Eye protection method and system for an intelligent terminal

Abstract

The present invention discloses an eye protection method and system for intelligent terminals, which relates to the technical field of eye protection terminals. Aiming at the problem of difficult monitoring of children's eye use safety, multi-source data such as children's facial images, interpupillary distance, device posture, and environmental stability are collected, and time-series data in a unified format is formed through noise filtering and synchronization processing; posture angles and device stability information are extracted through visual recognition and motion recognition sub-modules to comprehensively judge unhealthy eye use states; according to children's height, vision conditions, and parents' feedback, the eye use 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, parents can view data in real time, remotely enable the protection mode through the parent app, 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 use behaviors are significantly improved, and the risk of myopia occurrence is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of eye protection terminals, and specifically 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 prone on the desktop and stare at the screen at a close distance, or lie on the sofa or bed and view the device at a large elevation angle; in mobile occasions such as school buses or private cars, 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 in addition, parents cannot supervise all the time, if intelligent terminals are used at too close a 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 of 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 have variable postures, environmental shaking, or 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 ranging 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 ranging 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 there are highly individualized differences, 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 during a ride, 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 take into account individual characteristics such as children's height and vision, 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 terms of 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] For this reason, the present invention provides an eye - protection method and system for intelligent terminals. Summary of the Invention

[0006] (I) Technical Problems to be Solved

[0007] In view of 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, eye distance, device posture, and environmental stability, and forming time - series data in a unified format through noise filtering and synchronization processing; extracting posture angles and device stability information through visual recognition and motion recognition sub - modules to comprehensively judge unhealthy eye - using states; dynamically updating eye - using distance and posture thresholds based on children's height, vision conditions, and parents' feedback; real - time monitoring and triggering friendly reminders, recording feedback data 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, it significantly improves the accurate monitoring and personalized management of children's eye - using behaviors, effectively reduces the risk of myopia, and solves the technical problems raised in the background art.

[0008] (II) Technical Solutions

[0009] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0010] An eye protection method for a smart terminal, including: when the terminal starts or enters the usage phase, collecting original state data through a sensor array composed of a front camera, a depth / distance sensor, and an accelerometer and a gyroscope, performing multi-source alignment and noise filtering on it to eliminate low-light shooting jitter and sensor drift interference, and at the same time attaching a sampling frequency and an interpolation mark to the meta-information to form multi-source time-series data;

[0011] Based on the facial position data Extract head key points and calculate the pose angle , according to the device attitude data And the environmental stability data Determine 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 a risk determination, 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;

[0012] 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 , 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;

[0013] Real-time monitor the status feedback data. If any of them exceeds the limit, trigger an animated pop-up window or a 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 parent feedback and other information at the reminder moment and send it back to the adaptive algorithm, continuously monitor and correct to form a closed loop;

[0014] When the reminder record is completed and a usage log is generated, upload the relevant recognition results, threshold evolution, and parent feedback to the cloud. The parent terminal can accordingly present charts such as eye use duration and the number of unhealthy eye uses 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.

[0015] Preferably, the original state data collected by the sensor array includes facial position data , distance data , device attitude data , environmental stability data ; adopt a unified sampling clock mechanism to perform corresponding matching at the same moment t;

[0016] Preferably, after obtaining the original state data, perform denoising and time series alignment of unified format, and output aggregated data As the input data for the next posture and distance recognition, where:

[0017] Introduce a weighted Minkowski fusion model to perform multi-dimensional aggregation on different source data. In terms of time series alignment, output the aggregated data According to the timestamp synchronized with the highest sampling frequency. If there is no new reading from the low-frequency sensor, use the effective value of the previous moment and append an interpolation mark

[0018] Preferably, according to the facial position data Extract the unit vector of the head orientation , and according to the aggregated data Obtain the posture of the device in three-dimensional space, and convert it into the unit vector of the front of the device , and construct a posture angle after combination Measure the relative inclination between the child's head and the device, and identify unhealthy eye use postures;

[0019] Preferably, read the preprocessed distance data , and the environmental stability data Define a comprehensive health function ; when the comprehensive health function Exceeds the safety threshold , it is determined as an unhealthy eye use state, and the state feature vector Or the updated state feature vector Of specific components to locate the source of the anomaly, and record the specific threshold exceedance items according to the over-limit components (large posture angle, too close distance, too 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.

[0020] Preferably, read the child's height And vision status , and map them to obtain the height offset And the vision offset ; 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; finally integrate these historical behavior indicators with the parent feedback into a feedback correction amount ;

[0021] Preferably, incorporate the correction amount into an adaptive correction function and output the distance alarm threshold for the next cycle and posture deviation threshold ;

[0022] 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 deviation amount ;

[0023] 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 real - time distance threshold or the posture angle real - time posture threshold and other over - limit situations occur, mark it as about to trigger a reminder and send out the mark. If no over - limit situation occurs, maintain the normal monitoring state;

[0024] Preferably, after receiving the signal about to trigger a reminder, if a mild over - limit is detected for the first time, only prompt it in the form of a small pop - up animation or a slight vibration. If a severe over - limit is repeatedly detected and lasts for a long time, it can be upgraded to a voice reminder or a screen mask;

[0025] To avoid annoyance caused by overly frequent reminders, construct a reminder cooling function , where represents the time interval from the last reminder to the current time. If is too small, postpone the new reminder or downgrade the reminder intensity,

[0026] Query the reminder cooling function value before generating a reminder

[0027] If it is too low, the next reminder can be postponed or weakened to reduce the interference to users caused by continuous reminders; if it is relatively high, allow normal or enhanced reminder intensity to ensure the agile responsiveness of the system;

[0028] Preferably, generate an item record after each reminder, including the timestamp when the reminder is triggered, the reminder intensity / way, such as a slight pop - up window, a voice mask, etc., and the result of the detected or parent - confirmed user behavior change;

[0029] If the parent actively marks the reminder as invalid or a false alarm, add a feedback mark separately for subsequent correction;

[0030] Pack the above records and send them back to the adaptive threshold regulation so that the feedback correction amount and other correction amounts are updated after the next time window , and the personalized distance and posture thresholds for the next cycle, such as and , are adjusted;

[0031] For frequently occurring ineffective reminder scenarios, gradually weaken or postpone similar reminders; for successful reminders and cases where the user truly cooperates, retain or slightly relax the current threshold.

[0032] Preferably, according to usage item records and personalized thresholds, such as and After encryption and packaging, upload to the cloud server;

[0033] Enable the parent - side App to obtain the aggregated eye - using data uploaded by the terminal in real - time and visually display it in the interface;

[0034] If the child maintains a high - risk posture for a long time or frequently triggers reminders, enable the forced protection mode in the parent - side App, and the trigger record of the forced protection mode will be synchronously written to the cloud;

[0035] Preferably, aggregate the historical records from different families or terminals into a data set Y. Each record can include fields such as children's age, vision status, historical threshold evolution path, reminder trigger frequency, and parent feedback, etc.

[0036] Introduce a high - dimensional feature mapping function in the cloud , perform non - linear mapping on each record and construct a feedback status vector ; for the mapped vector sequence Apply clustering or association rule mining to automatically identify high - risk populations or special groups; generate a group - based eye - using risk report and recommend usage duration and posture correction strategies; visually present the above - mentioned analysis results in the cloud, including the probability distribution of group unhealthy eye - using and the average risk index for different age / vision segments, etc.

[0037] An eye - protection system for an intelligent terminal, including,

[0038] A data acquisition module. When the terminal starts or enters the usage session, collect raw state data through a sensor array composed of a front - facing camera, depth / distance sensors, and an accelerometer and gyroscope. Perform multi - source alignment and noise filtering on it to eliminate low - light shooting jitter and sensor drift interference. At the same time, attach the sampling frequency and interpolation mark in the meta - information to form multi - source time - series data;

[0039] A posture and distance recognition module. Based on the facial position data Extract the head key points and calculate the posture angle , based on 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 determination, thereby detecting the distance data Determine whether it exceeds the safe range and mark the unhealthy eye - using state, and output the over - limit information;

[0040] The adaptive threshold regulation module, after obtaining the unhealthy eye - using state and the corresponding over - limit information, the adaptive algorithm reads the difference parameter between the child's height and vision and the correction amount of the parent's feedback to update the alarm threshold and When a high - risk posture is frequently detected, it 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;

[0041] The reminder and feedback module monitors the status feedback data in real - time. If any over - limit occurs, it triggers an animated pop - up window or voice reminder, and calls the reminder cooling function to control the reminder frequency. If there are multiple over - limits within a short period of time, it increases the cooling decay weight. Subsequently, it records information such as the user's adjustment behavior and the parent's feedback at the reminder time and transmits it back to the adaptive algorithm, continuously monitoring and correcting to form a closed - loop;

[0042] The remote monitoring module, after completing the reminder record and generating the usage log, uploads the relevant recognition results, threshold evolution, and parent feedback to the cloud. The parent - end accordingly presents charts such as eye - using duration and the number of unhealthy eye - using 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 - using groups, and transmit the group report and personalized suggestions back to the terminal to continuously optimize the eye - protection strategy.

[0043] (III)Beneficial effects

[0044] The present invention provides an eye - protection method and system for intelligent terminals, having the following beneficial effects:

[0045] Relying on the integrated visual recognition sub - module and motion recognition sub - module, comprehensively processes elements such as the head tilt angle and distance information to obtain the posture angle and accurately measures the screen distance, realizing the 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 rapid posture changes of children, and generate marks of unhealthy eye - using states;

[0046] Introduce an adaptive algorithm, and dynamically regulate the distance alarm threshold and the posture deviation threshold individually and dynamically in combination with factors such as the child's height and vision condition. When it is detected that a certain child frequently has short - distance eye - using, the threshold will be tightened in the next cycle: if the parent repeatedly reports false alarms, the threshold will be appropriately 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.

[0047] According to the updated distance alarm threshold and 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 within a short period: 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;

[0048] The recognition results and reminder logs on the terminal side are uploaded to the cloud. Parents can view the children's eye usage data reports through the App or platform, and remotely enable the forced protection mode when necessary;

[0049] 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 recording and remote management improve the efficiency of parental monitoring and intervention, thus making the monitoring and correction of children's eye health more comprehensive, accurate, and sustainable. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic flow diagram of the eye protection method of the intelligent terminal of the present invention;

[0051] Figure 2 It is a schematic structural diagram of the eye protection system of the intelligent terminal of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0052] 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 work shall fall within the protection scope of the present invention.

[0053] Please refer to Figure 1 , the present invention provides an eye protection method for an intelligent terminal, including,

[0054] Step 1: When the terminal starts or enters the usage link, 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 the interference of low-light shooting jitter and sensor drift, and at the same time attach the sampling frequency and interpolation mark to the meta-information to form multi-source time-series data;

[0055] The content of the above step 1 includes the following:

[0056] Step 101: Synchronously collect multi-source data

[0057] Obtain the corresponding original state data from the following four types of sensors respectively:

[0058] Front camera: Output the facial position data of the child , (including key point coordinates or contour information), and 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 decomposed into three-axis angular velocity and acceleration components; Environmental detection module (which 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;

[0059] During the actual execution process, to ensure the temporal 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 timestamp marking is performed on other sampling sources, so that 、 、 and can be correspondingly matched at the same moment t; If some sensors lose frames or cannot be read for a short time, the data for that period is marked; Utilize 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).

[0060] Step 102, Noise filtering and temporal alignment

[0061] After obtaining the original state data, perform denoising and temporal alignment of unified format, and output the aggregated data as the input data for the next pose and distance recognition. The specific processing flow is as follows:

[0062] Execute an image stabilization algorithm on the child's facial position data to remove large-scale illumination jitters or instantaneous jumps of facial key points between video frames; For the distance data , use 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 attitude data and the environmental stability data , adopt an inclination angle noise rejection strategy to smooth instantaneous ultra-high frequency spike values;

[0063] All filtering processes record the filtering ratio and attach it to the data meta-information for subsequent algorithm evaluation of data reliability;

[0064] To obtain a comprehensive representation at the same time point t, a weighted Minkowski fusion model (not a simple average or weighted average) is introduced to perform multi-dimensional aggregation on different source data:

[0065]

[0066] Among them: , , , are the fusion weights of each source data, and the value range is , and:

[0067]

[0068] 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;

[0069] 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;

[0070] The aggregated data obtained through the above formula is a multi-dimensional composite quantity. 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 the Minkowski coefficient; 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 effective value is used and an interpolation mark is added to ensure a continuous and complete input sequence can be obtained in the next step of recognition.

[0071] When in use, by simultaneously filtering and time series aligning multi-source data, the misjudgment caused by a single sensor due to light, occlusion, or instantaneous interference can be significantly reduced; interpolation marks are made on the outputs of each sensor, which 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 multi-party information such as posture data, depth data, acceleration, and environmental stability; when aligning time series, not only interpolating missing values.

[0072] Step 2. Based on the facial position data Extract head key points and calculate the pose angle , based on the device pose data and environmental stability data Determine the terminal's tilt and environmental stability. If the terminal is found to be too low or the distance is too close, the risk assessment will be carried out immediately to detect the distance data. Whether it exceeds the safety range and marks the unhealthy eye status, outputting the limit-exceeding information;

[0073] The second step includes the following:

[0074] Step 201: Visual posture extraction and motion posture analysis

[0075] Receive facial position data , device posture data Finally, focus on obtaining core posture measurements related to the child's head and equipment orientation to quantify the degree of the child's neck tilt or tilt;

[0076] Based on facial position data The facial key point information extracted from the is used to estimate the unit vector in front of the child’s face, which is recorded as the head direction unit vector. Among them, the head pose is estimated by using the relative positions of facial key points using a PnP algorithm or a deep learning model (such as a 2D / 3D pose estimation model). These algorithms calculate the head pose angle by comparing the degree of match between facial key points and known models. Based on the positions of facial key points (such as the eyes, nose, and chin), the pitch, yaw, and roll angles of the child's head are calculated. These angle values can accurately represent the orientation of the child's head and can be converted into unit vectors.

[0077] If there are multiple key points (such as the center of the eyebrows, the tip of the nose, etc.), a heading vector is fitted through the pre-built face geometry model, and then normalized to obtain the head heading unit vector If the aggregate data Since the facial direction information is already included in , the corresponding component can be directly read and converted into a unit vector;

[0078] Based on aggregated data Get the device's posture in three-dimensional space, which can be converted into a unit vector in front of the device If the device's front orientation is consistent with the camera normal, the device orientation unit vector can be obtained by quaternion or Euler angle analysis. This also needs to be ensured to be the same unit vector as the head vector for subsequent calculations.

[0079] To measure the relative tilt between the child's head and the device, a posture angle ,as follows:

[0080]

[0081] in, Represents vector dot multiplication operation, Represents the vector norm, the angle range returned by arccos is ;

[0082] When the posture angle Too large (close to ), it means that the angle between the child and the device is too large (for example, too much lowering or raising the head), which can easily lead to unhealthy eye posture;

[0083] When used, pass the head towards the unit vector and the device's orientation unit vector The precise vectorization can still stably estimate the relative tilt of the child's head and posture angle in scenes with insufficient light or partial occlusion. It provides core posture indicators for subsequent multimodal comprehensive judgment, avoiding crude solutions that rely on single-point detection or simple angle thresholds; using the angle between the head and the device as a posture measurement method can not only identify lowering or tilting the head, but also maintain good universality in multi-angle usage scenarios (such as lying on the side or looking down).

[0084] Step 202: Distance information analysis and unhealthy eye status determination

[0085] In getting the posture angle Finally, distance data and environmental stability data The system uses key data such as the distance between the child and the screen and the surrounding stability to evaluate whether there is unhealthy eye use and the corresponding limit violation information. Finally, it outputs the core results used by the adaptive algorithm, namely the unhealthy eye use status mark and its specific threshold violation items;

[0086] Read pre-processed distance data , obtain the instant distance between the child and the screen, and refer to the environmental stability data Understand the degree of environmental turbulence or vibration, such as the data may increase significantly in the vehicle scene;

[0087] First define the state eigenvector Contains the following elements:

[0088]

[0089] in, is the posture angle; is the distance data, is the safety reference distance;

[0090] A measure of the overall degree of environmental vibration (derived from accelerometers, gyroscopes, or other sensors). It is used to describe the instantaneous state of posture, distance, and environment at the current moment t;

[0091] On this basis, a symmetric positive definite matrix Q that performs a weighted quadratic measure on the state feature vector is introduced, as well as a short-term dynamic summation term used to measure the impact of sharp changes in posture or distance over time. The comprehensive health function is defined as follows:

[0092]

[0093] To make the numerical values comparable and keep the dimensions consistent, the 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, the value can be compressed into an interval closer to the dimension of the original component, facilitating subsequent addition operations with another term.

[0094] The Euclidean norm can be used to instantaneously measure the changes in three-dimensional components; is a regulation factor used to control the influence weight of dynamic changes on the comprehensive health;

[0095] 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;

[0096] 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 exceeding items can be recorded according to the components exceeding the limit (large posture angle, short distance, strong environmental shaking);

[0097] 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.

[0098] Using this multi-dimensional health measure of the comprehensive health function , a unified assessment can be made for the deviation of the posture angle, abnormal distance, and strong environmental shaking, avoiding misjudgment of single variables, so as to facilitate targeted processing.

[0099] 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 of parental feedback , 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;

[0100] The content of the said Step 3 includes the following:

[0101] Step 301: Individual feature collection and historical usage analysis

[0102] 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, and prepare for the subsequent adaptive algorithm:

[0103] Read the child's height and vision status (if there is astigmatism or myopia tendency, mark it as the corresponding level), and map it to obtain the height offset and vision offset ;

[0104] Let the reference value represent the ideal height (or average height), represent the ideal vision state, without astigmatism or myopia;

[0105] If , then , indicating that the child is taller than the average height, and the distance threshold can be appropriately increased;

[0106] If the displayed myopia degree is significant, then , indicating that the distance or posture threshold may need to be tightened subsequently.

[0107] In this way, the differences in height and vision are quantified into scalars or small-range fractions that can be superimposed and used in subsequent calculations;

[0108] 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 parental 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);

[0109] Finally, integrate these historical behavior indicators with parental feedback into the feedback correction amount , representing the comprehensive correction amount of recent habit deviation or parental tolerance.

[0110] 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 threshold when there are children with existing myopia tendencies or who are used to getting close to the screen, and can also relax moderately when there are too many false alarms, mapping the parental feedback and historical posture data to the same superimposable or combinable feedback correction amount , which is more flexible and scalable than simply counting the number of times.

[0111] Step 302, Adaptive Threshold Correction and Dynamic Output

[0112] Let represent the distance alarm threshold, that is, if the distance between the child and the screen is lower than this value, a distance overlimit alarm will be triggered;

[0113] Let represent the posture deviation threshold, that is, if the posture angle (or other posture metrics) exceeds this value, a posture overlimit 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;

[0114] Incorporate correction amounts such as height, vision, and historical usage into the adaptive correction function to enhance the adaptability to extreme cases and avoid excessive frequent oscillations. The new threshold calculation formulas are defined as follows:

[0115]

[0116]

[0117] 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);

[0118] represents the height offset, represents the vision offset, represents the feedback correction amount;

[0119] The finally output and , which is the personalized distance and posture threshold for the next cycle;

[0120] If an unhealthy eye - using state is detected continuously for multiple times, a negative correction (tightening the threshold) can be appropriately increased in the feedback correction amount or the vision deviation amount to correct the risk behavior as soon as possible;

[0121] If parents frequently mark false alarms for reminders, it indicates that the threshold is too strict. The information can also be accumulated in the feedback correction amount for positive correction (relaxing the threshold).

[0122] When in use, the exponential correction method is used to quickly adapt to special individual differences (such as children with extremely high / short height or highly myopic children) while ensuring the coherence of the threshold. Linked with the unhealthy eye - using state and parents' feedback information, it can dynamically optimize between the detection rate and the false - alarm rate, improving practicality and user acceptance. If parents think the reminders are too frequent, the threshold 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.

[0123] 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 the information such as the user's adjustment behavior and parents' feedback at the reminder moment and send it back to the adaptive algorithm to continuously monitor and correct to form a closed - loop;

[0124] The above - mentioned Step Four includes the following contents:

[0125] Step 401: Threshold reading and real - time monitoring

[0126] 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 . Considering different times for updating the threshold, align the real - time data with timestamps to ensure that indicators such as the distance value and the posture angle can be correctly applied to the corresponding thresholds;

[0127] If the detected real - time distance exceeds the real - time distance threshold or the posture angle exceeds the real - time posture threshold and other situations of exceeding the limit, mark it as about to trigger a reminder and send out the mark. If there is no situation of exceeding the limit, maintain the normal monitoring state; at the same time, the trend of exceeding the limit can be statistically analyzed in the background;

[0128] During use, through threshold reading and data synchronization, it can always ensure that the application side and the algorithm side remain consistent after multiple threshold updates. The real-time threshold comparison mechanism enables rapid detection and reduces false negatives, enabling the system to respond immediately to abnormal eye use in children. Synchronously switching the personalized threshold in the time stamp dimension can be compatible with multi-frequency sensor data and asynchronous update scenarios, maximizing the reduction of false positives caused by delays or threshold misalignment.

[0129] Step 402, generating friendly reminders

[0130] After receiving the signal that a reminder is about to be triggered, determine the form and intensity of the reminder based on factors such as the cumulative over-limit duration, historical reminder records, and parental preferences:

[0131] If it is detected as slightly over the limit for the first time, it is only prompted by a small pop-up animation or a slight vibration. If it is repeatedly detected as severely over the limit and the duration is long, it can be upgraded to more obvious intervention means such as voice reminders or screen masking;

[0132] To avoid annoyance caused by overly frequent reminders, construct a reminder cooling function , where represents the time interval from the last reminder to the current moment. If is too small, postpone the new reminder or downgrade the reminder intensity. The example is as follows:

[0133] To make the reminder cooling function better reflect the cumulative effects of factors such as environmental shaking, abnormal posture / distance, and parental feedback in a short period, first define a weighted metric integrated over a time window as follows:

[0134] Let the short-term cumulative anomaly vector represent the cumulative anomaly features from the time after the last reminder to the current moment, which may include the following components:

[0135]

[0136] In the formula: can represent the difference amplitude between the distance and the personalized threshold (positive if the distance is too close);

[0137] is the deviation of the posture angle from the threshold ; characterizes the degree of environmental shaking (such as vehicle bumps); is used to represent the parent's subjective reaction to the reminder during this period (such as marking false positives or ineffectiveness multiple times);

[0138] Introduce a symmetric positive definite matrix , used to separately adjust the weights and cross - impacts of the above - mentioned 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 minimum value;

[0139] Integrate over the window to obtain a scalar that measures the cumulative anomaly intensity in the recent duration: :

[0140]

[0141] The larger it is, the more and more serious the unhealthy eye - using behaviors and adverse environmental factors of the user are during this period;

[0142] Based on this short - term cumulative anomaly intensity , construct a reminder cooling function :

[0143]

[0144] In the formula: , is the power exponent, used to increase the non - linear sensitivity to the cumulative anomaly; , is the basic cooling coefficient, used to adjust the overall reminder intensity benchmark;

[0145] , the sensitivity factor, controls the sensitivity to , the larger the value of, the more obvious the amplification effect on the anomaly;

[0146] Before generating a reminder, first query the value of the reminder cooling function :

[0147] If it is too low (indicating obvious short - term anomaly accumulation), the next reminder can be postponed or weakened to reduce the interference to the user caused by continuous reminders; if it is relatively high, normal or enhanced reminder intensity is allowed to ensure the agile responsiveness of the system;

[0148] The reminder content should be as concise and easy to understand as possible, such as a cartoon character guiding a child to look up or move away a little to reduce the rebellious psychology. For unstable environmental scenarios such as in a vehicle, if the environmental shaking index is higher than a certain value for a long time, it can also be prompted to temporarily stop using or stabilize the device before watching.

[0149] When in use, hierarchical reminders can avoid frequent interruptions due to minor or instantaneous over - limits, and can also intervene in continuous high - risk situations in a timely manner; through mechanisms such as cooling functions, reminder bombing is reduced, and the user's tolerance and cooperation with reminders are improved, realizing continuous adjustment of reminder frequency and intensity.

[0150] Step 403, reminder data recording and feedback back - transmission

[0151] An entry record is generated after each reminder, including the timestamp of the reminder triggering, the reminder intensity / method, such as a gentle pop-up window or voice masking, and the results of the user behavior change detected or confirmed by the parent (for example, whether the child sat up again, adjusted the distance, etc.);

[0152] If the parent actively marks the reminder as invalid or a false alarm, a feedback mark will be added for subsequent correction;

[0153] The above records are packaged and sent back to the adaptive threshold control, 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 they can Post-update feedback correction and other corrections to adjust and Waiting for the personalized distance and posture thresholds of the next cycle, for scenarios where reminders are frequently invalid, the exponential correction mechanism of the distance alarm threshold and posture deviation threshold that are dynamically adjusted in the previous step 302 will be used to gradually weaken or postpone similar reminders; for situations where reminders are successful and the user does cooperate, the current thresholds will be retained or slightly relaxed to reduce the possibility of false alarms in the future.

[0154] During use, by recording reminder moments and response behaviors in detail, it is possible to clearly determine which reminders are effective and which need improvement, and fully utilize the user's actual situation in the subsequent adaptive threshold calculation; linked with parental feedback, it can achieve rapid correction of false alarms or excessive reminders, and improve the practicality and friendliness of the overall system. After the reminder is triggered, the behavioral information of the child and the parent is fed back to step three to form a complete closed-loop feedback. This feedback is used to continuously improve the interval setting and improve the adaptability to the individual differences of minors in the next cycle; combined with the reminder cooling strategy and parent feedback mechanism, it can ensure that children are guided to maintain healthy eye posture in a timely manner at critical moments, and effectively reduce the adverse experience caused by false alarms and high-frequency reminders, and realize a more personalized, accurate and easy-to-accept children's eye protection monitoring solution.

[0155] Step 5. After the reminder record is completed and the usage log is generated, the relevant recognition results, threshold evolution and parent feedback are uploaded to the cloud. Parents can use this data to display real-time charts such as eye usage time and the number of times of unhealthy eye use. They can also remotely enable forced protection mode, conduct risk analysis on children of different ages and vision conditions, identify high-risk groups, and transmit group reports and personalized suggestions back to the terminal to continuously optimize eye protection strategies.

[0156] The step five includes the following:

[0157] Step 501: Synchronize cloud data with parent-side applications

[0158] According to the usage entry records and personalized thresholds, such as and After encryption and packaging, it is uploaded 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 the parent account ID during the upload process: if multiple children share a family account, different children are marked with distinguishing fields in the data structure to ensure accurate subsequent analysis and visualization;

[0159] Enable the parent - side App to obtain the summary of eye - usage data uploaded by the terminal in real - time, and visually display charts such as the usage duration trend, the statistics of unhealthy eye - usage times, and the average posture angle in 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 parent - side App, such as remotely reducing the device brightness or locking some entertainment functions, to let children take a short break or leave the screen; the trigger records of the forced protection mode will be synchronously written to the cloud and can be incorporated into the children's eye - usage behavior model in subsequent data analysis;

[0160] 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, meeting the immediate control requirements for eye - usage behavior in actual family scenarios. The personalized threshold strategy and real - time reminder records are uniformly uploaded to the cloud for the parent - side App to dynamically query or force intervention, breaking through the limitation that traditional single - end devices cannot be remotely operated and realizing more flexible cross - terminal collaborative management.

[0161] Step 502, Big data aggregation and group - based eye - usage analysis

[0162] Aggregate the historical records from different families or terminals into a dataset Y. Each record can include fields such as children's age, vision condition, historical threshold evolution path , reminder trigger frequency, and parent feedback, etc.;

[0163] 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 state vector , for example:

[0164]

[0165] 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 state vector is convenient for subsequent group clustering or risk prediction;

[0166] For the mapped vector sequence Apply clustering or association rule mining to automatically identify high-risk groups (e.g., those who frequently lower their heads and whose parents' feedback is insensitive) or special groups (such as those with a relatively young age but ineffective tightening of thresholds);

[0167] On this basis, multi-dimensional statistical indicators can be calculated to generate a group eye use risk report and recommend more scientific usage duration and posture correction strategies. For example, define a cluster center vector c to represent the typical eye use habits of a certain type of children, and compare it with the feedback status vector of an individual child The distance from the cluster center vector c to measure the degree of deviation:

[0168]

[0169] Among them, It 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 values may belong to the high-risk category, which is used to guide subsequent personalized improvement suggestions;

[0170] Visualize the above analysis results in the cloud, including the probability distribution of the group's unhealthy eye use, the average risk index for different age / vision segments, etc.; according to specific cluster labels or risk distributions, recommend the maximum daily usage duration, interval breaks, recommended distance, and posture alarm threshold adjustment plans to parents to better adapt to the child's personal situation and group rules; parents can forward these suggestions to the terminal again or manually enable the corresponding settings in the parent-side App to further optimize the child's eye protection strategy. First, collect children's eye use behavior data (such as eye distance, posture, usage duration, etc.) through multiple sensors (such as cameras, accelerometers, gyroscopes, distance sensors, etc.) to complete the data collection process, and synchronize, filter, and format these data.

[0171] 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.

[0172] For the personalized recommendation model, use machine learning algorithms (such as regression analysis, decision trees, neural networks, etc.) to process children's personal data (such as height, vision, eye use history, etc.) and generate recommendation plans based on these individual characteristics.

[0173] The dynamic threshold adjustment and parental feedback mechanism dynamically adjust the maximum usage duration, rest intervals, distance, and posture alarm thresholds based on real-time monitoring data. Parents can provide feedback through the App to adjust and optimize the recommended results.

[0174] Finally, through the feedback loop (recording parental feedback and children's behavior adjustments), the model and recommendation scheme are continuously optimized, thereby gradually improving accuracy and personalization according to the actual situation.

[0175] When in use, through big data algorithms such as high-dimensional feature extraction + clustering / association mining, potential patterns and differences in eye usage behaviors among different groups can be deeply discovered. The group eye usage risk report enables parents to understand the comparison situation of their children with the peer group, and more targeted regulation such as eye usage duration management and posture reminder settings can be carried out; the parent side can view in real time and remotely enable the forced protection mode, perform aggregated analysis on a large-scale historical record in the cloud, form a group risk assessment and strategy recommendations, and then feedback to the parent side or the terminal, so as to incorporate richer group experience into the adaptive regulation and reminder strategy.

[0176] Please refer to Figure 2 , the present invention provides an eye protection system for an intelligent terminal, including,

[0177] A data acquisition module, when the terminal starts or enters the usage session, collects original state data through a sensor array composed of a front camera, a depth / distance sensor, and an accelerometer and a 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 an interpolation mark to the meta-information to form multi-source time-series data;

[0178] A posture and distance recognition module, based on the facial position data extracts the head key points and calculates the posture angles , based on the device posture data and the environmental stability data determines the terminal tilt 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 marking the unhealthy eye usage state, and outputting the exceeded limit information;

[0179] An adaptive threshold regulation module, after obtaining the unhealthy eye usage state and the corresponding exceeded limit information, the adaptive algorithm reads the child's height and vision difference parameters and the parental feedback correction amount , updates 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;

[0180] The reminder and feedback module monitors the status feedback data in real time. If any of them exceeds the limit, it triggers an animated pop-up window or a voice reminder, and calls the reminder cooling function to control the reminder frequency. If the limit is exceeded multiple times in a short period, the cooling decay weight is increased. Subsequently, it records the user's adjustment behavior and parental feedback at the reminder moment and transmits the information back to the adaptive algorithm for continuous monitoring and correction to form a closed loop;

[0181] The remote monitoring module uploads relevant recognition results, threshold evolution, and parental feedback to the cloud after completing the reminder record and generating the usage log. Based on this, the parental end presents charts such as eye usage duration and the number of unhealthy eye usages in real time and can remotely enable the forced protection mode. It conducts risk analysis on children of different ages and vision conditions, identifies high-risk eye-using groups, and transmits the group report and personalized suggestions back to the terminal to continuously optimize the eye protection strategy.

[0182] 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 herein 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.

[0183] Those skilled in the art can clearly understand that for the convenience and conciseness 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.

[0184] In several embodiments provided in this 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, and there can be other division methods in actual implementation. 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 mutual coupling or direct coupling or communication connection can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be in an electrical, mechanical, or other form.

[0185] The units described as separate components may or may not be physically separated, and 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.

[0186] The above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims described above.

Claims

1. An eye protection method for an intelligent terminal, characterized in that: including, After detecting the startup of the terminal, call multi-source sensors to collect raw state data. After noise filtering, interpolation marking, and time-series synchronization alignment, generate multi-source time-series data in a unified format; Extract head key points and calculate the pose angle, and determine the tilting and bumping conditions of the terminal. If the detected distance data exceeds the limit, mark the unhealthy eye use state; Use an adaptive algorithm to read the height and vision difference parameters of children and combine the correction amount from parents' feedback. Update the formula to adjust the corresponding alarm thresholds respectively. If risky postures occur frequently, tighten the limit and output the latest personalized strategy; Compare the feedback data in real time to check if it exceeds the limit. If it is abnormal, trigger an animation or voice reminder, and call the reminder cooling function to suppress high-frequency alarms. Record the user's adjustment behavior and parents' feedback at the reminder moment and send it back; Among them, Incorporate the correction amount into the adaptive correction function to output the distance alarm threshold and pose deviation threshold for the next cycle. If the unhealthy eye use state is detected continuously for multiple times, add a negative correction to the feedback correction amount or vision offset amount; Map the new threshold to the executable configuration at the current moment. If an over-limit situation is detected, mark it as about to trigger a reminder and send out this mark. If no over-limit occurs, maintain the normal monitoring state; After receiving the signal about to trigger a reminder, if mild over-limit is detected for the first time, only prompt with a small pop-up animation or a slight vibration. If repeated severe over-limit is detected and lasts for a long time, upgrade it to a voice reminder or screen masking; Upload the recognition result and personalized threshold strategy to the cloud. Parents can view the eye use statistics in real time based on this and can remotely enable the forced protection mode, analyze the eye use risks of different age and vision groups, and send back improvement suggestions; After receiving the signal about to trigger a reminder, determine the form and intensity of the reminder according to the cumulative over-limit duration, historical reminder records, and parents' preferences.

2. The eye protection method for an intelligent terminal according to claim 1, characterized in that: Obtain raw state data from the sensor array, including facial position data, distance data, device attitude data, and environmental stability data. After adopting a unified sampling clock mechanism, match the collected data at the same moment; After obtaining the raw state data, perform denoising and time-series alignment in a unified format, and output aggregated data.

3. The eye protection method for an intelligent terminal according to claim 2, characterized in that: Introduce a weighted Minkowski fusion model to perform multi-dimensional aggregation on different source data, and output the aggregated data according to the time stamp synchronized with the highest sampling frequency; If there is no new reading from the low-frequency sensor, use the effective value of the previous moment and attach an interpolation mark.

4. The eye protection method for an intelligent terminal according to claim 1, characterized in that: Extract the unit vector of the head orientation from the facial position data, obtain the unit vector of the front of the device according to the aggregated data, and jointly construct the pose angle to measure the relative tilt between the child's head and the device, identify unhealthy eye use postures, and construct a comprehensive health function by reading the pre-processed distance data and environmental stability data.

5. The eye protection method for an intelligent terminal according to claim 4, characterized in that: When the comprehensive health function exceeds the safety threshold, it is determined as an unhealthy eye - using state; Analyze the specific components of the state feature vector or the updated state feature vector to locate the source of anomalies, and record the specific threshold - exceeding items according to the over - limit components; 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.

6. The eye - protection method for an intelligent terminal according to claim 5, wherein: Read the child's height and vision condition, and map them to obtain a height offset and a vision offset; Aggregate the usage records, including the average usage distance within the past week and the occurrence frequency of detected unhealthy postures; Compare with the parent feedback information and integrate it into a feedback correction amount.

7. The eye - protection method for an intelligent terminal according to claim 6, wherein: Construct a reminder cooling function by combining the reminder time and the user's adjustment behavior, and adjusting the reminder frequency according to the time - decay strategy and user feedback; If the time interval from the last reminder to the current time is less than expected, postpone the new reminder or downgrade the reminder intensity. Before generating a reminder, first query the value of the reminder cooling function. If it is lower than expected, postpone or weaken the next reminder; if it is higher, then give a normal or enhanced reminder.

8. The eye - protection method for an intelligent terminal according to claim 7, wherein: Generate an entry record after each reminder, including the timestamp of triggering the reminder, the reminder intensity and method, and detect or confirm the result of the user's behavior change; pack the entry record and send it back to the adaptive threshold regulation part to update the feedback correction amount and other correction amounts after the next time window, and adjust the personalized distance and posture thresholds for the next cycle; For frequently occurring scenarios where reminders are ineffective, gradually weaken or postpone similar reminders; for cases where the reminder is successful and the user does cooperate, then retain or slightly relax the current threshold.

9. The eye - protection method for an intelligent terminal according to claim 8, wherein: Pack the usage entry records and personalized thresholds and upload them to the cloud server, so that the parent terminal can obtain the summary of the eye - using 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, enable the forced protection mode on the parent terminal, and the trigger record of the forced protection mode is synchronously written to the cloud.

10. The eye - protection method for an intelligent terminal according to claim 9, wherein: Aggregate the historical records from different families or terminals into a data set, and each record includes the child's age, vision condition, historical threshold evolution path, reminder trigger frequency, and parent feedback; Introduce a high - dimensional feature mapping function in the cloud, perform non - linear mapping on each record and construct a feedback state vector; apply clustering or association rule mining to the mapped vector sequence to automatically identify high - risk populations or special groups, generate a group - based eye - using risk report, and recommend usage duration and posture correction strategies.

11. An eye protection system for a smart terminal, which applies the eye protection method described in any one of claims 1 to 10, and is characterized in that: including, A data acquisition module, after detecting the startup of the terminal, calls multi - source sensors to collect raw state data. After noise filtering, interpolation marking, and time - series synchronization alignment, it generates multi - source time - series data in a unified format; The posture and distance recognition module extracts head key points, calculates the posture angles, and determines the tilting and bumping conditions of the terminal. If the detected distance data exceeds the limit, an unhealthy eye use state is marked. The adaptive threshold regulation module uses an adaptive algorithm to read the height and vision difference parameters of children and combines the correction amount of parental feedback. The update formula adjusts the corresponding alarm thresholds respectively. If risky postures are frequent, the limit is tightened and the latest personalized strategy is output. The reminder and feedback module continuously compares whether the feedback data exceeds the limit in real time. If it is abnormal, it triggers an animation or voice reminder, and calls the reminder cooling function to suppress high-frequency alarms. It records the user's adjustment behavior and parental feedback at the reminder moment and transmits them back. The remote monitoring module uploads the recognition results and personalized threshold strategies to the cloud. Based on this, the parental end can view the eye use statistics in real time and can remotely enable the forced protection mode, analyze the eye use risks of different age and vision groups, and transmit back improvement suggestions.

Images