Child vision screening and tracking system and method based on computer vision

Abstract

The invention relates to the technical field of ophthalmology medical information systems, in particular to a child vision screening and tracking system and method based on computer vision, and the system comprises an eyeball movement analysis module which is used for collecting and analyzing eyeball movement data; the AR test module is used for carrying out eyeball movement enhancement and vision test; the vision development prediction module is used for predicting the vision development trend; the vision knowledge base is used for providing professional knowledge support; the intelligent assistant module is used for providing personalized suggestions; the system background is used for realizing data uploading and statistical analysis; according to the invention, a comprehensive and intelligent children vision health management platform is constructed. By integrating advanced technologies such as computer vision, augmented reality, artificial intelligence and the like, the system realizes full-process automation from data acquisition, analysis and processing to prediction and suggestion. The efficiency and accuracy of vision screening are greatly improved, and powerful technical support is provided for long-term management of vision health of children.

Description

Technical Field

[0001] The present invention relates to the technical field of ophthalmic medical information systems, and more specifically, to a computer vision-based children's vision screening and tracking system and method thereof. Background Art

[0002] In recent years, with the popularity of electronic devices and the increase in learning pressure, children's myopia has become increasingly serious and has become a global public health issue. Traditional children's vision screening methods mainly rely on regular hospital examinations or vision screening organized by schools. However, these methods have many shortcomings.

[0003] First, traditional vision testing methods often rely on children's subjective feedback, such as identifying letters or graphics on a vision chart. The accuracy of this method is easily affected by children's attention, comprehension and cooperation. It is especially difficult to obtain accurate test results for younger children.

[0004] Secondly, traditional vision screening can usually only provide vision status at a certain point in time, making it difficult to continuously and dynamically monitor the development trend of children's vision. This makes it difficult to detect and intervene in vision problems at an early stage, missing the best time for prevention and treatment.

[0005] Furthermore, existing vision examination methods often lack interest and interactivity, making it difficult to mobilize children's enthusiasm, which not only affects the accuracy of the examination, but also makes regular vision examinations an unpleasant thing, reducing the compliance of parents and children.

[0006] In addition, most existing children's vision management systems are limited to data recording and simple statistical analysis, and lack the ability to predict future vision development. This makes it difficult to develop personalized vision care plans and fails to provide strong data support for public health decision-making.

[0007] Finally, existing vision screening systems are often isolated and difficult to effectively integrate data and share information with other health management systems, which limits the comprehensive and systematic management of children's vision health.

[0008] Based on the above problems, there is an urgent need for a vision screening and tracking system that can objectively, accurately and continuously monitor the development of children's vision and has interesting and predictive capabilities. The present invention is proposed to solve these technical problems. Summary of the invention

[0009] The present invention proposes a child vision screening and tracking system and method based on computer vision, aiming to solve the problems of strong subjectivity, poor continuity, lack of fun, weak prediction ability and low system integration in the prior art. Through innovative system architecture design and application of advanced algorithms, the present invention realizes objective measurement, dynamic tracking, fun interaction, intelligent prediction and comprehensive management of child vision.

[0010] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0011] Children's vision screening and tracking system based on computer vision, including:

[0012] Eye movement analysis module, used to collect and analyze eye movement data;

[0013] AR test module for eye movement enhancement and vision testing;

[0014] Vision development prediction module, used to predict vision development trend;

[0015] Vision knowledge base, used to provide professional knowledge support;

[0016] Smart assistant module for providing personalized suggestions;

[0017] System backend, used for data upload and statistical analysis;

[0018] Among them, the eye movement analysis module, AR test module, vision development prediction module, vision knowledge base, intelligent assistant module and system background form a closed-loop vision management system through data interaction.

[0019] Preferably, the eye movement analysis module comprises:

[0020] An image acquisition unit, used for acquiring eye movement images;

[0021] An image preprocessing unit, used for graying and filtering the collected image;

[0022] A feature extraction unit, used for extracting eye movement features from the preprocessed image;

[0023] A data generation unit, used for generating a training set and a test set based on the extracted features;

[0024] The feature extraction unit extracts eye movement features through top hat transformation denoising, edge detection, skeleton extraction and Hough circle detection steps.

[0025] Preferably, the top hat transformation denoising step in the feature extraction unit adopts the following formula:

[0026]

[0027] Among them, A is the grayscale image of the image at time t, u t is the top-hat transformation mask at time t, γ is the opening operation, Open operation for morphology.

[0028] Preferably, the Hough circle detection step in the feature extraction unit adopts the following formula:

[0029] (x) 2 +(yb) 2 =r 2

[0030] Among them, (a, b) are the coordinates of the center of the circle, r is the radius of the circle, and (x, y) are the coordinates of any point on the circle.

[0031] Preferably, the AR test module comprises:

[0032] An eye movement enhancement unit, used to enhance eye movement features using deep learning technology;

[0033] An eye movement recognition unit, used to compare the enhanced eye movements with the training set to generate a prediction model;

[0034] Vision testing unit, used to simulate vision testing scenarios through AR technology and provide testing instructions;

[0035] The eye movement enhancement unit performs binarization, filling and corrosion processing on the eye movement data.

[0036] Preferably, the corrosion processing step in the eye movement enhancement unit adopts the following formula:

[0037]

[0038] Among them, A is the eye movement binary map, B is the erosion operator kernel, x is the coordinate of the point to be eroded, B(-x) is the kernel with the origin at x, and E(A, B) is the erosion result.

[0039] Preferably, the vision development prediction module comprises:

[0040] A diopter prediction unit, used for predicting diopter;

[0041] An axial length prediction unit, used for predicting the axial length;

[0042] Among them, the refractive power prediction unit adopts a multivariate linear regression algorithm, and the axial length prediction unit adopts a long short-term memory network algorithm.

[0043] Preferably, the diopter prediction unit adopts the following multivariate linear regression formula:

[0044] y=μ+δ0+δ1x1+δ2x2+δ3x3+δ4x4+δ5x5

[0045] Among them, y is the predicted refractive power, μ is the intercept, δ1, δ2, δ3, δ4, δ5 are regression coefficients, and x1, x2, x3, x4, x5 are the influencing factors of axial length, current refractive power, amblyopia, myopia and astigmatism, respectively.

[0046] Preferably, the axial length prediction unit adopts the following long short-term memory network formula:

[0047] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0048] i t =o(W i ·[h t-1 ,x t ]+b i )

[0049] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0050] c t =f t *c t-1 +i t *tanh(W c ·[h t-1 ,x t ]+b c )

[0051] h t =o t *tanh(c t )

[0052] Among them, f t ,i t , o t They are forget gate, input gate and output gate respectively, c t is the unit state, h t is the hidden state, x t is the input, W f , W i , W o , W c is the weight matrix, bf , b i , b o , b c is the bias vector, σ is the sigmoid function, and tanh is the hyperbolic tangent function.

[0053] The child vision screening and tracking method based on computer vision includes the following steps:

[0054] Collect eye movement data;

[0055] Preprocess and extract features of the collected data;

[0056] Eye movement enhancement and vision testing based on the extracted features;

[0057] Use predictive models to predict vision development trends;

[0058] Provide personalized health recommendations based on prediction results;

[0059] Upload the test data to the system backend for statistical analysis;

[0060] Among them, the feature extraction, eye movement enhancement, and vision development prediction all adopt the corresponding algorithms.

[0061] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0062] The system of the present invention includes an eye movement analysis module, an AR test module, a vision development prediction module, a vision knowledge base, an intelligent assistant module, and a system background. These modules form a closed-loop vision management ecosystem through close data interaction. The innovation and beneficial effects of the system are mainly reflected in the following aspects:

[0063] First, from a macro perspective, the present invention builds a comprehensive and intelligent children's vision health management platform. By integrating advanced technologies such as computer vision, augmented reality, and artificial intelligence, the system realizes the automation of the entire process from data collection, analysis and processing to prediction and recommendation. This not only greatly improves the efficiency and accuracy of vision screening, but also provides strong technical support for the long-term management of children's vision health.

[0064] Secondly, the eye movement analysis module of the present invention uses advanced computer vision technology to capture children's eye movements through a high-speed camera and extract key features using a series of image processing algorithms. This method greatly improves the objectivity and accuracy of vision measurement and effectively solves the problem of traditional methods over-relying on children's subjective feedback. Especially for children who are younger or less cooperative, this non-contact measurement method shows obvious advantages.

[0065] Furthermore, the present invention innovatively introduces AR technology into the vision test process. The AR test module conducts vision tests through vivid and interesting virtual scenes, which greatly improves the fun of the test and the participation of children. This not only makes the vision test an interesting experience, but also obtains more accurate test results. The combination of AR technology and traditional vision test methods reflects the innovative thinking of the present invention in solving technical contradictions.

[0066] In addition, the vision development prediction module of the present invention uses advanced machine learning algorithms, such as multivariate linear regression and long short-term memory network (LSTM). The coordinated application of these algorithms enables the system to accurately predict the future vision development trend of children based on historical data and current test results. The realization of this function fills the gap in the existing technology in vision development prediction and provides an important basis for early intervention and personalized management.

[0067] The smart assistant module provides users with personalized health advice based on the prediction results and vision knowledge base. This intelligent health management method greatly improves the pertinence and effectiveness of vision care. At the same time, the presentation of these suggestions through AR technology further enhances user experience and compliance.

[0068] Finally, the data management and analysis functions of the system background enable the present invention to not only serve individual children, but also provide data support for large-scale public health decision-making. Through statistical analysis of massive data, the general laws and regional characteristics of children's vision development can be revealed, providing a scientific basis for the formulation of targeted vision protection policies.

[0069] In general, the present invention achieves functional complementarity and effect superposition through the organic combination and collaborative work of multiple modules. The system not only solves various problems existing in the prior art, but also achieves a qualitative leap in the management of children's vision health. It provides a one-stop solution for accurate screening, dynamic tracking, intelligent prediction and comprehensive management of children's vision. In the context of the increasingly serious problem of myopia in children, it has important practical significance and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a flowchart of the overall system of the present invention.

[0071] Figure 2 It is a logic block diagram of the eye movement analysis module of the present invention.

[0072] Figure 3 It is a logic block diagram of the AR test module of the present invention.

[0073] Figure 4 This is a logical block diagram of the vision development prediction module of the present invention.

[0074] Figure 5 This is a logic block diagram of the intelligent assistant module of the present invention.

[0075] Figure 6 It is a logic block diagram of the system background of the present invention. DETAILED DESCRIPTION

[0076] The scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. The embodiments in the present invention and all other embodiments obtained by ordinary persons in the art without making creative work are within the scope of protection of the present invention.

[0077] like Figure 1-6 As shown, the computer vision-based children's vision screening and tracking system of the present invention includes an eye movement analysis module 1, an AR test module 2, a vision development prediction module 3, a vision knowledge base 4, an intelligent assistant module 5 and a system background 6. These modules form a closed-loop vision management system through close data interaction.

[0078] The eye movement analysis module 1 is the basis of the whole system, which is responsible for collecting and analyzing the eye movement data of children. In practical applications, this module can be integrated into smart glasses or dedicated vision detection equipment to capture the tiny movements of children's eyes through a high-speed camera. Preferably, the sampling rate of the camera should be no less than 240Hz to ensure that the details of rapid eye movements (such as saccades) are captured.

[0079] The AR test module 2 is a major innovation of the present invention. It combines augmented reality (AR) technology with traditional vision tests, which not only improves the accuracy of the test, but also greatly enhances the participation of children. For example, the system can present various interesting vision test graphics in the AR environment, such as animated characters or interesting geometric figures, to guide children to complete the test tasks. This method is particularly suitable for children aged 3-12 years old and can effectively reduce the problem of distraction during the test.

[0080] The vision development prediction module 3 is one of the core functions of the system. It predicts the future vision development trend of children by analyzing historical data and current test results. The prediction accuracy of this module is directly related to the effectiveness of subsequent intervention measures. In one embodiment of the present invention, this module uses advanced machine learning algorithms, such as multivariate linear regression and long short-term memory network (LSTM), to improve the accuracy of prediction.

[0081] The Vision Knowledge Base 4, as the "brain" of the system, stores a large amount of professional knowledge and data related to children's vision development. This includes but is not limited to the normal vision development curve, common vision problems and their characteristics at all ages, and effective vision care methods. The content of the knowledge base should be updated regularly to reflect the latest medical research results.

[0082] The intelligent assistant module 5 is an important interface for the system to interact with users (children and their parents). Based on the results of vision development prediction and the support of the vision knowledge base, the intelligent assistant can provide personalized vision care advice for each child. For example, for children who are detected to have myopia tendencies, the system may recommend increasing outdoor activity time or adjusting eye habits.

[0083] The system backend 6 is responsible for the unified management and analysis of data. It not only stores the historical test data of each child, but also performs statistical analysis on the overall data to provide strong support for public health decision-making. For example, by analyzing the vision development trend of children in a certain area, it can provide data support for the local education department to formulate vision protection policies.

[0084] Furthermore, the eye movement analysis module 1 comprises an image acquisition unit 11, an image preprocessing unit 12, a feature extraction unit 13 and a data generation unit 14. The coordinated work of these units ensures high-quality acquisition and processing of eye movement data.

[0085] The image acquisition unit 11 is responsible for capturing the eye movement images of the child. In order to obtain a clear image, the present invention recommends using a high-definition camera with a resolution of not less than 1080p, and ensuring that the ambient light is sufficient but not glaring. The ideal lighting condition should be between 300-500 lux, which helps to reduce the interference caused by eye reflection.

[0086] The image preprocessing unit 12 performs preliminary processing on the collected original image, including graying and filtering. Graying can simplify the subsequent processing steps, while filtering helps to remove image noise. The present invention preferably uses Gaussian filtering, the kernel size of which can be set to 3x3 or 5x5, and the σ value can be selected between 0.5-1.5, so that small noise points can be effectively removed while retaining the eyeball contour.

[0087] The feature extraction unit 13 is the core of eye movement analysis. It extracts useful eye movement features from the preprocessed image through a series of complex image processing steps. These steps include top hat transform denoising, edge detection, skeleton extraction and Hough circle detection. Each step is carefully designed to ensure that the subtle changes in eye movement can be accurately captured.

[0088] The data generation unit 14 converts the extracted features into training sets and test sets that can be used for machine learning. This process involves operations such as feature selection and data normalization. For example, for the feature of eye movement speed, it can be normalized to a range of 0-1 to facilitate subsequent model training.

[0089] In one embodiment of the present invention, the top-hat transform denoising step in the feature extraction unit 13 adopts the following formula:

[0090]

[0091] Among them, A is the grayscale image of the image at time t, u t is the top-hat transformation mask at time t, γ is the opening operation, is a morphological opening operation. The advantage of the top hat transform is that it can effectively remove background noise while retaining the detailed features of eye movement. t The size of is usually chosen to be 1 / 10 to 1 / 20 of the image size, which can achieve a good balance between denoising and preserving details.

[0092] In one embodiment of the present invention, the Hough circle detection step is formulated as follows:

[0093] (x) 2 +(yb) 2 =r 2

[0094] Wherein, (a, b) is the coordinate of the center of the circle, r is the radius of the circle, and (x, y) is the coordinate of any point on the circle. Hough circle detection can effectively identify the position and size of the pupil. In children's vision screening, the present invention usually sets the detection radius range to 2-5 mm, which is consistent with the variation range of normal children's pupil size.

[0095] In one embodiment of the present invention, the AR test module 2 includes an eye movement enhancement unit 21, an eye movement recognition unit 22, and a vision test unit 23. The coordinated work of these three units makes the vision test process more accurate and interesting.

[0096] The eye movement enhancement unit 21 uses deep learning technology to enhance the eye movement features. Specifically, a convolutional neural network (CNN) can be used to process the eye image to extract more robust features. In practice, the present invention finds that the use of a deep network structure similar to ResNet, combined with appropriate data enhancement techniques (such as random rotation, scaling, etc.), can significantly improve the quality of feature extraction.

[0097] The eye movement recognition unit 22 compares the enhanced eye movement with the pre-prepared training set to generate a prediction model. A machine learning algorithm such as a support vector machine (SVM) or a random forest can be used here. Experience shows that for high-dimensional features such as eye movement, SVM with RBF kernel can usually achieve better classification results.

[0098] The vision testing unit 23 simulates the vision test scene through AR technology and provides test instructions. For example, the system can present the letter "E" of different sizes and directions in the AR environment, and ask the child to indicate the direction of the letter opening through eye movement. This method can not only accurately test vision, but also improve children's participation. Preferably, the test should start from 0.1 on the vision chart and gradually increase to 1.5 or higher to comprehensively evaluate the child's vision condition.

[0099] In general, the children's vision screening and tracking system of the present invention provides an innovative solution for children's vision health management by integrating computer vision, augmented reality and artificial intelligence technologies. The system can not only accurately collect and analyze eye movement data, but also improve children's cooperation through vivid and interesting AR tests. In addition, the system's prediction function and personalized suggestions can help parents and doctors detect and intervene in potential vision problems in a timely manner, which is of great significance for preventing children from myopia.

[0100] In one embodiment of the present invention, the corrosion processing step in the eye movement enhancement unit 21 adopts the following formula:

[0101]

[0102] Among them, A is the eye movement binary image, B is the erosion operator kernel, x is the coordinate of the point to be eroded, B(-x) is the kernel with the origin at x, and E(A,B) is the erosion result. Erosion operation plays a vital role in image processing, especially in processing eye movement images. It can effectively remove small noise points in the image and make the eye contour clearer.

[0103] In practical applications, the choice of the erosion operator kernel B directly affects the processing effect. The present invention preferably uses a 3x3 square kernel, which can remove most of the noise points while retaining the main features. For some special cases, such as when there are eyelashes around the eyes, a 5x5 cross kernel can be considered to obtain a better denoising effect. The erosion operation is usually used in conjunction with the dilation operation to form an opening operation or a closing operation to further optimize the image quality.

[0104] In one embodiment of the present invention, the core components of the vision development prediction module 3 include a diopter prediction unit 31 and an axial length prediction unit 32. These two units respectively use different algorithms to predict key indicators of children's vision development.

[0105] The diopter prediction unit 31 uses a multiple linear regression algorithm. This algorithm is selected because the change in diopter is usually linearly related to multiple factors. For example, age, eye habits, outdoor activity time, etc. may all have an impact on diopter. Multiple linear regression can effectively capture the complex relationship between these factors, thereby giving a more accurate prediction.

[0106] The axial length prediction unit 32 uses a long short-term memory network (LSTM) algorithm. The change in axial length has obvious time series characteristics, especially in the rapid growth period of children, and its change may show a nonlinear trend. As a special recurrent neural network, LSTM can effectively handle this long-term dependency, so it is particularly suitable for predicting axial length.

[0107] In one embodiment of the present invention, the multivariate linear regression formula used by the diopter prediction unit 31 is as follows:

[0108] y=μ+δ0+δ1x1+δ2x2+δ3x3+δ4x4+δ5x5

[0109] Among them, y is the predicted refractive power, μ is the intercept, δ0, δ1, δ2, δ3, δ4, δ5 are regression coefficients, and x1, x2, x3, x4, x5 are influencing factors such as axial length, current refractive power, amblyopia, myopia and astigmatism, respectively.

[0110] In practical applications, the selection and weight adjustment of these influencing factors are crucial. For example, for children aged 6-12, the axial length x1 and current refractive power x2 usually have a larger weight, because this age group is the critical period for the development of myopia. For younger children, such as 3-5 years old, the amblyopia factor x3 may need to be given a higher weight, because this is the golden period for amblyopia treatment.

[0111] The system will continuously collect data in actual operation and automatically adjust these coefficients through machine learning algorithms to adapt to the characteristics of children in different regions and age groups. Preferably, the system will conduct a comprehensive model update every 3-6 months to ensure the accuracy of the prediction.

[0112] In one embodiment of the present invention, the axial length prediction unit 32 adopts a long short-term memory network (LSTM) algorithm, and its core formula is as follows:

[0113] f t =o(W f ·[h t-1 ,x t ]+b f )

[0114] i t=σ(W i ·[h t-1 ,x t ]+b i )

[0115] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0116] c t =f t *c t-1 +i t *tanh(W c ·[h t-1 ,x t ]+b c )

[0117] h t =o t *tanh(c t )

[0118] Among them, f t ,i t , o t They are forget gate, input gate and output gate respectively, c t is the unit state, h t is the hidden state, x t is the input, W f , W i , W o , W c is the weight matrix, b f , b i , b o , b c is the bias vector, σ is the sigmoid function, and tanh is the hyperbolic tangent function.

[0119] The advantage of LSTM is that it can effectively handle long-term dependencies. This feature is particularly important in the prediction of axial length. For example, changes in children's axial length in the preschool period (3-6 years old) may affect the development of vision in the school age (7-12 years old). Traditional regression models have difficulty capturing this long-term impact, while LSTM can handle this problem well.

[0120] In practical applications, the present invention needs to carefully select the hyperparameters of LSTM. For example, the number of hidden layers is usually set to 2-3 layers, and the number of neurons in each layer can be adjusted between 64-256. Regarding the selection of the time step, considering that children's vision examinations are usually conducted every 3-6 months, the present invention can set each time step to 3 months and use the data of the past 2-3 years as the input sequence. This can capture long-term trends without introducing too much noise.

[0121] In one embodiment of the present invention, a computer vision-based children's vision screening and tracking method is also disclosed, which includes the following steps: collecting eye movement data, data preprocessing and feature extraction, eye movement enhancement and vision testing, vision development prediction, providing personalized health advice, data uploading and statistical analysis.

[0122] In the step of collecting eye movement data, the system uses a high-speed camera to capture the child's eye movements. Preferably, the sampling rate should reach at least 240Hz to ensure that rapid eye movements such as saccades can be captured. At the same time, in order to obtain stable image quality, it is recommended to test in an environment with sufficient light but not glaring, and the ideal light intensity should be between 300-500 lux.

[0123] In the data preprocessing and feature extraction steps, the system first grayscales and filters the original image. Grayscale can reduce the complexity of the data, while filtering can remove noise in the image. Next, the system will apply techniques such as top hat transformation and edge detection to extract key features of eye movement. These features may include the position, size, movement speed and direction of the pupil.

[0124] In the eye movement enhancement and vision test steps, the system uses deep learning technology to enhance the extracted features and then presents the vision test content in the AR environment. For example, the system may display the letter "E" of different sizes and directions on the screen of AR glasses or smart devices, and ask children to indicate the direction of the letter opening through eye movement. This method can not only accurately test vision, but also improve children's participation.

[0125] The vision development prediction step is the core of the whole method. Based on historical data and current test results, the system uses algorithms such as multivariate linear regression and LSTM to predict the future vision development trend of children. This includes the prediction of refraction and axial length. The accuracy of the prediction results is directly related to the effectiveness of subsequent intervention measures.

[0126] Based on the prediction results, the system will provide personalized health recommendations. These recommendations may include adjusting eye habits, increasing outdoor activity time, doing specific eye exercises, etc. Preferably, these recommendations should take into account factors such as the child's age and living environment to ensure their feasibility and effectiveness.

[0127] Finally, the system will upload the test data to the backend for statistical analysis. This not only helps track the vision development of individual children, but also provides strong support for public health decision-making. For example, by analyzing the vision development trend of children in a certain area or school, data support can be provided for the formulation of targeted vision protection policies.

[0128] In general, the children's vision screening and tracking method proposed in the present invention provides a comprehensive, accurate and efficient solution for children's vision health management by combining computer vision, augmented reality and artificial intelligence technology. This method can not only detect potential vision problems in a timely manner, but also effectively prevent the occurrence and development of myopia in children through prediction and personalized suggestions. In the context of the increasingly serious myopia at a younger age, the present invention has important practical significance and application value.

[0129] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto; any person familiar with the art who, within the scope disclosed by the present invention, makes substitutions or changes based on the scheme and improved concepts of the present invention shall be covered within the protection scope of the present invention.

Claims

1. A children's vision screening and tracking system based on computer vision, characterized in that: include: Eye movement analysis module, used to collect and analyze eye movement data; AR test module for eye movement enhancement and vision testing; Vision development prediction module, used to predict vision development trend; Vision knowledge base, used to provide professional knowledge support; Smart assistant module for providing personalized suggestions; System backend, used for data upload and statistical analysis; Among them, the eye movement analysis module, AR test module, vision development prediction module, vision knowledge base, intelligent assistant module and system background form a closed-loop vision management system through data interaction.

2. The system according to claim 1, characterized in that The eye movement analysis module comprises: An image acquisition unit, used for acquiring eye movement images; An image preprocessing unit, used for graying and filtering the collected image; A feature extraction unit, used for extracting eye movement features from the preprocessed image; A data generation unit, used for generating a training set and a test set based on the extracted features; The feature extraction unit extracts eye movement features through top hat transformation denoising, edge detection, skeleton extraction and Hough circle detection steps.

3. The system according to claim 2, characterized in that The top hat transform denoising step in the feature extraction unit adopts the following formula: Among them, A is the grayscale image of the image at time t, u t is the top-hat transformation mask at time t, γ is the opening operation, and ° is the morphological opening operation.

4. The system according to claim 2, characterized in that The Hough circle detection step in the feature extraction unit adopts the following formula: (x-a) 2 +(y-b) 2 =r 2 Among them, (a, b) are the coordinates of the center of the circle, r is the radius of the circle, and (x, y) are the coordinates of any point on the circle.

5. The system according to claim 1, characterized in that The AR test module includes: An eye movement enhancement unit, used to enhance eye movement features using deep learning technology; An eye movement recognition unit, used to compare the enhanced eye movements with the training set to generate a prediction model; Vision testing unit, used to simulate vision testing scenarios through AR technology and provide testing instructions; The eye movement enhancement unit performs binarization, filling and corrosion processing on the eye movement data.

6. The system according to claim 5, characterized in that The corrosion processing step in the eye movement enhancement unit adopts the following formula: Among them, A is the eye movement binary map, B is the erosion operator kernel, x is the coordinate of the point to be eroded, B(-x) is the kernel with the origin at x, and E(A, B) is the erosion result.

7. The system according to claim 1, characterized in that The vision development prediction module includes: A diopter prediction unit, used for predicting diopter; An axial length prediction unit, used for predicting the axial length; Among them, the refractive power prediction unit adopts a multivariate linear regression algorithm, and the axial length prediction unit adopts a long short-term memory network algorithm.

8. The system according to claim 7, characterized in that The diopter prediction unit uses the following multivariate linear regression formula: y=μ+δ0+δ1x1+δ2x2+δ3x3+δ4x4+δ5x5, where y is the predicted refractive power, μ is the intercept, δ0, δ1, δ2, δ3, δ4, δ5 are regression coefficients, and x0, x1, x2, x3, x4, x5 are the influencing factors of axial length, current refractive power, amblyopia, myopia and astigmatism respectively.

9. The system according to claim 7, characterized in that The axial length prediction unit adopts the following long short-term memory network formula: f t =σ(W f ·[h t-1 ,x t ]+b f ) i t =σ(W i ·[h t-1 ,x t ]+b i ) the t =σ(W o ·[h t-1 ,x t ]+b o ) c t =f t *c t-1 +i t *tanh(W c ·[h t-1 ,x t ]+b c ) h t =o t *tanh(c t ) Among them, f t ,i t , o t They are forget gate, input gate and output gate respectively, c t is the unit state, h t is the hidden state, x t is the input, W f , W i , W o , W c is the weight matrix, b f , b i , b o , b c is the bias vector, σ is the sigmoid function, and tanh is the hyperbolic tangent function.

10. A method for screening and tracking children's vision based on computer vision, characterized in that: The following steps are involved: Collect eye movement data; Preprocess and extract features of the collected data; Eye movement enhancement and vision testing based on the extracted features; Use predictive models to predict vision development trends; Provide personalized health recommendations based on prediction results; Upload the test data to the system backend for statistical analysis; Among them, the feature extraction, eye movement enhancement, and vision development prediction all adopt the corresponding algorithms.

Images