Eye behavior intervention control method, smart wearable device and readable medium

Abstract

This disclosure provides an intervention and control method for eye-use behavior, a smart wearable device, and a computer-readable medium. The intervention and control method includes: generating an intervention command and suspending it in response to eye load meeting a triggering condition; identifying the user's state related to eye-use behavior based on collected eye movement and head movement information; maintaining the suspension of the intervention command when the identification result is a focused state; determining a task boundary for executing the intervention command when a change in the user's state from a focused state to a non-focused state is detected; determining constraints for intervention strategy selection based on environmental feature data and the user's historical intervention response data in response to the task boundary, the constraints including the sensitivity level of the current scene and the user's desensitization index; determining a target intervention strategy from multiple candidate intervention strategies based on the constraints; releasing the suspension of the intervention command and executing the target intervention strategy. This disclosure can protect flow state, eliminate social interference, and prevent desensitization adaptation.

Description

Technical Field

[0001] This disclosure relates to the fields of smart wearable devices and human-computer interaction technology, and in particular to a method for intervening and controlling eye behavior, a smart wearable device, and a computer-readable medium. Background Technology

[0002] Eye use behavior monitoring and intervention is one of the important application areas of smart wearable devices. Currently, such devices typically use built-in sensors to monitor eye distance, head posture, and other factors, and trigger alerts when eye use behavior exceeds preset thresholds.

[0003] However, in practical applications, the above intervention methods have the following shortcomings: First, the timing of intervention lacks correlation with the user's real-time cognitive state; the user may be in a highly focused state when the reminder is triggered, and the intervention itself interferes with the user's cognitive coherence. Second, the feedback method is relatively fixed and difficult to adapt to different usage scenarios; it may cause unnecessary interference in quiet environments, and may fail to effectively attract the user's attention in noisy environments. Furthermore, after using the same feedback method for a long time, the user's response to the reminder may gradually decrease, affecting the continued effectiveness of the intervention. Summary of the Invention

[0004] This disclosure provides an intervention and control method for eye use behavior, a smart wearable device, and a computer-readable medium. The purpose of this disclosure is to address the technical problems existing in current eye use behavior monitoring devices, such as inappropriate intervention timing, feedback modal conflicts with the scenario, and the potential for desensitization with long-term use, and to achieve dynamic adaptive adjustment of intervention timing and feedback strategies.

[0005] In a first aspect, embodiments of this disclosure provide an intervention control method for eye-use behavior, comprising: generating an intervention command and suspending it in response to an eye load meeting a triggering condition; identifying a user state related to eye-use behavior based on collected eye movement information and head movement information; maintaining the suspension of the intervention command when the identification result is a focused state; determining a task boundary for executing the intervention command when a change in the user state from a focused state to a non-focused state is detected; determining constraints for selecting an intervention strategy based on environmental feature data and the user's historical intervention response data in response to the task boundary, the constraints including the sensitivity level of the current scene and the user's desensitization index; determining a target intervention strategy among multiple candidate intervention strategies based on the constraints; releasing the suspension of the intervention command and executing the target intervention strategy.

[0006] In some embodiments, identifying user states related to eye-use behavior includes: extracting multidimensional eye-use behavior features based on the eye movement information and the head movement information; comprehensively processing the multidimensional eye-use behavior features to obtain a state quantity reflecting the user's level of focus; and determining whether the user is in a focused state based on the comparison result of the state quantity with a first threshold, or based on the fluctuation range of the state quantity within a preset time window.

[0007] In some embodiments, detecting a change in the user's state from a focused state to a non-focused state includes: detecting the magnitude of change of a state quantity representing the user's level of focus within a continuous time window; and determining that a change from a focused state to a non-focused state occurs when the magnitude of the change exceeds a second threshold.

[0008] In some embodiments, the state quantity is a cognitive load index, which is obtained by weighted fusion of the multidimensional eye-use behavior features; the change amplitude detection includes calculating the first difference or equivalent rate of change of the time series of the cognitive load index, wherein the multidimensional eye-use behavior features include blink frequency, gaze saccade range and head posture change amplitude.

[0009] In some embodiments, determining the sensitivity level of the current scene includes: acquiring environmental feature data of the current scene, the environmental feature data including ambient acoustic noise floor energy, ambient illuminance, and current scene identification information; determining the sensitivity level of the current scene by performing weighted fusion based on the acoustic difference between the ambient acoustic noise floor energy and a quiet reference threshold, and the brightness difference between the ambient illuminance and a working environment brightness reference threshold, combined with the current scene identification information; wherein the sensitivity level is used to characterize the degree to which the current scene restricts the physical interference of the intervention strategy, and is quantitatively characterized by a scene invasiveness penalty factor.

[0010] In some embodiments, determining the user's desensitization index includes: obtaining the user's response results to each intervention in historical intervention events, wherein the response results are used to characterize whether the user adjusted their eye load to within the triggering condition within a preset time after the intervention; based on the response results, identifying historical intervention events in which the user did not adjust their eye load to within the triggering condition within the preset time after the intervention; performing attenuation weighting on the identified historical intervention events according to the time interval between them and the current time; and recursively updating the current desensitization index based on the historical desensitization index and in combination with the attenuation weighting results.

[0011] In some embodiments, determining a target intervention strategy from multiple candidate intervention strategies based on the constraints includes: obtaining the expected success rate of each candidate intervention strategy under the current desensitization index; obtaining the preset basic physical intrusion degree of each candidate intervention strategy, and adjusting the basic physical intrusion degree based on the sensitivity level to obtain the corresponding interference cost; and selecting a target intervention strategy from the multiple candidate intervention strategies based on the expected success rate and interference cost of each candidate intervention strategy.

[0012] In some embodiments, after executing the target intervention strategy, the method further includes: determining a positive reward value based on changes in eye load before and after the intervention; determining a negative reward value based on the degree of interference caused by the target intervention strategy to the current scene; determining a comprehensive reward value based on the positive reward value and the negative reward value; and updating the expected success rate of the target intervention strategy based on the comprehensive reward value.

[0013] In a second aspect, embodiments of this disclosure provide a smart wearable device, including: an eye-tracking camera for collecting eye-tracking information; an inertial measurement unit for collecting head motion information; one or more processors; and a memory storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the processors implement the intervention control method described in the first aspect of this disclosure.

[0014] Thirdly, a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the intervention control method described in the first aspect of the present disclosure.

[0015] This disclosure identifies user eye-related states and determines task boundaries by detecting state changes, delaying intervention until the user transitions from a focused to a non-focused state. This improves the alignment between intervention and user behavior rhythms and reduces interference with the focus process. Simultaneously, by introducing scene sensitivity levels and adaptively selecting based on the interference characteristics of different intervention strategies, the system can choose relatively low-interference feedback methods under different environmental conditions, thereby reducing the impact on the surrounding environment and user experience. Furthermore, by constructing a desensitization index based on historical intervention responses and combining it with intervention effect feedback to form a closed-loop update mechanism, the system can dynamically adjust strategy selection based on user responses to different intervention strategies, thus mitigating the adaptability to a single intervention method and improving the sustained effectiveness of the intervention strategy. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall framework of an eye behavior intervention and control method according to an embodiment of this disclosure.

[0017] Figure 2This is a flowchart of an intervention and control method for eye use behavior according to an embodiment of this disclosure. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions of this disclosure will be described in detail below with reference to the accompanying drawings.

[0019] Exemplary embodiments will be described more fully below with reference to the accompanying drawings; however, these exemplary embodiments may be embodied in different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will enable those skilled in the art to fully understand the scope of this disclosure.

[0020] Where there is no conflict, the various embodiments of this disclosure and the features thereof in the embodiments may be combined with each other.

[0021] As used herein, the term “and / or” includes any and all combinations of one or more related enumerated entries.

[0022] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. As used herein, the singular forms “a” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “made of” are used in this specification, the presence of the stated feature, integral, step, operation, element, and / or component is specified, but the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded.

[0023] Unless otherwise specified, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and will not be interpreted as having an idealized or overly formal meaning, unless expressly so defined herein.

[0024] Existing methods of intervening in eye use behavior have the following problems in practical applications: Intervention triggering timing is usually determined based on a single physical threshold, lacking a comprehensive assessment of the user's current cognitive state. This can easily trigger intervention during the user's sustained focus, thus affecting the continuity of use. The feedback method is usually a fixed modality and does not adapt to the environmental context. In different use cases, the intervention method may not be sufficiently adapted to the environment. As the number of interventions increases, users may develop an adaptation effect to repetitive feedback methods, leading to a decrease in the effectiveness of the intervention response.

[0025] To address the aforementioned issues, this implementation proposes an adaptive feedback method for eye-use behavior based on cognitive state perception and scene context analysis. By constructing a cognitive load assessment mechanism, an environmental context analysis mechanism, and a dynamic adjustment mechanism based on historical responses, adaptive optimization of intervention timing and feedback strategies can be achieved.

[0026] This disclosure reconstructs the traditional linear triggering logic of "exceeding the boundary triggers an alarm," upgrading it into an edge-cloud collaborative decision-making and control flow encompassing four stages: state awareness, multi-dimensional analysis, adaptive scheduling, and reinforcement evolution. The system achieves closed-loop dynamic intervention of user eye-tracking behavior through an eye-tracking camera, inertial measurement unit (IMU), photosensor, and external collaborative IoT devices located inside the smart glasses.

[0027] Specifically, this disclosure achieves adaptive feedback of eye-use behavior through the following technical means: First, by sensing the user's cognitive load state and identifying the "task boundary" where their attention is located, the timing of the intervention instruction is controlled within the natural gaps of the user's cognitive activities, so as to achieve the intervention timing that matches the cognitive state and avoid interrupting the state of deep concentration.

[0028] Second, implicit temperature feedback and ecological collaborative visual cues are introduced. By combining spatiotemporal scene information (such as environmental noise, illuminance, network identifiers, etc.), the algorithm automatically matches the most suitable feedback modality for the current environment, so that the intervention method is adapted to the environmental context and social awkwardness is eliminated.

[0029] Third, construct a reinforcement learning closed loop based on intervention effectiveness assessment, dynamically calculate the individual's "desensitization index" to the current feedback method, and automatically upgrade or switch feedback strategies based on historical response results to achieve dynamic evolution of intervention methods and prevent sensory fatigue and feedback desensitization in long-term use.

[0030] Figure 1 This is a schematic diagram of the overall framework of an eye behavior intervention and control method according to an embodiment of this disclosure. Figure 1 It demonstrates a four-stage decision flow, comprising stage one (cognitive state assessment and task boundary detection), stage two (contextual analysis and desensitization index calculation), stage three (adaptive scheduling of feedback modalities), and stage four (intervention effectiveness assessment and strategy evolution).

[0031] like Figure 1 As shown, the intervention and control methods for eye use behavior include the following stages: When visual load exceeds the limit, the intervention process is triggered, an intervention command is generated, and the process is put into a suspended state.

[0032] The process then proceeds to Phase One, which assesses the user's cognitive state. Specifically, by collecting the user's eye movement and head movement information, visual behavior characteristics are extracted, and the Cognitive Load Index (CLI) is calculated. Based on this, it is determined whether the user is currently in a task boundary state; if not, the intervention command is suspended.

[0033] When the user's state is detected to have reached the task boundary, the process enters Phase Two, which analyzes the context and the user's historical response characteristics. Specifically, this includes calculating the Scene Intrusion Index (CIP) for the current scenario and the User Desensitization Index (DI) to determine constraints.

[0034] In Phase Three, based on the Cognitive Load Index (CLI), Scene Intrusion Index (CIP), and Desensitization Index (DI), a multi-objective optimization function is constructed to uniformly evaluate multiple candidate intervention strategies and select the optimal feedback strategy m. Candidate intervention strategies include thermosensitive feedback, visual-coordinated feedback, and tactile feedback. After determining the target intervention strategy, the corresponding feedback control is implemented.

[0035] The fourth stage then proceeds to evaluate the intervention's effectiveness. This involves obtaining the change in visual distance Δd before and after the intervention and calculating the immediate reward value. This is used to characterize the overall effect of this intervention.

[0036] Based on the immediate reward value, the policy value parameters of the current state-action pair are updated using reinforcement learning methods, thereby adjusting the expected returns of each candidate intervention strategy under different states.

[0037] After completing the strategy update, end the current intervention cycle and wait for the next visual load trigger.

[0038] Figure 2 This is a flowchart of an intervention and control method for eye use behavior according to an embodiment of the present disclosure, illustrating the main logic from triggering to execution.

[0039] Firstly, referring to Figure 2 This disclosure provides a method for intervening and controlling eye-use behavior, including: S21. In response to the eye load meeting the triggering condition, generate an intervention command and put it into a suspended state; S22. Based on the collected eye movement and head movement information, identify the user status related to eye use behavior; S23. When the recognition result is a focused state, the intervention command is suspended; when the user state is detected to change from a focused state to a non-focused state, it is determined as the task boundary where the intervention command can be executed. S24. In response to the task boundary, determine the constraints for selecting the intervention strategy based on environmental feature data and the user's historical intervention response data. The constraints include the sensitivity level of the current scenario and the user's desensitization index. S25. Based on the constraints, determine the target intervention strategy from multiple candidate intervention strategies; S26. Release the suspended state of the intervention command and execute the target intervention strategy.

[0040] This disclosure provides a method for intervening and controlling eye-use behavior. The method is based on multimodal perception and adaptive decision-making mechanisms to achieve dynamic intervention and control of user eye-use behavior. The method includes the following steps: Step S21, Triggering and Suspending First, the system continuously collects user eye distance and head posture data using a ranging sensor and / or a posture sensor. When the system detects that the current eye-related parameters meet preset eye load triggering conditions, it determines that the current eye load is abnormal or requires intervention, and generates a corresponding intervention command. Unlike traditional instant triggering, this intervention command is placed in a suspended state and is not executed immediately.

[0041] The triggering conditions for eye strain can be determined based on one or more eye-related indicators, including but not limited to: eye distance, head posture, continuous eye use duration, visual stability, blinking characteristics, or other parameters that can characterize visual fatigue or poor eye use behavior. Eye strain meeting the triggering conditions may include: an eye distance less than a preset distance threshold, and / or a continuous eye use duration exceeding a preset duration threshold; however, this disclosure is not limited to these, and other equivalent or extended triggering methods are also within the scope of this disclosure.

[0042] Step S22, Cognitive State Recognition Subsequently, based on eye movement information (including blink frequency, changes in fixation point, etc.) and head movement information (including angular velocity, posture changes, etc.), the system identifies the user's state related to eye-use behavior. Specifically, the system extracts eye features at high frequency using a medial eye-tracking camera and combines this with IMU data to assess head and neck posture stability, calculating the user's current cognitive load index. When the cognitive load index remains at a high value, the identification result is a focused state, and the intervention command is suspended at this time.

[0043] Step S23, Task Boundary Detection Furthermore, the process of changing the user's state is continuously monitored. Specifically, the time series of the cognitive load index is calculated using first-order difference. When a sharp drop in the cognitive load index is detected (i.e., the first-order difference exceeds a preset threshold), it is determined that the user's state has changed from a focused state to a non-focused state. The current moment is identified as the task boundary (i.e., the user's natural attention gap), which serves as the optimal time for intervention.

[0044] Step S24: Determine the constraints After detecting the task boundary, the system acquires environmental characteristic data and the user's historical intervention response data, and determines the constraints for selecting the intervention strategy accordingly. These constraints include at least the sensitivity level of the current scenario and the user's desensitization index.

[0045] The sensitivity level of the current scene is determined by combining the absolute illuminance collected by the ambient photoresistor, the ambient acoustic noise floor energy collected by the low-frequency microphone, and spatiotemporal anchor points (such as campus LAN SSID matching). The user's desensitization index is calculated based on historical intervention response logs using a recursive model with time decay.

[0046] Step S25, Strategy Selection Based on the above constraints, the system optimizes and selects from multiple candidate intervention strategies to determine the target intervention strategy. These candidate intervention strategies include various modalities such as implicit temperature feedback, ecological collaborative visual cues, and tactile frequency-modal feedback.

[0047] Step S26: Implement the intervention Finally, the system releases the suspension of the intervention command and executes the target intervention strategy to intervene in the user's eye-use behavior.

[0048] After the intervention is implemented, the system evaluates the intervention effect through closed-loop monitoring, generates a reward value, and updates the strategy value parameters through reinforcement learning algorithms, thereby realizing the self-evolution of the intervention strategy.

[0049] In some embodiments, identifying user states related to eye-use behavior includes: Multidimensional eye-use behavior features are extracted based on the eye movement information and the head movement information; The multidimensional eye-use behavior characteristics are comprehensively processed to obtain a state quantity that reflects the user's level of focus; Based on the comparison result between the state quantity and the first threshold, or based on the fluctuation range of the state quantity within a preset time window, it is determined whether the user is in a focused state.

[0050] In this embodiment of the disclosure, the process of identifying user states related to eye-use behavior includes: First, multidimensional eye-use behavior features are extracted based on the eye movement and head movement information. These features include at least blink frequency, gaze movement, and head posture stability. Specifically, a focused state corresponds to a state of deep immersion in learning, characterized by reduced blink frequency, relatively stable gaze, and minimal changes in head posture.

[0051] Subsequently, the multidimensional eye-use behavior features are fused to obtain a state quantity reflecting the user's level of focus. The fusion process can be implemented using methods such as weighted summation, normalization, or nonlinear transformation. For example, a cognitive load index can be obtained as the state quantity by weighted fusion of blink frequency, gaze range, and head posture changes.

[0052] Furthermore, the system compares the state quantity with a preset first threshold, or analyzes the fluctuation of the state quantity within a preset time window. When the state quantity exceeds the first threshold, or the fluctuation amplitude of the state quantity within the time window is less than a preset stability threshold, the system determines that the user is in a focused state; otherwise, it determines that the user is in a non-focused state.

[0053] Based on the above judgment results, the system continuously monitors the user's status and triggers subsequent intervention procedures when it detects a change from a focused state to a non-focused state.

[0054] In some embodiments, detecting a change in the user's state from a focused state to a non-focused state includes: The magnitude of change in the state quantity representing the user's level of focus is detected within a continuous time window; When the magnitude of the change exceeds the second threshold, it is determined that the state has changed from a focused state to a non-focused state.

[0055] In this embodiment of the disclosure, the process of detecting a change in user state from a focused state to a non-focused state includes: Based on the state variables reflecting the user's level of focus obtained in the aforementioned embodiments, the magnitude of change of these state variables is detected within a continuous time window. The time window is a sliding window, for example, with a window length of 1-2 seconds, and windows sliding with a 50% overlap rate.

[0056] Specifically, the magnitude of change of the state quantity is obtained by calculating the change value or rate of change of the state quantity between adjacent time windows. For example, the difference between the mean of the state quantity in the current window and the mean of the state quantity in the previous window can be calculated, or the first-order difference calculation can be performed on the time series of the state quantity to obtain the magnitude of change.

[0057] When the change exceeds a preset second threshold, it is determined that the user's state has changed abruptly, that is, from a focused state to a non-focused state. This determines the current moment as the task boundary, that is, the user's natural attention gap, which is the best time to execute the suspension intervention command.

[0058] In some embodiments, the state quantity is a cognitive load index, which is obtained by weighted fusion of the multidimensional eye-use behavior features; The change magnitude detection includes calculating the first difference or equivalent rate of change of the time series of the cognitive load index. The multidimensional eye behavior characteristics include blinking frequency, gaze range, and head posture variation.

[0059] In this embodiment of the disclosure, the state quantity is the Cognitive Load Index (CLI), which is obtained by weighted fusion of multidimensional eye-use behavior features.

[0060] Specifically, the system extracts eye features at high frequency using a medial eye-tracking camera and combines this with IMU data to assess head and neck posture stability. The cognitive load index CLI(t) is a weighted fusion of features from three dimensions: blink frequency, saccade path variance, and head micro-movements. The calculation formula is as follows: in: This represents the blink frequency within the current time window. The time window is a fixed-length window, such as 1 to 2 seconds, and is implemented using a sliding window with a 50% overlap rate. A CLI value is calculated for each time window, forming a time series.

[0061] This represents the baseline blink frequency when the user is relaxed. This value is automatically updated based on the user's historical data, and the longer the user wears the device, the closer it is to the user's true average. Blink frequency decreases significantly during deep focus, and the exponential term causes it to exhibit non-linear amplification.

[0062] The variance of the eye saccade path represents the intensity of the viewpoint shift. During deep reading, the gaze is often locked on a specific area, resulting in minimal variance.

[0063] The variance of the head's three-axis angular velocity extracted from the IMU characterizes the head's attitude stability.

[0064] To prevent extremely small positive numbers with a denominator of zero.

[0065] To normalize the feature weights, fixed empirical values ​​determined based on experimental data can be used.

[0066] When CLI(t) remains at a high value, the system determines that the user is in a state of deep focused flow and keeps the intervention command suspended.

[0067] Furthermore, the magnitude of change detection is achieved by performing first-order difference or equivalent rate of change calculation on the time series of the cognitive load index. Specifically, the algorithm performs first-order difference calculation (derivative operation) on the CLI(t) sequence to find the "task boundary" where the cognitive load drops sharply. When the rate of change (derivative) exceeds a preset threshold, it is determined that the user's state has changed abruptly (i.e., from a focused state to a non-focused state), and the current moment is determined as the task boundary.

[0068] In practical applications, the physical manifestations of this sudden change in state include, but are not limited to: the user suddenly raising their head significantly to shift their gaze, or starting to rub their eyes frequently or blink continuously due to fatigue, indicating that the user has been released from a state of deep concentration and entered a natural rest interval.

[0069] The threshold is a fixed threshold set based on group experience. Since the CLI remains relatively stable when the user is focused, but fluctuates drastically at task boundaries with a drastic rate of change, the fixed threshold is sufficient to meet the requirements.

[0070] Once the task boundary (natural rest interval) is detected, the intervention strategy selection and execution process begins.

[0071] In some embodiments, determining the sensitivity level of the current scenario includes: Acquire environmental feature data of the current scene, including ambient acoustic noise floor energy, ambient illuminance, and current scene identification information; Based on the acoustic difference between the ambient acoustic noise floor energy and the quiet reference threshold, and the brightness difference between the ambient illuminance and the working environment brightness reference threshold, and combined with the current scene identification information, a weighted fusion is performed to determine the sensitivity level of the current scene. The sensitivity level is used to characterize the degree to which the current scene restricts the physical interference of the intervention strategy, and is quantitatively characterized by the scene invasiveness penalty factor.

[0072] In this embodiment of the disclosure, the sensitivity level of the current scene is used to characterize the degree to which the current scene limits the physical interference of the intervention strategy, and it is obtained by comprehensively evaluating the environmental acoustic features, lighting features and scene semantic information.

[0073] Specifically, firstly, environmental characteristic data of the current scene is acquired. This environmental characteristic data includes ambient acoustic noise floor energy, ambient illuminance, and current scene identification information. For example, the ambient acoustic noise floor energy N(t) can be obtained by a low-frequency microphone on the device, the ambient illuminance L(t) can be obtained by an ambient photoresistor, and the current scene identification information can be obtained by WiFi SSID matching, GPS positioning, or user-defined scene tags, such as campus, library, or home environment.

[0074] Subsequently, the physical interference sensitivity of the current scene is evaluated based on the acoustic difference between the ambient acoustic noise floor energy and the preset quiet reference threshold, and the brightness difference between the ambient illuminance and the preset working environment brightness reference threshold.

[0075] Among them, the quiet reference threshold The preset value is within the range of 30~40dB, which is the reference threshold for ambient brightness in the working environment. The preset value can be set to a range of 500 to 750 lux.

[0076] In one specific embodiment, by weighted fusion of acoustic difference, brightness difference, and current scene identification information, a scene invasiveness penalty factor CIP(t) is obtained to quantify the sensitivity level, which is expressed as follows: ) in: The semantic indicator variable corresponding to the scene identifier is set to 1 when a high-constraint scene such as a campus or library is identified, and 0 otherwise. Other scene identifier information (such as the "classroom mode" manually set by the user) can be encoded as 0 / 1 values ​​in a similar way to participate in the fusion.

[0077] The ambient acoustic noise floor energy at the current moment; This is the preset quiet reference threshold.

[0078] The ambient illuminance at the current moment; This is the preset reference threshold for ambient brightness in the working environment.

[0079] α1, α2, and α3 are weighting coefficients that can be adjusted according to the application scenario.

[0080] In this definition, the scene invasiveness penalty factor CIP(t) is used to numerically represent the sensitivity level. The higher the CIP(t) value, the quieter the current scene (such as evening self-study or library), and the greater the penalty should be for the system to apply strong invasive feedback such as mechanical vibration or buzzing. Conversely, the lower the CIP(t) value, the more tolerant the current scene is of higher intensity physical intervention.

[0081] The main technical objective of this module is to prevent users from feeling embarrassed by inappropriate reminder methods in quiet, socially sensitive settings. Since the primary users are children and teenagers, this embodiment prioritizes mitigating the social pressures they face while focused on studying. Through this design, the system can automatically reduce the priority of highly intrusive feedback in quiet environments, prioritizing implicit feedback methods such as temperature sensing.

[0082] In some embodiments, determining the user's desensitization index includes: Obtain the user's response results to each intervention in historical intervention events. The response results are used to characterize whether the user has adjusted the eye load to within the trigger condition within a preset time after the intervention. Based on the response results, historical intervention events were identified in which the eye load was not adjusted to within the triggering condition within a preset time after the intervention. For identified historical intervention events, a decay weighting is applied based on the time interval between them and the current time. Based on the historical desensitization index and combined with the attenuation weighted result, the current desensitization index is obtained through recursive updating.

[0083] In this embodiment of the disclosure, the process of determining the user de-identification index includes: First, obtain the user's response results from historical intervention events. These response results characterize whether the user adjusted their eye load to within the triggering conditions within a preset time (i.e., whether they made substantial posture corrections). It should be noted that if the user fails to complete the correction within the preset time, it is considered a non-response.

[0084] Secondly, based on the response results, intervention events that were not corrected within a preset time (i.e., events with a response result of "no response") are identified.

[0085] Subsequently, the identified unresponsive events are weighted and attenuated according to their time interval from the current time. The shorter the time interval, the greater the attenuation weight, indicating that the recent unresponsive events have a more significant impact on the current desensitization index.

[0086] Finally, the current desensitization index is obtained by iteratively updating the index based on the historical desensitization index (DI). :

[0087] in: The desensitization index at the time of the last intervention (historical desensitization index). This is the user's response indicator after the last intervention. If the user completes the correction within the preset time, the value is 1; otherwise, the value is 0. This is the time interval since the last intervention; This is the historical memory forgetting factor, with a value between 0 and 1, used to control the degree of influence of the historical desensitization index on the current value; This is the time decay constant, used to control the effect of the time interval on the decay rate. It can be 0.5, with a preferred range of 0.4 to 0.6.

[0088] The core of this recursive update mechanism is: if a user has continuously ignored system reminders recently (i.e., R=0 and Δt is extremely short), the desensitization index... The exponential surge indicates that users have become severely desensitized to the current intervention strategy, and the system must switch feedback strategies to break the users' "resistance".

[0089] In this embodiment, the system maintains a corresponding desensitization index for each candidate intervention strategy. When the system switches to a new intervention strategy, the historical desensitization index corresponding to that strategy is updated. Initialized to 0, the current desensitization index is only affected by currently unresponsive events.

[0090] In some embodiments, determining a target intervention strategy from a plurality of candidate intervention strategies based on the constraints includes: Obtain the expected success rate of each candidate intervention strategy under the current desensitization index; Obtain the preset basic physical intrusion level for each candidate intervention strategy, and adjust the basic physical intrusion level based on the sensitivity level to obtain the corresponding interference cost; Based on the expected success rate and interference cost of each candidate intervention strategy, a target intervention strategy is selected from the multiple candidate intervention strategies.

[0091] In this embodiment, the process of determining the target intervention strategy includes: First, obtain the expected success rate of each candidate intervention strategy under the current desensitization index.

[0092] Secondly, the basic physical intrusion of each candidate strategy is obtained and adjusted in conjunction with the sensitivity level of the current scene to obtain the corresponding interference cost.

[0093] Finally, based on the following optimization objective function, the target intervention strategy is selected from the candidate strategies: First, obtain the expected success rate of each candidate intervention strategy under the current desensitization index. The expected success rate is a dynamically updated parameter, representing the percentage of success at the current desensitization index (DI). nThe expected probability that the adopted strategy can substantially correct the user's eye-use behavior is given. This value is assigned a prior probability based on experimental baseline data during the system cold start phase (for example, setting the success rate of temperature feedback in a low desensitization state to 0.85); in actual operation, this value is dynamically output by the reinforcement learning model and is equivalently mapped from the state-action value function in the reinforcement learning model.

[0094] Secondly, obtain the preset basic physical intrusion level for each candidate strategy. And adjust it according to the current scene sensitivity level CIP(t) to obtain the corresponding interference cost, where This is a balancing coefficient used to adjust the weight of the interference cost in the objective function.

[0095] Finally, based on the following optimization objective function, a target intervention strategy is selected from the candidate strategy set M. :

[0096] The physical meaning of this objective function is: to select the strategy with the largest difference between the expected success rate and the cost of interference, that is, to maximize the probability of successful intervention while minimizing social interference in the current scenario.

[0097] In practical applications, the system calculates the objective function value for each candidate strategy and selects the strategy with the largest value as the target intervention strategy. .

[0098] In this embodiment, the candidate intervention strategies include, but are not limited to, the following three: Strategy A (Implicit Thermal Feedback): Selected when the scene sensitivity level is extremely high (such as a quiet scene) and the desensitization index is low. The system abandons mechanical vibration and outputs a progressive control curve to drive a miniature thermal hysteresis element, generating a thermal cues that smoothly transition from body temperature to warning temperature (such as 39°C~41°C), achieving zero noise and minimizing social interference.

[0099] Strategy B (Ecosystem Collaboration Visual Cueing): Selected when in a home interconnected scenario (e.g., the device senses a specific home LAN SSID environment) and the desensitization index is low. The system issues cross-device control commands through IoT protocols, using external smart devices (such as smart study lamps and tablets) to execute visual cues, such as adjusting the lamp's color temperature from cool white light to warm yellow light, or controlling the screen edge to produce a breathing halo effect.

[0100] Strategy C (Haptic Frequency Variable Feedback): Selected when the desensitization index exceeds the high-risk threshold, indicating that conventional intervention has failed. The system breaks the inherent single-frequency vibration rhythm and generates an asymmetric, pseudo-random tactile pulse sequence, producing a strong, unexpected tactile stimulus similar to a "rapid heartbeat," thus reawakening the user's proprioception.

[0101] It should be noted that the above policy pool can be expanded according to actual needs, including but not limited to the three policies mentioned above. This disclosure focuses on intervention and control methods and does not impose excessive constraints on specific hardware implementations.

[0102] In some embodiments, after executing the targeted intervention strategy, the method further includes: Positive return values ​​are determined based on changes in eye load before and after the intervention; The negative reward value is determined based on the degree of interference generated by the target intervention strategy in the current scenario; Based on the positive return value and the negative return value, determine the overall return value; The expected success rate of the target intervention strategy is updated based on the overall return value.

[0103] In this embodiment of the disclosure, the closed loop of intervention effectiveness evaluation and strategy evolution includes the following processes: After implementing the targeted intervention strategy, the system immediately opens the intervention result observation window. The system uses an end-to-end eye-tracking camera and a depth sensor to calculate the change in visual distance Δd before and after the intervention, representing the degree of improvement in the user's eye distance.

[0104] Simultaneously, the system obtains the current scene sensitivity level CIP(t) and combines it with a preset scene tolerance threshold. Assess the extent to which this intervention disrupts the current scenario.

[0105] System defines an instant reward function as follows:

[0106] in: Δd represents the change in eye distance before and after the intervention; Correct the distance threshold for the target; CIP(t) represents the sensitivity level of the current scene (scene invasiveness penalty factor). This represents the scene tolerance threshold. η1 and η2 are the positive and negative return coefficients, respectively.

[0107] This reward function is used to comprehensively represent the net benefit of this intervention, where: a positive reward is generated when the user effectively corrects their eye posture; and a negative penalty is generated when the interference caused by the intervention to the scene exceeds the tolerance threshold.

[0108] The scene tolerance threshold It can be dynamically determined according to the scenario type, including but not limited to: determining based on preset rules, obtaining by looking up a table based on scenario tags, or being customized by the user through the application interface.

[0109] Based on the aforementioned comprehensive return value The system updates the strategy evaluation result of the target intervention strategy in the current state and uses it to calculate the expected success rate in the subsequent intervention strategy selection process.

[0110] In some embodiments, updating the expected success rate of the targeted intervention strategy based on the overall return value includes: Obtain user status information when executing the target intervention strategy; Using the comprehensive reward value and the user status information as input, the strategy value parameters corresponding to the user status information and the target intervention strategy are updated through an adaptive learning algorithm, wherein the adaptive learning algorithm includes a reinforcement learning algorithm. The updated strategy value parameters are mapped to the expected success rate of the target intervention strategy under the user's state information.

[0111] In this embodiment, the strategy value parameter is used to characterize the expected long-term returns of different intervention strategies under a given user state. It is obtained by learning and updating the historical intervention execution results. The update process iteratively adjusts the strategy value based on the reward signal after the intervention execution, so that the priority of high-return strategies gradually increases in the corresponding state, and the priority of low-return strategies gradually decreases, thereby forming a strategy evaluation mechanism that dynamically evolves with user behavior feedback.

[0112] In some embodiments, the policy value parameter can be updated using any reinforcement learning or adaptive learning method, including but not limited to value function-based update methods, policy gradient-based methods, or other online learning methods. The system adjusts the policy value parameter according to the current state, the selected intervention policy, and the corresponding reward value, and uses the updated policy value to characterize the expected effectiveness of the intervention policy in that state.

[0113] In one alternative implementation, the process of updating the expected success rate based on the comprehensive reward value is implemented through a reinforcement learning framework, which is based on a Markov decision process and includes the following steps: 1) Definition of state space and action space To prevent strategy rigidity, the system introduces a reinforcement learning algorithm based on Markov decision process (MDP) to continuously update the expected success rate matrix of candidate intervention strategies.

[0114] The system will use the user state space Defined as:

[0115] in: The cognitive load index is used to characterize the user's current level of focus. The user's anonymization index; The sensitivity level is determined by the scene.

[0116] based on It can further determine whether the user is in a focused or unfocused state. Motion space. The selected intervention strategy m (i.e. the target intervention strategy) includes multiple candidate strategies such as implicit temperature feedback, ecological synergistic visual cue, and tactile frequency-modulated feedback.

[0117] 2) Strategy value parameter update After each intervention is completed and the observation window ends, the system obtains the current state. Execution of actions and the corresponding overall return value The policy value parameters are then updated using a reinforcement learning algorithm. In some embodiments, the reinforcement learning algorithm employs Q-learning, and its update method is as follows:

[0118] in: Indicates the state Next action The strategic value of (i.e., the selected target intervention strategy); The learning rate; γ is the discount factor; The new state after the intervention; This indicates the highest value among all candidate strategies in the new state.

[0119] This update mechanism considers not only the immediate benefits of this intervention. Also through This allows for the estimation of future long-term returns, thereby optimizing strategy selection.

[0120] In this embodiment of the disclosure, the strategy value parameter is stored in the form of a state-action pair. After each intervention is executed and the corresponding comprehensive reward value is obtained, the system only processes the current state. With the selected action Corresponding strategy value parameters The system updates policy value parameters without adjusting those of other unvisited state-action pairs. This local update-based policy value parameter update mechanism enables the system to learn and dynamically adapt to user behavior responses online while maintaining manageable computational complexity.

[0121] As the number of interventions increases, the system gradually accumulates the value assessment results of each state-action pair, thereby constructing a strategy value evaluation system for specific users and realizing the personalized evolution of intervention strategies.

[0122] 3) Initialization mechanism During the initial operation phase of the system, the strategy value parameter It has preset initial values. These initial values ​​are set based on empirical or experimental data to achieve a cold start. For example, the initial Q values ​​of different intervention strategies vary depending on the sensitivity level of the scene. In quiet scenes such as libraries, the initial Q value of implicit temperature feedback is higher, while the initial Q value of mechanical vibration is lower.

[0123] 4) Expected success rate mapping Updated strategy value parameters The expected success rate is converted through a normalized mapping function, which includes, but is not limited to, the Softmax function. After mapping, the higher the policy value, the greater its corresponding expected success rate. For example, if the reinforcement learning algorithm uses Q-learning, the Q value is an absolute value (e.g., policy A has a score of 8.5, policy B has a score of 2.1), while the expected success rate is a probability value between 0 and 1. The Q value of each candidate policy is mapped to a probability distribution using a normalized function. After mapping, the higher the Q value of the policy, the higher its corresponding expected success rate (e.g., policy A has a success rate of 80%, policy B has a success rate of 20%).

[0124] The expected success rate is fed back into the intervention strategy selection process to calculate the optimal intervention strategy. Through this closed-loop update mechanism, the system can dynamically adjust strategy preferences based on user feedback, achieving personalized intervention.

[0125] It should be noted that the Q-learning method described above is only one implementation method. This disclosure is not limited to this method. Other reinforcement learning methods or adaptive learning methods can also be used to update the policy evaluation parameters.

[0126] Secondly, embodiments of this disclosure provide an intervention and control device for eye-use behavior, comprising: The load monitoring module is used to generate an intervention command and put it into a suspended state in response to the eye load meeting the triggering conditions; The status recognition module is used to identify user status related to eye use behavior based on the collected eye movement and head movement information. The task boundary detection module is used to suspend the intervention command when the recognition result is a focused state; and to determine the task boundary where the intervention command can be executed when the user's state changes from a focused state to a non-focused state. The constraint determination module is used to determine the constraints for intervention strategy selection based on environmental feature data and the user's historical intervention response data in response to the task boundary. The constraints include the sensitivity level of the current scene and the user's desensitization index. The strategy selection module is used to determine the target intervention strategy from multiple candidate intervention strategies based on the constraints. The execution module is used to release the suspended state of the intervention command and execute the target intervention strategy.

[0127] This disclosure also provides an intervention and control device for eye use behavior. This device is used to implement each step in the above-described intervention and control method. Its specific implementation is basically the same as that of the method embodiment, and will not be described again here.

[0128] Intervention control devices include: The load monitoring module is used to generate an intervention command in response to the eye load meeting the triggering conditions, and to set the intervention command to a suspended state; The state recognition module is used to extract eye-use behavior features based on the collected eye movement information and head movement information, and to calculate the cognitive load index (CLI) that represents the user's cognitive load in order to identify the user's eye-use state. The task boundary detection module is used to determine the user's state evolution process based on the changing trend of the cognitive load index (CLI); when the user is in a focused state, the intervention command is kept in a suspended state; when the user's state changes from a focused state to a non-focused state, it is determined as the task boundary where the intervention command can be executed. The constraint determination module is used to calculate the scene intrusion degree (CIP) of the current scene based on environmental feature data after detecting the task boundary, and to calculate the user's desensitization index (DI) based on historical intervention response data, thereby determining the constraint conditions for the selection of intervention strategies. The strategy selection module is used to construct a multi-objective decision function based on the cognitive load index CLI, scene intrusion CIP and desensitization index DI, to comprehensively evaluate multiple candidate intervention strategies and determine the optimal target intervention strategy. The execution module is used to release the suspended state of the intervention command and execute the target intervention strategy.

[0129] It should be noted that the above-mentioned functional modules can be implemented entirely or partially by software, hardware, firmware, or any combination thereof. In one specific embodiment, each module can be implemented by the processor of the smart wearable device calling computer program instructions stored in the memory.

[0130] Optionally, the above functional modules can also be designed using a hardware description language and integrated into an application-specific integrated circuit or a field-programmable gate array.

[0131] Thirdly, embodiments of this disclosure provide a smart wearable device, including: Eye-tracking cameras are used to collect eye-tracking information; An inertial measurement unit (IMU) is used to collect head motion information. One or more processors; A memory having stored one or more programs that, when executed by one or more processors, cause the processors to implement any of the intervention control methods described above.

[0132] Fourthly, embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the intervention control method described above.

[0133] The intervention and control method and device for eye use behavior disclosed herein have the following beneficial effects: 1) In response to the problem of “unreasonable intervention timing and easy interruption of focus”, this disclosure identifies the user’s eye-related state and determines the task boundary by combining state change detection. The intervention time is postponed to the stage when the user changes from a focused state to a non-focused state, thereby avoiding triggering intervention during deep focus, improving the rationality of intervention timing and reducing interference with the user’s focus process.

[0134] 2) To address the issue that "intervention methods are prone to causing social interference", this disclosure introduces a scene sensitivity level to constrain intervention strategies in different environments and adaptively selects them based on the interference cost of each intervention strategy. This allows the system to prioritize low-interference methods in quiet or sensitive scenes, thereby reducing the impact on the surrounding environment and the user's social experience.

[0135] 3) To address the issue of "intervention strategies are prone to reduced adaptability", this disclosure constructs a desensitization index based on historical intervention responses and combines it with intervention effect feedback to form a closed-loop update mechanism. This enables the system to dynamically adjust strategy selection based on user responses to different intervention strategies, thereby mitigating the decline in user adaptability to fixed intervention methods and improving the sustainability of intervention effects.

[0136] To enable those skilled in the art to more clearly understand the technical solutions provided by the embodiments of this disclosure, the technical solutions provided by the embodiments of this disclosure will be described in detail below through specific embodiments: Example 1 In one specific embodiment, the process of applying the above-described intervention decision-making method for eye use behavior is as follows: Scenario setting: The user is a junior high school student who wears smart glasses to do close-range writing assignments in a quiet library environment.

[0137] (1) Triggering and Suspension: The system detects that the user’s eye distance is 28cm (less than the preset safe distance threshold of 30cm) through the distance sensor for a continuous preset time. It determines that the eye load meets the triggering condition, generates an intervention command and puts it into suspension state.

[0138] (2) User State Recognition: The system extracts multi-dimensional eye-use behavior features based on the collected eye movement and head movement information, including blink frequency, gaze saccade characteristics, and head posture stability; by weighted fusion processing of the multi-dimensional eye-use behavior features, the cognitive load index is obtained as a state quantity representing the user's level of focus. At this time, the user is focused on solving the problem, the cognitive load index remains at a high level and fluctuates little, the user is determined to be in a focused state, and the intervention command is kept in a suspended state.

[0139] (3) Task boundary detection: During continuous monitoring, after 10 minutes, the user finished solving a problem, let out a sigh of relief, stretched, and blinked repeatedly. The system performed first-order difference calculation on the time series of the cognitive load index, detected that the change exceeded the preset threshold, and determined that the user's state changed from a focused state to a non-focused state (the physical manifestation of this state change was that the user suddenly raised his head significantly, stretched, and blinked repeatedly), thus identifying it as the task boundary for which an intervention instruction could be executed.

[0140] (4) Determination of constraints: The system acquires environmental characteristic data of the current scene, including ambient acoustic noise energy and ambient illuminance, finds that the light is sufficient but extremely quiet, and combines scene identification information (such as WiFi SSID matching the library network) to determine that the current scene is a library scene, and calculates that the scene intrusion penalty coefficient CIP is extremely high; at the same time, the system checks the historical logs and finds that the user did not ignore the reminder today, and the system calculates the user's desensitization index DI based on historical intervention response data, which is low; thus, the constraints for the selection of intervention strategies are determined, including scene sensitivity level (characterized by CIP) and user desensitization index DI.

[0141] (5) Intervention Strategy Selection: Based on constraints, the system obtains the expected success rate and basic physical intrusion level of each candidate intervention strategy. Due to the extremely high penalty of CIP, mechanical vibration will deduct a large number of points; while implicit temperature feedback not only has an expected success rate... The system is highly sensitive to temperature and completely silent, resulting in the highest overall benefit score. Therefore, the system selects implicit temperature feedback as the target intervention strategy.

[0142] (6) Execute the intervention: The system releases the suspended state of the intervention command and executes the target intervention strategy. Specifically, the micro thermal hysteresis element in the temple of the glasses is activated, and the temperature rises smoothly to 40°C within 3 seconds, giving the skin a clear but silent thermal reminder, prompting the user to adjust their eye behavior.

[0143] (7) Effect evaluation: The user felt the warmth, realized that he was lying too low, and increased the reading distance to 45cm. The system calculates the change in visual distance Δd before and after the intervention through end-to-end eye movement and depth sensor, and generates a positive reward value based on the degree of improvement in visual distance.

[0144] (8) Strategy Update: The evaluation window ends, the system determines that the line of sight has significantly improved, and the overall reward value is updated. The value is a high positive number. The system takes the current user state information (including cognitive load index, desensitization index, and scene sensitivity level) and the comprehensive reward value as input, and updates the policy value parameter (Q value) of the state-action pair "library scene + temperature-sensitive feedback in a low desensitization state" through a reinforcement learning algorithm, and maps the updated Q value to the expected success rate. In similar scenarios in the future, the system will be more inclined to choose this strategy.

[0145] Technical effect: In this embodiment, by performing intervention when the user is at the natural task boundary, and combining the scene sensitivity level and the user desensitization index to select strategies, and performing closed-loop updates based on the intervention effect, the technical effect of improving the effectiveness of intervention is achieved while reducing social interference.

[0146] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0147] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in connection with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in connection with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of this disclosure as set forth by the appended claims.

Claims

1. A method for intervening and controlling eye-use behavior, characterized in that, include: In response to the eye strain meeting the triggering conditions, an intervention command is generated and placed in a suspended state; Based on the collected eye movement and head movement information, identify user status related to eye use behavior; When the recognition result indicates a focused state, the intervention command is suspended; when a change in the user's state from a focused state to a non-focused state is detected, it is determined as a task boundary where the intervention command can be executed. In response to the task boundary, constraints for the selection of intervention strategies are determined based on environmental characteristic data and the user's historical intervention response data. These constraints include the sensitivity level of the current scenario and the user's desensitization index. Based on the constraints, the target intervention strategy is determined from multiple candidate intervention strategies; Release the suspended state of the intervention command and execute the target intervention strategy.

2. The intervention and control method according to claim 1, characterized in that, Identify user states related to eye-use behavior, including: Multidimensional eye-use behavior features are extracted based on the eye movement information and the head movement information; The multidimensional eye-use behavior characteristics are comprehensively processed to obtain a state quantity that reflects the user's level of focus; Based on the comparison result between the state quantity and the first threshold, or based on the fluctuation range of the state quantity within a preset time window, it is determined whether the user is in a focused state.

3. The intervention and control method according to claim 2, characterized in that, The detection of a change in the user's state from a focused state to a non-focused state includes: The magnitude of change in the state quantity representing the user's level of focus is detected within a continuous time window; When the magnitude of the change exceeds the second threshold, it is determined that the state has changed from a focused state to a non-focused state.

4. The intervention and control method according to claim 3, characterized in that, The state quantity is the cognitive load index, which is obtained by weighted fusion of the multidimensional eye-use behavior features; The change magnitude detection includes calculating the first difference or equivalent rate of change of the time series of the cognitive load index. The multidimensional eye behavior characteristics include blinking frequency, gaze range, and head posture variation.

5. The intervention and control method according to claim 1, characterized in that, Determining the sensitivity level of the current scenario includes: Acquire environmental feature data of the current scene, including ambient acoustic noise floor energy, ambient illuminance, and current scene identification information; Based on the acoustic difference between the ambient acoustic noise floor energy and the quiet reference threshold, and the brightness difference between the ambient illuminance and the working environment brightness reference threshold, and combined with the current scene identification information, a weighted fusion is performed to determine the sensitivity level of the current scene. The sensitivity level is used to characterize the degree to which the current scene restricts the physical interference of the intervention strategy, and is quantitatively characterized by the scene invasiveness penalty factor.

6. The intervention and control method according to claim 1, characterized in that, The determination of the user's desensitization index includes: Obtain the user's response results to each intervention in historical intervention events. The response results are used to characterize whether the user has adjusted the eye load to within the trigger condition within a preset time after the intervention. Based on the response results, historical intervention events were identified in which the eye load was not adjusted to within the triggering condition within a preset time after the intervention. For identified historical intervention events, a decay weighting is applied based on the time interval between them and the current time. Based on the historical desensitization index and combined with the attenuation weighted result, the current desensitization index is obtained through recursive updating.

7. The intervention and control method according to claim 1, characterized in that, Based on the aforementioned constraints, the target intervention strategy is determined from multiple candidate intervention strategies, including: Obtain the expected success rate of each candidate intervention strategy under the current desensitization index; Obtain the preset basic physical intrusion level for each candidate intervention strategy, and adjust the basic physical intrusion level based on the sensitivity level to obtain the corresponding interference cost; Based on the expected success rate and interference cost of each candidate intervention strategy, a target intervention strategy is selected from the multiple candidate intervention strategies.

8. The intervention and control method according to claim 7, characterized in that, Following the implementation of the target intervention strategy, the following is also included: Positive return values ​​are determined based on changes in eye load before and after the intervention; The negative reward value is determined based on the degree of interference generated by the target intervention strategy in the current scenario; Based on the positive return value and the negative return value, determine the overall return value; The expected success rate of the target intervention strategy is updated based on the overall return value.

9. A smart wearable device, characterized in that, include: Eye-tracking cameras are used to collect eye-tracking information; An inertial measurement unit (IMU) is used to collect head motion information. One or more processors; A memory having stored one or more programs that, when executed by one or more processors, cause the processors to implement the intervention control method according to any one of claims 1 to 8.

10. A computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the intervention control method according to any one of claims 1 to 8.

Images