Eye protection method and system for intelligent terminal

Abstract

The invention discloses an eye protection method and system of an intelligent terminal, relates to the technical field of eye protection terminals, and aims to solve the problem of child eye use safety monitoring, multi-source data such as child face images, eye distances, equipment postures and environmental stability are collected, and time sequence data in a unified format is formed through noise filtering and synchronous processing; the posture angle and equipment stability information is extracted through the visual recognition and motion recognition sub-module, and the unhealthy eye using state is comprehensively judged; dynamically updating an eye using distance and a posture threshold value according to the height and the vision condition of the child and parent feedback; real-time monitoring is carried out, friendly reminding is triggered, and feedback data are recorded for closed-loop optimization; through cloud management, a parent-side App can check data in real time, a protection mode can be remotely started, and personalized eye protection suggestions are provided based on big data analysis. Through real-time remote intervention, accurate monitoring and personalized management of eye using behaviors of children are remarkably improved, and the risk of myopia is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of eye protection terminals, and particularly to an eye protection method and system for intelligent terminals. Background Art

[0002] With the popularization of mobile Internet, children increasingly rely on various intelligent terminals (such as tablet computers, smart phones, and e-readers, etc.) in their daily life and study. In a home scenario, children often lie on the table and stare at the screen at a close distance, or lie on the sofa or bed and watch the device at a large elevation angle; in a moving occasion such as a school bus or a private car, the terminal is in a bumpy environment, and children may unconsciously get closer to the screen or adopt an improper sitting posture during the shaking of the vehicle body. Since the visual system of children has not yet fully developed and their attention is easily distracted, and parents cannot supervise all the time, if intelligent terminals are used at an overly close distance or in an incorrect posture for a long time, it is extremely easy to cause eye fatigue and vision damage, and there are health risks such as deepening myopia or cervical fatigue. On the other hand, although each terminal manufacturer has successively launched eye protection functions, most of them focus on blue light filtering or simple distance reminders, which are not sufficient to adapt to the usage scenarios where children's postures change frequently, the environment shakes, or there are large individual differences, resulting in the eye protection effect being difficult to obtain reliable guarantee.

[0003] In the Chinese invention patent with the application publication number of CN104052871A, it relates to a mobile terminal eye protection device and method, which consists of three parts: a distance measurement unit, a central processing unit, and a warning and prompting unit. Its working principle is: when the eye protection function of the mobile terminal is turned on, the distance measurement unit measures the distance between the user and the mobile terminal in real time and transmits the measurement result to the central processing unit. When the distance between the user and the mobile terminal is always less than the set distance threshold within the set time threshold, the warning and prompting unit sends a warning and prompting message to the user to prompt the user to maintain a healthy viewing distance until the distance between the user and the mobile terminal is greater than the set distance threshold again, and the mobile terminal resumes normal display, and the warning and prompting message disappears, so as to achieve maintaining a healthy viewing distance between the user and the mobile terminal and achieve the purpose of preventing myopia and protecting the eyes.

[0004] Given that the eye - using scenarios of children are complex and highly individualized, conventional methods that solely rely on the front - facing camera to detect the face or simply use a fixed - distance threshold for simple reminders are difficult to identify various unhealthy eye - using behaviors promptly and accurately. Specifically: when children quickly change their postures or are in a bumpy environment while riding in a vehicle, using only single - vision detection is likely to result in the loss of facial key points or distorted distance data; if the threshold strategy does not combine individual characteristics such as children's height and eyesight, it may also cause frequent false alarms or missed alarms, thus undermining the reminder effect and even causing user resistance. In addition, for some parents who hope to remotely view and intervene in their children's eye - using situations, the local reminder mode of existing terminals also lacks unified data management and personalized cloud - based analysis support, further restricting large - scale promotion and application in different devices or multi - user scenarios. In summary, a technical solution with higher flexibility and scalability in detection accuracy, diversity of posture recognition, and personalized threshold regulation is needed to more comprehensively ensure the eye - using safety of children in complex environments.

[0005] To this end, the present invention provides an eye - protection method and system for intelligent terminals. Summary of the Invention

[0006] (I) Technical Problems to be Solved Aiming at the deficiencies of the prior art, the present invention provides an eye - protection method and system for intelligent terminals. By collecting multi - source data such as children's facial images, inter - eye distances, device postures, and environmental stability, and through noise filtering and synchronization processing, time - series data in a unified format is formed; through the visual recognition and motion recognition sub - modules, posture angles and device stability information are extracted to comprehensively judge unhealthy eye - using states; according to children's height, eyesight conditions, and parents' feedback, the eye - using distance and posture thresholds are dynamically updated; real - time monitoring and triggering of friendly reminders are carried out, and feedback data is recorded for closed - loop optimization; through cloud management, the parent - side App can view data in real time, remotely enable the protection mode, and provide personalized eye - protection suggestions based on big - data analysis. Through real - time remote intervention, the accurate monitoring and personalized management of children's eye - using behaviors are significantly improved, the risk of myopia is effectively reduced, and the technical problems raised in the background art are solved.

[0007] (II) Technical Solutions To achieve the above objectives, the present invention is implemented through the following technical solutions: An eye - protection method for intelligent terminals includes: when the terminal starts or enters the usage stage, raw state data is collected through a sensor array composed of a front - facing camera, a depth / distance sensor, and an accelerometer and a gyroscope, and multi - source alignment and noise filtering are performed on it to eliminate the interference of low - light shooting jitter and sensor drift. At the same time, the sampling frequency and interpolation mark are added to the meta - information to form multi - source time - series data; Based on the facial position data Extract head key points and calculate pose angles , based on the device pose data and environmental stability data Judge the terminal tilt and environmental stability. If it is found that the indicators of excessive head-down or too close distance exceed the limit, immediately conduct a risk judgment, so as to detect the distance data Whether it exceeds the safe range and mark the unhealthy eye use state, and output the over-limit information; After obtaining the unhealthy eye use state and the corresponding over-limit information, the adaptive algorithm reads the child height and vision difference parameters and the parent feedback correction amount , and update the alarm threshold and , when a high-risk pose is frequently detected, automatically tighten the distance alarm limit and generate a new strategy, adapt to different usage habits and reduce false alarms, and output personalized update results; Real-time monitor the status feedback data. If any of them exceeds the limit, trigger an animated pop-up window or voice reminder, and call the reminder cooling function to control the reminder frequency. If it exceeds the limit multiple times in a short period of time, increase the cooling decay weight. Then record the user's adjustment behavior and parent feedback and other information at the reminder moment to be passed back to the adaptive algorithm, and continuously monitor and correct to form a closed loop; When the reminder record is completed and the usage log is generated, upload the relevant recognition results, threshold evolution, and parent feedback to the cloud. The parent end can accordingly present charts such as eye use duration and unhealthy eye use times in real time and can remotely enable the forced protection mode, conduct risk analysis on children of different ages and vision conditions, identify high-risk eye use groups, and send the group report and personalized suggestions back to the terminal to continuously optimize the eye protection strategy.

[0008] Preferably, the original state data is collected by the sensor array, including face position data , distance data , device pose data , environmental stability data ; Adopt a unified sampling clock mechanism to perform corresponding matching at the same moment t; Preferably, after obtaining the original state data, perform denoising and unified format time series alignment, and output aggregated data as the input data for the next pose and distance recognition, where: Introduce a weighted Minkowski fusion model to perform multi-dimensional aggregation on different source data. In terms of time series alignment, the aggregated data is output according to the timestamp synchronized with the highest sampling frequency. If there is no new reading from the low-frequency sensor, the previous moment's valid value is used and an interpolation mark is added Preferably, according to the face position data extract the unit vector of the head orientation , according to the aggregated data Obtain the attitude of the device in three-dimensional space, and it can be converted into the unit vector of the front of the device , and construct the pose angle after combination Measure the relative inclination between the child's head and the device, and identify unhealthy eye-using postures; Preferably, read the preprocessed distance data , environmental stability data Define the comprehensive health function ; When the comprehensive health function exceeds the safety threshold , it is determined as an unhealthy eye-using state, and the state feature vector or the updated state feature vector can be further analyzed for specific components to locate the source of the anomaly, and the specific threshold-exceeding items are recorded according to the over-limit components (large pose angle, short distance, strong environmental vibration); when the comprehensive health function is not greater than the safety threshold , it is regarded as a normal eye-using state and no alarm is generated.

[0009] Preferably, read the child's height and vision condition , and map them to obtain the height offset and vision offset ; Aggregate the usage records for a recent period, including the average usage distance within the past week and the occurrence frequency of detected unhealthy postures ; Compare with the parent feedback information; finally integrate these historical behavior indicators with the parent feedback into the feedback correction amount ; Preferably, incorporate the correction amount into the adaptive correction function and output the distance alarm threshold and the posture deviation threshold ; If an unhealthy eye-using state is detected continuously for multiple times, a negative correction can be added to the feedback correction amount or the vision offset ; Preferably, map the new threshold to the executable configuration at the current time t, including the real-time distance threshold and the real-time posture threshold ; If the real-time distance is less than the real-time distance threshold or the pose angle is greater than the real-time posture threshold and other over-limit situations occur, it is marked as about to trigger a reminder and the mark is sent out. If no over-limit occurs, the normal monitoring state is maintained; Preferably, after receiving a reminder signal about to be triggered, if mild overlimit is detected for the first time, it is only prompted in the form of a small pop-up window animation or a slight vibration. If severe overlimit is repeatedly detected and lasts for a long time, it can be upgraded to voice reminder or screen masking; To avoid annoyance caused by overly frequent reminders, a reminder cooling function is constructed , where represents the time interval from the last reminder to the current moment. If is too small, new reminders are postponed or the reminder intensity is downgraded, Before generating a reminder, the reminder cooling function is queried first Value: If it is too low, the next reminder can be postponed or weakened to reduce user interference caused by continuous reminders; if it is relatively high, normal or enhanced reminder intensity is allowed to ensure the system's agile responsiveness; Preferably, an item record is generated after each reminder, including the timestamp when the reminder is triggered, the reminder intensity / method, such as a slight pop-up window, voice masking, etc., and the result of the detected or parent-confirmed user behavior change; If the parent actively marks the reminder as invalid or a false alarm, a feedback mark is added separately for subsequent correction; The above records are packaged and transmitted back to the adaptive threshold regulation so that the feedback correction amount is updated in the next time window and other correction amounts to adjust and and other personalized distance and posture thresholds in the next cycle; For frequently occurring scenarios where reminders are ineffective, similar reminders are gradually weakened or postponed; for cases where the reminder is successful and the user does cooperate, the current threshold is retained or slightly relaxed.

[0010] Preferably, according to the usage item records and personalized thresholds, such as and after being encrypted and packaged, they are uploaded to the cloud server; Enable the parent-side App to obtain the summary of eye usage data uploaded by the terminal in real time and visualize it on the interface; If the child maintains a high-risk posture for a long time or frequently triggers reminders, the forced protection mode is enabled in the parent-side App, and the trigger record of the forced protection mode is synchronously written to the cloud; Preferably, the historical records from different families or terminals are aggregated into a dataset Y, and each record can include fields such as the child's age, vision condition, historical threshold evolution path, reminder trigger frequency, and parent feedback; A high-dimensional feature mapping function is introduced in the cloud to perform non-linear mapping on each record and construct a feedback state vector ; for the mapped vector sequence Using clustering or association rule mining to automatically identify high-risk groups or special groups; generating a group eye use risk report and recommending usage duration and posture correction strategies; visually presenting the above analysis results in the cloud, including the probability distribution of unhealthy eye use in the group, the average risk index for different age / vision segments, etc.

[0011] An eye protection system for an intelligent terminal, including A data acquisition module that, when the terminal starts or enters the usage phase, collects raw state data through a sensor array composed of a front camera, depth / distance sensors, and an accelerometer and gyroscope, performs multi-source alignment and noise filtering on it to eliminate low-light shooting jitter and sensor drift interference, and at the same time attaches a sampling frequency and interpolation mark to the meta-information to form multi-source time-series data; A posture and distance recognition module based on facial position data Extracts head key points and calculates posture angles and, based on the device posture data and environmental stability data determines the tilt of the terminal and environmental stability. If it is found that the indicators of excessive head-down or too-close distance exceed the limit, a risk determination is immediately made, thereby detecting whether the distance data exceeds the safe range and marks the unhealthy eye use state, and outputs the over-limit information; An adaptive threshold adjustment module that, after obtaining the unhealthy eye use state and the corresponding over-limit information, reads the child's height and vision difference parameters and the parent feedback correction amount through an adaptive algorithm to update the alarm threshold and when a high-risk posture is frequently detected, automatically tightens the distance alarm limit and generates a new strategy, adapts to different usage habits and reduces false alarms, and outputs personalized update results; A reminder and feedback module that continuously monitors the status feedback data. If any over-limit occurs, it triggers an animated pop-up window or voice reminder, and calls a reminder cooling function to control the reminder frequency. If there are multiple over-limits in a short period of time, it increases the cooling decay weight. Subsequently, it records information such as the user's adjustment behavior and parent feedback at the reminder moment and transmits it back to the adaptive algorithm, continuously monitoring and correcting to form a closed loop; A remote monitoring module that, after completing the reminder record and generating a usage log, uploads the relevant recognition results, threshold evolution, and parent feedback to the cloud. The parent end can accordingly present charts such as eye use duration and the number of unhealthy eye use times in real time and can remotely enable a forced protection mode, perform risk analysis on children of different ages and vision conditions, identify high-risk eye use groups, and transmit the group report and personalized suggestions back to the terminal to continuously optimize the eye protection strategy.

[0012] (III) Beneficial effects The present invention provides an eye protection method and system for intelligent terminals, which have the following beneficial effects: Relying on the integrated visual recognition sub-module and motion recognition sub-module, elements such as head tilt and distance information are comprehensively processed to obtain the posture angle And accurately measure the screen distance to achieve timely capture of abnormalities such as excessive head-down or too close distance. This multi-modal recognition can still maintain a high accuracy in complex scenarios such as vehicle bumps or children's rapid posture changes, and generate marks of unhealthy eye use states; Introduce an adaptive algorithm, and combine factors such as children's height and vision conditions to adjust the distance alarm threshold And the posture deviation threshold For personalized dynamic regulation. When it is detected that a certain child frequently uses the eyes at a close distance, the threshold will be tightened in the next cycle: if parents report false alarms many times, the threshold will be moderately relaxed, which can ensure that individual differences are fully incorporated into the strategy and improve the adaptability to children of different ages or vision conditions.

[0013] According to the updated distance alarm threshold And the posture deviation threshold Perform real-time monitoring on the terminal side, and guide children to correct their postures through animated pop-up windows or voice reminders. To prevent excessive and frequent reminders from causing interference, a reminder cooling function is set, fully considering the aggregation effect of multiple triggers in a short period of time: each reminder will record the user's adjustment behavior and parental feedback at the reminder moment and send it back to the adaptive algorithm to form a complete closed loop, greatly enhancing the user-friendliness and sustainable improvement ability of the system; The recognition results and reminder logs on the terminal side are uploaded to the cloud. Parents can view the children's eye use data reports through the App or platform, and remotely enable the forced protection mode when necessary; Therefore, in this solution, multi-sensor fusion and high-precision posture determination can effectively reduce the false detection and missed detection rates. Second, the adaptive threshold and personalized strategy highlight differential protection. Third, the closed-loop reminder record and remote management improve the efficiency of parental monitoring and intervention, thereby making the monitoring and correction of children's eye health more comprehensive, accurate, and sustainable. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic flow chart of the eye protection method for the intelligent terminal of the present invention; Figure 2 It is a schematic structural diagram of the eye protection system for the intelligent terminal of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0015] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] Please refer to Figure 1 , the present invention provides an eye protection method for an intelligent terminal, including, Step 1: When the terminal starts or enters the usage stage, collect the original state data through the front camera, depth / distance sensor, and accelerometer and gyroscope to form a sensor array, perform multi-source alignment and noise filtering on it to eliminate low-light shooting jitter and sensor drift interference, and at the same time attach the sampling frequency and interpolation mark in the meta-information to form multi-source time-series data; The content of the above Step 1 includes the following: Step 101: Synchronous multi-source data collection Obtain the corresponding original state data from the following four types of sensors respectively: Front camera: Output the facial position data of the child , (including key point coordinates or contour information), the sampling interval is consistent with the video frame rate; Depth / distance sensor: Output the distance data between the child and the screen , recorded as a scalar or a shallow depth map; Accelerometer and gyroscope: Comprehensively output the device attitude data , which can be split into three-axis angular velocity and acceleration components; Environmental detection module (can be based on the fluctuation characteristics of the accelerometer gyroscope or an additional vibration sensor): Environmental stability data , used to indicate whether it is in a bumpy or violently shaking state; In the actual execution process, to ensure the timing consistency of the data, a unified sampling clock mechanism is adopted. If the sampling frequencies of the sensors are inconsistent, the highest frequency is used as the benchmark, and interpolation or caching with time stamps is performed on other sampling sources to make , , and corresponding matches can be made at the same moment t; if some sensors lose frames or cannot be read for a short time, mark the data for that period; Use the output by the environmental detection module to directly quantify external shaking or vibration, and evaluate the scene stability from the initial acquisition stage, providing a reference basis for subsequent recognition in complex environments (such as bumpy rides).

[0017] Step 102: Noise filtering and timing alignment After obtaining the original state data, perform denoising and time series alignment with unified format, and output aggregated data As the input data for the next step of pose and distance recognition, the specific processing flow is as follows: For the child's facial position data Execute the image stabilization algorithm to remove large-scale illumination jitter between video frames or instantaneous jumps of facial key points; for the distance data Adopt a fixed threshold to check for out-of-range or invalid depth readings (such as saturation values caused by being too close), and perform interpolation or marking; for the device pose data And the environmental stability data Adopt an inclination angle noise rejection strategy to smooth the instantaneous ultra-high frequency spike peaks; All filtering processes record the filtering ratio and attach it to the data meta-information for subsequent algorithm evaluation of data reliability; To obtain a comprehensive representation at the same time point t, introduce a weighted Minkowski fusion model (not simple average or weighted average) to perform multi-dimensional aggregation on different source data:

[0018] Where: 、 、 、 Are the fusion weights of each source data, and the value range is And:

[0019] In the formula: Represents the vector norm or scalar absolute value of the corresponding sensor output (if it is an image or key point vector, the Euclidean distance method can be used) to ensure unified measurement; Is the Minkowski coefficient, which is used to regulate the sensitivity of different source data to deviation values during fusion, and can take values greater than 2 to enhance the robustness to extreme value interference; The aggregated data obtained through the above formula Is a multi-dimensional synthesis quantity, and the characteristics of each sensor for time t are retained in its structure, and the influence of different source data on the overall determination is moderately amplified or weakened through weights and Minkowski coefficients; in terms of time series alignment, the aggregated data Is output according to the time stamp synchronized with the highest sampling frequency. If there is no new reading from the low-frequency sensor, the previous moment's valid value is used and an interpolation mark is added to ensure that a continuous and complete input sequence can be obtained in the next step of recognition.

[0020] During use, by simultaneously filtering and time-series aligning multi-source data, the misjudgment caused by light, occlusion, or instantaneous interference of a single sensor can be significantly reduced; interpolation marking of the outputs of each sensor helps to quickly identify the data credibility and perform dynamic threshold adjustment in the subsequent stage of unhealthy eye use state judgment; applying the weighted Minkowski fusion model to the preprocessing of multi-source data in the children's eye use scenario can improve the robustness to outliers while taking into account various information such as posture data, depth data, acceleration, and environmental stability; when performing time-series alignment, not only interpolate missing values.

[0021] Step 2. Based on the facial position data Extract head key points and calculate the posture angles , according to the device posture data and the environmental stability data Judge the tilt of the terminal and the environmental stability. If it is found that the indicators of excessive head-down or too close distance exceed the limit, immediately perform risk judgment, so as to detect the distance data whether it exceeds the safe range and mark the unhealthy eye use state, and output the over-limit information; The said Step 2 includes the following contents: Step 201. Visual posture extraction and motion posture analysis Receive the facial position data , the device posture data After that, focus on obtaining the core posture quantities related to the child's head and device orientation to quantify the degree of neck tilt or skew of the child; According to the facial key point information extracted from the facial position data , estimate the unit vector in the front of the child's face, denoted as the head orientation unit vector ; among them, through the relative positions of the facial key points, use the PnP algorithm or a deep learning model (such as a 2D / 3D pose estimation model) to estimate the head posture. These algorithms calculate the head posture angle by comparing the matching degree of the facial key points with the known model. According to the positions of the facial key points (such as eyes, nose, and chin), calculate the pitch, yaw, and roll angles of the child's head. These angle values can accurately represent the child's head orientation and are thus converted into a unit vector; If there are multiple key points (such as the center of the eyebrows, the tip of the nose, etc.), fit an orientation vector through a pre-constructed face geometric model and then normalize it to obtain the head orientation unit vector , if the aggregated data already contains facial direction information, the corresponding components can be directly read and converted into a unit vector; According to the aggregated data Obtain the posture of the device in three-dimensional space, and its posture can be converted into the unit vector of the front of the device ; If the front-facing direction of the device is consistent with the normal line of the camera, the unit vector of the device orientation can be obtained through quaternion or Euler angle analysis, and it is also necessary to ensure that it is a unit vector like the head vector for subsequent calculations; To measure the relative inclination between the child's head and the device, a posture angle is constructed , as follows:

[0022] Among them, represents the vector dot product operation, represents the vector norm, and the angle range returned by arccos is ; When the posture angle is too large (close to ), it means that the angle between the child and the device is too large (such as excessive head-down or head-up), which is likely to lead to unhealthy eye use postures; During use, through the precise vectorization of the unit vector of the head orientation and the unit vector of the device orientation , the relative inclination of the child's head can be estimated more stably in some scenes with insufficient light or local occlusion. The posture angle provides a core posture index for subsequent multi-modal comprehensive judgment, avoiding the rough scheme that relies on single-point detection or simple angle thresholds; using the angle between the two vectors of the head and the device as the posture measurement method can not only identify head-down or head-tilt, but also maintain good universality in multi-angle use scenarios (such as lying on the side or looking down).

[0023] Step 202, Distance Information Analysis and Unhealthy Eye Use State Judgment After obtaining the posture angle , it is necessary to combine key data such as distance data and environmental stability data to evaluate the distance between the child and the screen and the surrounding stability, and finally output whether there is an unhealthy eye use state and the corresponding over-limit information, and finally output the core result for the adaptive algorithm, that is, the unhealthy eye use state mark and its specific threshold over-limit items; Read the preprocessed distance data to obtain the instant distance between the child and the screen, and at the same time refer to the environmental stability data to understand the degree of environmental bumps or jitters. For example, this data may increase significantly in a car ride scenario; First, define the state feature vector including the following elements:

[0024] Among them, is the posture angle; is the distance data, is the safety reference distance; characterizes the comprehensive degree of environmental shaking in terms of numerical value (derived from accelerometers, gyroscopes or other sensors); as a whole is used to describe the instantaneous state of posture, distance and environment at the current moment t; On this basis, a symmetric positive definite matrix Q that performs weighted quadratic measurement on the state feature vector is introduced, as well as a short-term dynamic summation term, which is used to measure the impact of sharp changes in posture or distance over time, and the comprehensive health function is defined as follows:

[0025] To make the numerical values comparable and keep the dimensions consistent, corresponding weight factors can be placed at the diagonal elements of Q (such as , , amplification or scaling factors), and the cross elements can be set to 0 or small values according to actual needs; finally, by taking the square root, this value can be compressed into an interval closer to the dimension of the original component, which is convenient for subsequent additive operations with another term.

[0026] The Euclidean norm can be used to instantaneously measure the changes in the three-dimensional components; is the adjustment factor, which is used to control the influence weight of dynamic changes on the comprehensive health; is the time window; is a symmetric positive definite matrix to ensure that the quadratic form is non-negative; in specific implementations, the diagonal elements can be taken to be all >0, and the cross elements are moderately 0 or small values; When the comprehensive health function exceeds a certain safety threshold , it can be determined as an unhealthy eye use state, and the specific components of the state feature vector or the updated state feature vector can be further analyzed to locate the source of the anomaly, and the specific threshold exceeded items can be recorded according to the components exceeding the limit (large posture angle, short distance, strong environmental shaking); When the comprehensive health function is not greater than the safety threshold , it is regarded as a normal eye use state and no alarm is generated.

[0027] The comprehensive health function This multi-dimensional health metric can uniformly evaluate posture angle deviation, distance anomalies, and severe environmental vibrations, avoiding misjudgment of single variables and facilitating targeted handling.

[0028] Step 3: After obtaining the unhealthy eye use status and corresponding over-limit information, the adaptive algorithm reads the difference parameter between the child's height and vision and the correction amount feedback from the parents , and updates the alarm threshold and , when a high-risk posture is frequently detected, automatically tighten the distance alarm limit and generate a new strategy, adapt to different usage habits and reduce false alarms, and output personalized update results; The content of Step 3 is as follows: Step 301: Individual feature collection and historical usage analysis After receiving the unhealthy eye use status mark and its over-limit information, generate an individualized correction factor by integrating multiple information such as the child's height and vision status and historical posture data, preparing for the subsequent adaptive algorithm: Read the child's height and vision status (mark as the corresponding level if there is astigmatism or myopia tendency), and map it to obtain the height offset and vision offset ; A reference value can be set to represent the ideal height (or average height), and represent the ideal vision state, without astigmatism or myopia; If , then , indicating that the child is taller than the average height and the distance threshold can be appropriately increased; If the myopia degree is significantly shown, then , indicating that the distance or posture threshold may need to be tightened subsequently.

[0029] In this way, the differences in height and vision are quantified into scalars or small-range scores that can be superimposed and used in subsequent calculations; Aggregate the usage records for a recent period, including the average usage distance within the past week and the frequency of detected unhealthy postures ; compare with the parent feedback information (if the parent has marked a certain reminder as very necessary multiple times, it means that the sensitivity to this anomaly needs to be improved. If it is marked as a false alarm multiple times, the corresponding threshold can be appropriately relaxed); Finally, integrate these historical behavior indicators with the parent feedback into a feedback correction amount , representing the comprehensive correction amount of recent habit deviation or parent tolerance.

[0030] Through the height offset and the vision offset 、 ; Quantify the objective physiological differences and subjective usage preferences of children, and provide specific correction coefficients for the subsequent calculation of adaptive thresholds; automatically tighten the thresholds for children with existing myopia tendencies or those who are used to getting close to the screen, and can also moderately relax them when there are too many false alarms, mapping the parent feedback and historical posture data to the same feedback correction amount that can be superimposed or combined , which is more flexible and scalable than simply counting the number of times

[0031] Step 302, Adaptive Threshold Correction and Dynamic Output Let represent the distance alarm threshold, that is, if the distance between the child and the screen is lower than this value, a distance overrun alarm will be triggered; Let represent the posture deviation threshold, that is, if the posture angle (or other posture metrics) exceeds this value, a posture overrun alarm will be triggered: Let and be general recommended values, such as a distance of 30 cm and a maximum head deviation angle of a certain degree, which will be adaptively updated through the following formula later; Incorporate correction amounts such as height, vision, and historical usage into the adaptive correction function to enhance the adaptation ability to extreme cases and avoid excessive frequent oscillations. Define the new threshold calculation formulas as follows respectively:

[0032]

[0033] In the formula: , , and , , are weight parameters used to adjust the response sensitivity to each correction amount (can take positive or negative values according to specific tuning strategies); represents the height offset, represents the vision offset, represents the feedback correction amount; The finally output and are the personalized distance and posture thresholds for the next cycle; If an unhealthy eye - using state is detected continuously for multiple times, a negative correction (tighten the threshold) can be moderately increased in the feedback correction amount or the vision offset to correct the risk behavior as soon as possible; If parents frequently mark false alarms for reminders, it indicates that the threshold is too strict, and this information can also be accumulated into the feedback correction amount. Make a positive correction (relax the threshold) in it.

[0034] When in use, the exponential correction method is used to quickly adapt to special individual differences (such as children with extremely high / low height or highly myopic children) while ensuring the coherence of the threshold. Linked with the unhealthy eye use state and the feedback information from parents, it can dynamically optimize between the detection rate and the false alarm rate, improving the practicality and user acceptance; if parents think the reminder is too frequent, it will be appropriately relaxed to improve the overall adaptability. At the same time, the updated threshold will be read in real time by Step Four to form the judgment standard for the next usage cycle, and then further accept a new round of feedback in the closed loop for iterative optimization.

[0035] Step Four: Real-time monitor the status feedback data, including distance, posture, and environmental data. If any of them exceeds the limit, trigger an animated pop-up window or voice reminder, and call the reminder cooling function to control the reminder frequency. If it exceeds the limit multiple times in a short period, increase the cooling decay weight. Subsequently, record information such as the user's adjustment behavior and parents' feedback at the reminder moment and send it back to the adaptive algorithm for continuous monitoring and correction to form a closed loop. The above-mentioned Step Four includes the following contents: Step 401: Threshold reading and real-time monitoring Map the new threshold to the executable configuration at the current moment t, including the real-time distance threshold and the real-time posture threshold , considering that the threshold update times are different, align the real-time data with timestamps to ensure that distance values and posture angles and other indicators can be correctly applied to the corresponding thresholds; If the detected real-time distance is greater than the real-time distance threshold or the posture angle is greater than the real-time posture threshold and other situations of exceeding the limit occur, mark it as about to trigger a reminder and send out this mark. If no situation of exceeding the limit occurs, maintain the normal monitoring state; at the same time, the trend of exceeding the limit can be statistically analyzed in the background; When in use, through threshold reading and data synchronization, it can always ensure that the application side and the algorithm side are consistent after multiple threshold updates. The real-time threshold comparison mechanism realizes rapid detection and reduces missed reports, enabling the system to react immediately to abnormal eye use of children. Synchronously switch the personalized threshold in the time dimension of the timestamp, which can be compatible with multi-frequency sensor data and asynchronous update scenarios, and maximize the reduction of false alarms caused by delays or threshold misalignment.

[0036] Step 402: Generate friendly reminders After receiving the upcoming reminder signal, determine the form and intensity of the reminder based on factors such as the cumulative over-limit duration, historical reminder records, and parental preferences: If mild over-limit is detected for the first time, it is only prompted by a small pop-up animation or a slight vibration. If severe over-limit is repeatedly detected and lasts for a long time, it can be upgraded to more obvious intervention means such as voice reminder or screen masking; To avoid annoyance caused by overly frequent reminders, a reminder cooling function is constructed where represents the time interval from the last reminder to the current moment. If is too small, the new reminder is postponed or the reminder intensity is downgraded. The example is as follows: To make the reminder cooling function better reflect the cumulative impact of factors such as environmental shaking, abnormal posture / distance, and parental feedback in a short period, a weighted metric integrated over a time window is first defined as follows: Let the short-term cumulative anomaly vector represent the anomaly features accumulated from the time after the last reminder to the current moment, which may include the following components:

[0037] In the formula: can represent the difference amplitude between the distance and the personalized threshold (positive if the distance is too close); is the deviation of the posture angle from the threshold ; characterizes the degree of environmental shaking (such as vehicle bumpiness); is used to represent the parent's subjective reaction to the reminder during this period (such as marking false alarms or ineffectiveness multiple times); Introduce a symmetric positive definite matrix to separately adjust the weights and cross-influences of the above components on the cooling function; if some components do not need to be coupled, the corresponding non-diagonal elements can be set to 0 or a very small value; Integrate over the window to obtain a scalar measuring the cumulative anomaly intensity in the recent duration:

[0038] The larger is, the more and more serious the user's unhealthy eye use behaviors and adverse environmental factors are during this period; Based on this short-term cumulative anomaly intensity :

[0039] In the formula: , is the power exponent, used to increase the non-linear sensitivity to cumulative anomalies; , is the basic cooling coefficient, used to adjust the overall reminder strength benchmark; , the sensitivity factor, controls the sensitivity to , the larger the value, the more obvious the amplification effect on anomalies; Before generating a reminder, first query the reminder cooling function Value: If it is too low (indicating obvious short-term anomaly accumulation), the next reminder can be postponed or weakened to reduce user interference caused by continuous reminders; if it is higher, normal or enhanced reminder intensity is allowed to ensure the system's agility and responsiveness; The reminder content should be as concise and easy to understand as possible. For example, a cartoon character can guide children to look up or move away a little to reduce the rebellious psychology. For unstable environmental scenarios such as riding, if the environmental shaking index is higher than a certain value for a long time, it can also prompt to temporarily stop using or stabilize the device before watching.

[0040] When in use, hierarchical reminders can avoid frequent interruptions due to minor or instantaneous over-limit situations, and can also intervene in continuous high-risk situations in a timely manner; by mechanisms such as cooling functions, reminder bombing is reduced, and users' tolerance and cooperation with reminders are improved, realizing continuous adjustment of reminder frequency and intensity.

[0041] Step 403, reminder data recording and feedback backhaul An entry record is generated after each reminder, including the timestamp of the triggered reminder, reminder intensity / method, such as a slight pop-up window, voice mask, etc., and the result of the detected or parent-confirmed user behavior change (for example, whether the child sits up straight again, whether the distance is adjusted, etc.); If the parent actively marks the reminder as ineffective or a false alarm, a feedback mark is added separately for subsequent correction; Pack the above records and backhaul them to the adaptive threshold regulation, such as the dynamic adjustment of the distance alarm threshold and the posture deviation threshold, to adapt to the individual differences of different children, so that the feedback correction amount is updated in the next time window and other correction amounts, thereby adjusting and and the personalized distance and posture thresholds for the next cycle, etc. For frequently occurring reminder ineffective scenarios, according to the exponential correction mechanism of the dynamic adjustment of the distance alarm threshold and the posture deviation threshold in the previous step 302, similar reminders will be gradually weakened or postponed; for the situation where the reminder is successful and the user does cooperate, the current threshold will be retained or slightly relaxed to reduce the possibility of future false alarms.

[0042] During use, by recording the reminder time and response behavior in detail, it is possible to clearly judge which reminders are effective and which need improvement, and make full use of the actual situation of the user in subsequent adaptive threshold calculation; linked with parental feedback, it can quickly correct false alarms or over-reminders, improving the overall usability and friendliness of the system. The behavior information of children and parents after reminder triggering is sent back to Step 3 to form a complete closed-loop feedback, and this feedback is used to continuously improve the threshold setting, enhancing the adaptability to individual differences of minors in the next cycle; combined with the reminder cooling strategy and parental feedback mechanism, it can not only ensure timely guidance for children to maintain a healthy eye use posture at critical moments, but also effectively reduce the bad experiences brought by false alarms and frequent reminders, realizing a more personalized, accurate and acceptable children's eye protection monitoring solution.

[0043] Step Five: After completing the reminder record and generating the usage log, upload the relevant recognition results, threshold evolution and parental feedback to the cloud. Based on this, the parental end can present charts such as eye use duration and the number of unhealthy eye use times in real time and remotely enable the forced protection mode, conduct risk analysis on children of different ages and vision conditions, identify high-risk eye use groups, and send the group report and personalized suggestions back to the terminal to continuously optimize the eye protection strategy; The said Step Five includes the following contents: Step 501: Cloud data synchronization and parental end application According to the usage item records and personalized thresholds, such as and after encryption and packaging, upload them to the cloud server; to ensure data isolation between different children's accounts and devices, it is necessary to perform matching authentication of the unique device ID and parental account ID during the upload process: if there are multiple children sharing a family account, make a distinguishing field mark for different children in the data structure to ensure accurate subsequent analysis and visualization; Enable the parental end App to obtain the summary of eye use data uploaded by the terminal in real time and visually display charts such as the trend of use duration, statistics of the number of unhealthy eye use times, and average posture angle on the interface; if parents find that their children maintain a high-risk posture for a long time or frequently trigger reminders, they can immediately enable the forced protection mode in the parental end App, such as remotely reducing the device brightness or locking some entertainment functions, allowing children to take a short break or leave the screen; the trigger record of the forced protection mode will be synchronized and written into the cloud, and it can be incorporated into the children's eye use behavior model in subsequent data analysis; During use, through data synchronization and authentication, remote monitoring can be achieved while ensuring privacy and security. The forced protection mode of the parent-side App can intervene across time and space to meet the immediate control requirements for eye-using behaviors in actual family scenarios. The personalized threshold strategy and real-time reminder records are uniformly uploaded to the cloud for dynamic query or forced intervention by the parent-side App, breaking through the limitation that traditional single-end devices cannot be remotely operated and realizing more flexible cross-terminal collaborative management.

[0044] Step 502, Big data aggregation and group eye-using analysis Aggregate the historical records from different families or terminals into a data set Y. Each record may include fields such as children's age, vision status, historical threshold evolution path , reminder trigger frequency, and parent feedback, etc.; Introduce a high-dimensional feature mapping function in the cloud , perform non-linear mapping on each record to capture the complex interaction relationship between age - posture - feedback, and construct a feedback status vector , for example:

[0045] Among them, represents the original vector of the th record, can be a polynomial basis function, a kernel function, or other advanced mappings. The mapped feedback status vector is convenient for subsequent group clustering or risk prediction; For the mapped vector sequence , use clustering or association rule mining to automatically identify high-risk groups (such as those who frequently lower their heads and are insensitive to parent feedback) or special groups (such as those with a small age but ineffective tightening of thresholds); On this basis, multi-dimensional statistical indicators can be calculated to generate a group eye-using risk report and recommend more scientific usage duration and posture correction strategies. For example, define a clustering center vector c to represent the typical eye-using habits of a certain type of children, and compare the feedback status vector of a single child with the clustering center vector c to measure its deviation degree:

[0046] Among them, can be specified as the Minkowski distance or other advanced distance metrics (such as taking the norm after weighting with a symmetric positive definite matrix), those with too large Visualize the above analysis results in the cloud, including the probability distribution of the group's unhealthy eye use and the average risk index for different age / vision segments; according to specific clustering labels or risk distributions, recommend to parents the maximum daily usage duration, interval breaks, recommended distance, and posture alarm threshold adjustment plans to better adapt to the child's individual conditions and group patterns; parents can send these suggestions to the terminal again or manually enable the corresponding settings in the parent app to further optimize the child's eye protection strategy. First, collect the eye use behavior data of children (such as eye distance, posture, usage duration, etc.) through multiple sensors (such as cameras, accelerometers, gyroscopes, distance sensors, etc.), complete the data collection process, and perform synchronization, filtering, and formatting on these data.

[0047] Then, the group behavior analysis should include using data clustering methods (such as K-means, DBSCAN, etc.) to analyze the behavior patterns of children of different ages and vision conditions, and generate a health benchmark based on group behavior for each child. For example, for children in different age groups, the group data may show the average eye use duration and frequently occurring posture deviations of children in a certain age group, and the system provides a reference based on this.

[0048] For the personalized recommendation model, use machine learning algorithms (such as regression analysis, decision trees, neural networks, etc.) to process the child's personal data (such as height, vision, eye use history, etc.), and generate a recommendation plan based on these individual characteristics.

[0049] The dynamic threshold adjustment and parent feedback mechanism dynamically adjust the maximum usage duration, rest interval, distance, and posture alarm threshold according to real-time monitoring data, and parents provide feedback through the app to adjust and optimize the recommendation results.

[0050] Finally, continuously optimize the model and recommendation plan through a feedback loop (recording parent feedback and child behavior adjustments), so as to gradually improve the accuracy and personalization according to the actual situation.

[0051] When in use, through big data algorithms such as high-dimensional feature extraction + clustering / association mining, potential rules and differences in eye use behavior among different groups can be deeply discovered. The group eye use risk report enables parents to understand the comparison situation of their children with the peer group, and more targeted regulation such as eye use duration management and posture reminder settings can be carried out; the parent app can view in real time and remotely enable the forced protection mode, aggregate and analyze large-scale historical records in the cloud, form a group risk assessment and strategy recommendations, and then feedback to the parent app or the terminal, so as to incorporate richer group experience into the adaptive regulation and reminder strategy.

[0052] Please refer to Figure 2 , the present invention provides an eye protection system for an intelligent terminal, including, Data acquisition module: When the terminal starts or enters the usage phase, it collects raw state data through the sensor array composed of a front camera, depth / distance sensors, and an accelerometer and gyroscope, performs multi-source alignment and noise filtering on it to eliminate low-light shooting jitter and sensor drift interference, and at the same time attaches the sampling frequency and interpolation mark to the meta-information to form multi-source time-series data; Posture and distance recognition module: Based on the facial position data Extract head key points and calculate the posture angle , according to the device posture data And environmental stability data Judge the terminal tilt and environmental stability. If it is found that the indicators of excessive head-down or too close distance exceed the limit, immediately perform a risk judgment, thereby detecting the distance data Whether it exceeds the safe range and mark the unhealthy eye use state, and output the over-limit information; Adaptive threshold regulation module: After obtaining the unhealthy eye use state and the corresponding over-limit information, the adaptive algorithm reads the parameters of the height and vision differences of children and the correction amount of parents' feedback , update the alarm threshold And , when a high-risk posture is frequently detected, automatically tighten the distance alarm limit and generate a new strategy, adapt to different usage habits and reduce false alarms, and output personalized update results; Reminder and feedback module: Real-time monitor the status feedback data. If any of them exceeds the limit, trigger an animated pop-up window or voice reminder, and call the reminder cooling function to control the reminder frequency. If the limit is exceeded multiple times in a short period of time, increase the cooling decay weight. Subsequently, record the user's adjustment behavior and parents' feedback and other information at the reminder moment to be passed back to the adaptive algorithm, and continuously monitor and correct to form a closed loop; Remote monitoring module: When the reminder record is completed and the usage log is generated, upload the relevant recognition results, threshold evolution, and parents' feedback to the cloud. The parent end accordingly presents charts such as eye use duration and unhealthy eye use times in real time and can remotely enable the forced protection mode, perform risk analysis on children of different ages and vision conditions, identify high-risk eye use groups, and send the group report and personalized suggestions back to the terminal to continuously optimize the eye protection strategy.

[0053] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0054] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0055] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only for some logical function divisions. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0056] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0057] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application and should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An eye protection method for a smart terminal, characterized in that: include, After detecting that the terminal is started, multi-source sensors are called to collect raw status data. After noise filtering, interpolation marking and time series synchronization alignment, multi-source time series data in a unified format is generated; Extract key points of the head and calculate the posture angle, and determine the terminal's tilt and bump conditions. If the monitoring detection distance data exceeds the limit, mark the unhealthy eye status; Use an adaptive algorithm to read the height and vision difference parameters of children and combine them with the correction amount provided by parents to update the formula and adjust the corresponding alarm thresholds. If risky postures occur frequently, tighten the limits and output the latest personalized strategy. Compare the feedback data in real time to see if it exceeds the limit. If it is abnormal, trigger animation or voice reminder, call reminder cooling function to suppress high-frequency alarms, record user adjustment behavior and parent feedback at the reminder time and send them back; The recognition results and personalized threshold strategies are uploaded to the cloud, and parents can view eye usage statistics in real time and remotely enable forced protection mode, analyze eye risks of different age and vision groups, and send back improvement suggestions.

2. The eye protection method of a smart terminal according to claim 1, characterized in that: The sensor array collects and acquires raw state data, including facial position data, distance data, device posture data, and environmental stability data. After adopting a unified sampling clock mechanism, the collected data are matched at the same time. After acquiring the raw state data, denoising and timing alignment in a unified format are performed, and aggregated data is output.

3. The eye protection method of a smart terminal according to claim 2, characterized in that: The weighted Minkowski fusion model is introduced to aggregate different source data in multiple dimensions, and the aggregated data is output according to the timestamp synchronized with the highest sampling frequency; If there is no new reading of the low-frequency sensor, the valid value at the last moment is used and an interpolation mark is added.

4. The eye protection method of a smart terminal according to claim 1, characterized in that: The head orientation unit vector is extracted from the facial position data, and the unit vector of the front of the device is obtained from the aggregated data. The posture angle is constructed after combination, the relative inclination between the child's head and the device is measured, unhealthy eye postures are identified, and a comprehensive health function is constructed by reading the pre-processed distance data and environmental stability data.

5. The eye protection method of a smart terminal according to claim 4, characterized in that: When the comprehensive health function exceeds the safety threshold, it is judged as an unhealthy eye state; Analyze the specific components of the state feature vector or the updated state feature vector to locate the source of the abnormality, and record the specific threshold exceeding items according to the exceeding components; when the comprehensive health function is not greater than the safety threshold, it is regarded as a normal eye state and no alarm is generated.

6. The eye protection method of a smart terminal according to claim 5, characterized in that: Read the child's height and vision status, and map them to obtain height offset and vision offset; Aggregate usage records. Including, the average usage distance in the past week and the frequency of unhealthy postures detected; Compare with parents’ feedback information and integrate it into feedback correction amount.

7. The eye protection method of a smart terminal according to claim 6, characterized in that: The correction amount is incorporated into the adaptive correction function, and the distance alarm threshold and posture deviation threshold of the next cycle are output. If unhealthy eye conditions are detected for multiple consecutive times, a negative correction is added to the feedback correction amount or vision offset.

8. The eye protection method of a smart terminal according to claim 7, characterized in that: Map the new threshold to the executable configuration at the current moment. If an over-limit situation is detected, mark it as a reminder that will be triggered and send the mark. If no over-limit situation occurs, maintain the normal monitoring state. After receiving a warning signal that is about to be triggered, if a slight over-limit is detected for the first time, only a small pop-up animation or a slight vibration will be used as a reminder. If a serious over-limit is detected repeatedly and lasts for a long time, it will be upgraded to a voice reminder or screen mask.

9. The eye protection method of a smart terminal according to claim 8, characterized in that: By adjusting the reminder time and user adjustment behavior, the reminder frequency is adjusted in combination with the time decay strategy and user feedback, and a reminder cooling function is constructed; If the interval between the last reminder and the current moment is too short as expected, the new reminder will be postponed or the reminder intensity will be downgraded. Before generating a reminder, the reminder cooling function value will be queried. If it is lower than expected, the next reminder will be postponed or weakened; if it is higher, the reminder intensity will be normal or increased.

10. The eye protection method of a smart terminal according to claim 9, characterized in that: After each reminder, an entry record is generated, including the timestamp of the reminder triggering, the reminder intensity and method, and the result of the user behavior change is detected or confirmed; the entry record is packaged and sent back to the adaptive threshold control part, so that it updates the feedback correction amount and other correction amounts after the next round of time window, and adjusts the personalized distance and posture thresholds for the next cycle; For scenarios where reminders are frequently invalid, similar reminders will be gradually weakened or postponed; for situations where reminders are successful and the user does cooperate, the current threshold will be retained or slightly relaxed.

11. The eye protection method of a smart terminal according to claim 10, characterized in that: The usage item records and personalized thresholds are packaged and uploaded to the cloud server, so that parents can obtain the summary of eye usage data uploaded by the terminal in real time and display it visually in the interface; If the child maintains a high-risk posture for a long time or frequently triggers reminders, the forced protection mode is enabled on the parent side, and the trigger records of the forced protection mode are synchronously written to the cloud.

12. The eye protection method of a smart terminal according to claim 11, characterized in that: Aggregate historical records from different families or terminals into a data set, where each record contains the child's age, vision status, historical threshold evolution path, reminder trigger frequency, and parent feedback; A high-dimensional feature mapping function is introduced in the cloud to perform nonlinear mapping on each record and construct a feedback state vector. Clustering or association rule mining is applied to the mapped vector sequence to automatically identify high-risk groups or special groups, generate group eye risk reports, and recommend usage time and posture correction strategies.

13. An eye protection system for a smart terminal, characterized in that: include, The data acquisition module detects that the terminal is started, calls multi-source sensors to collect raw state data, and generates multi-source time series data in a unified format after noise filtering, interpolation marking and time series synchronization alignment; The posture and distance recognition module extracts key points of the head and calculates the posture angle, and determines the tilt and bump status of the terminal. If the monitoring detection distance data exceeds the limit, it marks the unhealthy eye status; The adaptive threshold control module uses an adaptive algorithm to read the parameters of children's height and vision differences and combines the correction amount provided by parents to update the formula and adjust the corresponding alarm thresholds. If risky postures are frequent, the limit is tightened and the latest personalized strategy is output; The reminder and feedback module compares the feedback data in real time to see if it exceeds the limit. If it is abnormal, it triggers animation or voice reminders, calls the reminder cooling function to suppress high-frequency alarms, and records the user's adjustment behavior and parent feedback at the reminder time and transmits them back; The remote monitoring module uploads the recognition results and personalized threshold strategies to the cloud. Parents can view eye usage statistics in real time and remotely enable forced protection mode, analyze the eye risks of different age and vision groups, and send back improvement suggestions.

Images