them, (X tar ,Y tar ) is the center coordinate point of the target circle, (X pos ,Y pos ) is the coordinate point of the center of the current circle, Step is the pixel step size for each trial, and n is the theoretical number of trials per second.

[0063] Step 6: English Input Module and Testing. The English Input Module allows users to complete the twenty-six English letters and even Chinese characters through eye movement input, enabling basic communication and information recording. The test set a set of English input tasks for users, consisting of six English letters randomly generated by the system backend. The random content was determined using pseudo-random numbers generated by the rand() function, and the srand() function was used to set the random seed to generate different random number sequences. During the experiment, the system recorded the time required for the subject to complete the target input, which was used to measure the input efficiency of the module.

[0064] Step 7: In order to enhance the user's interactive experience and make the system more suitable for actual application scenarios, the present invention integrates the English input module with the mouse control module, allowing patients to switch seamlessly between different tasks. The option originally used to exit the system has been adjusted to a module switching function, allowing users to switch freely between different modes. Each time the system switches to the English input module, the system automatically captures the window handle where the mouse cursor is currently located. After the user completes text input, the system will store the input content to the clipboard and automatically paste it to the target window previously captured. The handle is a unique identifier used to identify and manage system resources in computer programming, representing a specific object in the operating system. Through handle management, the system can ensure that the input content will not be misled to other windows. When the user uses the mouse control module in scenarios such as web browsing, online chatting, document editing, etc., he can switch to the English input module at any time to realize functions such as web search, text input, and message sending.

[0065] Step 8: Car control module and testing. Users can control the car forward, backward, left, and right by eye movement, completing various paths. The test set the car's completion path to two forward movements, two 90° left turns, and two forward movements and two 90° right turns. The forward, backward, left, and right commands for the car correspond to upward, downward, left, and right eye movements, respectively. After executing these commands, the car should return to its original position and resume its starting angle. The experiment evaluated the system's control accuracy and stability in a dark environment by recording the car's actual completed path and completion time.

[0066] Step 9: Eye tracking module and test. This module uses the nine-point calibration method to map the eye movement amplitude to screen coordinates. In order to provide more intuitive real-time eye movement visual feedback, the present invention also uses QML technology under the QT framework to design a 3D eye model that can be updated in real time with eye movement. The eye model is as follows: Figure 8 The test design continued with the mouse control module, but with a difference in the location of the visual stimulus targets. The yellow target circles appeared randomly instead of in a preset sequence. This was because the core objective of the mouse control module experiment was to verify the accuracy of the command decoding algorithm, while this experiment focused on examining the system's performance in eye movement perception.

[0067] System test results

[0068] The test results of the present invention were conducted in a simulation environment. The present invention first tested the accuracy of the algorithm model, and then evaluated the system performance by testing each module. The test was conducted on 20 groups of experiments. Each group of experiments contained eight eye movement directions and resting state data, and each type of data contained 100 trials to ensure the diversity of the training set and the generalization ability of the model. The training results are as follows Figure 9 Across 20 experiments, the model achieved an average classification accuracy of 84.36% for nine categories of eye movement data, with a standard deviation of 3.95. This result demonstrates that the model maintains high classification accuracy with a low standard deviation, indicating good stability and robustness.

[0069] In the mouse control module, according to the theoretical formula, the target coordinates and pixel step size Step are 10, and the theoretical number of trials per second is 5. The theoretical completion time of the first task is 25.44 seconds; the second is 7.28 seconds; the third is 1.76 seconds; the fourth is 13.02 seconds; and the fifth is 24.66 seconds.

[0070] Table 1 Mouse control module test results

[0071]

[0072]

[0073] As shown in Table 1, the test completion time deviated slightly from the theoretically calculated value, maintaining a consistent overall trend. In particular, the gap between the results and the theoretical prediction narrowed significantly for shorter single-task times. This indicates that the impact of accumulated eye control errors and system response delays is minimal during shorter task cycles, resulting in test times closer to the theoretical value. Furthermore, during the test, the cursor movement maintained high stability and tracking accuracy, enabling a relatively accurate mapping of the user's eye movements.

[0074] In the English output module, the present invention has carried out ten tests, and the test results are as follows Figure 10As shown in the figure, the longest input time was 62.32 seconds, the shortest was 32.75 seconds, and the average input time was 46.97 seconds. Theoretically, each level of selection can be completed within three trials, meaning that the user can select an English letter after two level selections. Based on the system response time, the input of each letter should theoretically be completed in approximately 2 seconds. Misselections leading to additional adjustments or re-entry contribute to the longer overall input time. During the test, users also needed to consider the location of the target letter and adjust their eye movement direction to accurately select it. This additional cognitive and executive process further prolonged the input time. As users become more familiar with the system, the time they spend selecting and adjusting their eye movement direction gradually decreases. After repeated practice, the input time for six letters can be stabilized between 30 and 40 seconds. Therefore, the English output module can assist users in completing short text input tasks.

[0075] In the eye tracking module, test results are shown in Table 2. Across ten tests, five tasks took an average of 1.69 seconds to complete, with a maximum of 2.48 seconds. Compared to the theoretical completion time of 200ms, the simulated completion time is primarily determined by reaction time and system errors. These results demonstrate that eye tracking technology, tracking the eye every 200ms, can accurately determine eye position in approximately 2 seconds.

[0076] Table 2 Eye tracking module test results

[0077]

[0078] In the vehicle control module under dark scenes, the present invention conducted five tests, and the test results are shown in Table 3 below.

[0079] Table 3 Test results of the car control module in dark scenes

[0080]

[0081] The results show that all five tests successfully reached the target position, indicating that the system has high stability in dark environments, can accurately analyze eye movement signals and control the car to run along the expected trajectory. In the five tests, only in the second experiment did the car execute an additional forward command and a backward command to return to the correct path due to untimely eye rotation. In terms of task completion time, the average completion time of the experiment was 33.69 seconds, the shortest time was 28.84 seconds, and the longest time was 40.27 seconds (due to additional correction operations). Therefore, the magnetic contact lens combined with the eye movement detection system can still work stably in a dark environment, can accurately analyze eye movement signals, and realize the control of the car.

Claims

1. An eye-movement interaction system based on magnetic contact lenses, the system comprising: Magnetic corneal contact lens, magnetic field signal acquisition module, signal processing and eye movement recognition module, interactive output module; The magnetic contact lens is worn on the eyes of the target person; the magnetic field signal acquisition module is a pair of frame glasses, which are matched with the target person. Each frame of the frame glasses is equipped with a circle of Hall effect sensors and a corresponding acquisition controller for detecting magnetic field changes; the sampling data of the Hall effect sensors is transmitted to the signal processing and eye movement recognition module for eye movement recognition, and the recognition results are sent to the interactive output module, which controls the interactive target based on the recognition results; The magnetic field signal acquisition module adopts a multi-threaded architecture, and the data collected by each Hall effect sensor is set up as an independent sub-thread for transmission. The magnetic field signal acquisition module also includes a data buffer pool. The buffer pool caches in the following manner: after the acquisition controller completes a three-way handshake with all Hall effect sensors and establishes a connection, the acquisition controller controls the Hall effect sensors to continuously acquire magnetic field change signals and continuously writes newly received TCP data packets to the buffer pool until the number of stored data points exceeds N. The TCP data packets are data packaged by the Hall effect sensors using the TCP protocol, and the data packets include magnetic field strength information of eight channels at the same time. When the amount of data stored in the buffer pool exceeds N data points, the system automatically removes the earliest N milliseconds of data and unpacks the newly entered data for transmission. The buffer pool uses low-byte data to store low addresses and high-byte data to store high addresses. Therefore, the magnetic field strength of a single channel at a single time point requires 4 bytes to be fully represented and must be restored in order from low byte to high byte. The eight channels share one buffer pool. The specific processing method in the signal processing and eye movement recognition module is: Step 1: Preprocessing. The preprocessing method includes two stages: data prediction and update. In the forecasting phase, it is estimated and covariance estimation Calculated by the following formulas: in, is a priori estimate of the magnetic field strength, represents the posterior magnetic field strength at time k-1, Prior error covariance matrix, P k-1 represents the posterior error covariance matrix at time k-1, A is the state transfer matrix, B is the control input matrix, u k-1 is the control input, Q is the process noise covariance matrix, and the superscript T indicates the transpose; In the update phase, the predicted magnetic field strength is corrected using the observed data; the update equation is as follows: Among them, K k is the gain matrix, H is the measurement matrix, R is the observation noise covariance matrix, z k is the measured value, is the posterior state estimate, P k is the posterior error covariance matrix, I is the identity matrix Step 2: Calculate the mean, maximum, and minimum values ​​of the magnetic field data within the sampling window as features for pattern recognition; Step 3: Support vector machine builds an eye tracking classification model to identify eye movements.

2. The eye-movement interaction system based on magnetic contact lenses according to claim 1, characterized in that: The number of the Hall effect sensors is 8, N=100, and 3200 bytes are parsed within a 100ms sampling window to fully restore the eight-channel magnetic field strength data. The depacketization formula for the TCP data packet is as follows: Where V[i] represents the magnetic field strength measured by the sensor at a single channel and a single time point, and D[i] represents a single byte in the buffer pool.

3. The eye-movement interaction system based on magnetic contact lenses according to claim 1, characterized in that: When training the eye tracking classification model constructed by the support vector machine, a magnetic field signal acquisition module is used to collect sample signals, and the sample signals are labeled automatically or manually; the label is the movement direction of the eyeball.

4. The eye-movement interaction system based on magnetic contact lenses according to claim 1, characterized in that: The interactive target is the mouse. Moving the eyeball up, down, left, and right corresponds to moving the mouse up, down, left, and right, respectively, achieving basic cursor positioning functions. The remaining four directions are left-click, right-click, middle-click, and exit functions.

5. The eye-movement interaction system based on magnetic contact lenses according to claim 1, characterized in that: The interactive target is the 26 English letters. The first layer of coding divides the 26 English letters into multiple subgroups and maps them to the four eye movement directions of up, down, left and right respectively; the remaining four directions correspond to the space, period, delete and exit functions respectively; the second layer of coding provides eight directions of detailed selection, each direction corresponds to a specific letter, and one is reserved for returning to the first layer function; the user selects the final target character through the second eye movement.

6. The eye-movement interaction system based on magnetic contact lenses according to claim 1, characterized in that: Before the eye movement interaction system interacts, the system must be calibrated first. The calibration method is as follows: A nine-point grid calibration method is used to establish the mapping relationship between the magnetic field space and the screen space; the calibration points are evenly distributed in a 3×3 matrix; the system displays nine balls in sequence at the upper left, upper, upper right, left, center, right, lower left, lower, and lower right positions on the screen, and requires the eyes to fixate in the corresponding directions; weighted center of mass positioning is performed during the fixation process.

Images