Real-time eye recognition and visual inspection process control method

Abstract

The present application relates to a real-time eye identification and visual detection process control method, a temporal edge disturbance stimulus is applied in a visual field to induce a spontaneous dominant response of a user's main eye, a difference in the perception response of a non-gaze area to the disturbance stimulus is collected to determine the current dominant state of the main eye; based on the micro-pressure difference distribution of different quadrant regions of the eye socket, a visual attention focus is predicted in real time, a spatial geometric deformation is performed on a region of interest (ROI) in a visual detection process, and the detection region is dynamically adapted to the gaze shift trend of the user; two physically independent visual detection links, a feature target strong discrimination process driven by the main eye and an environment contour weak confirmation process driven by the auxiliary eye, are respectively constructed, a logical thermal impedance adjustment mechanism is introduced, a buffer time window is set to absorb the processing fluctuations caused by state switching, and based on the statistical information of the main eye identification frequency, a mapping relationship between the eye identification dominant frequency and the user intention type is constructed to dynamically match the visual detection process path.

Description

Technical Field

[0001] This invention relates to eye recognition and visual detection, specifically a real-time eye recognition and visual detection process control method. Background Technology

[0002] Existing technologies, such as Chinese patent CN111949131A, describe an eye-tracking interaction method, system, and device based on eye-tracking technology. Although the use of a sensing area adsorption mechanism and eye movement behavior feature judgment improves the stability and accuracy of the interaction, this approach still has several significant technical limitations and application drawbacks, specifically in the following aspects: First, this technology judges the interaction intention based on whether the gaze cursor enters the sensing area and combines parameters such as nystagmus and saccade distance. This is a passive judgment mechanism based on the gaze point and behavioral parameter thresholds. It lacks the ability to identify and analyze the dynamic dominance between the dominant and secondary eyes during eye movement, and it cannot identify the difference in the user's binocular responses to disturbance stimuli. Therefore, it cannot achieve dynamic differentiation of the visual dominant path, resulting in the recognition of task intentions relying on the spatial layout and setting of external sensing targets, and lacking sufficient response to changes in the user's internal visual preferences and cognitive state.

[0003] Secondly, this patent employs a template-matching intent model learning method, obtaining a user's subjective eye-tracking intent model through offline training for subsequent prediction. This method has limited flexibility and cannot dynamically adapt to sudden state changes in real-time task environments. Especially in multi-task switching and high-frequency interaction scenarios, the original model cannot adjust its structure or perceptual weights in real time, easily leading to misjudgment or false triggering of instructions, thus affecting task continuity and the stability of visual detection. Furthermore, although this patent uses machine learning to train and model eye-tracking data, its focus is still on the extraction and thresholding of static behavioral features, without introducing a dynamic state vector structure with strong time correlation. Therefore, it cannot capture the temporal evolution trajectory of user eye-tracking behavior, nor can it extract trend signals of task behavior switching. This makes the method difficult to cope with user attention drift. The system's ability to match and control complex visual processes is further affected by highly dynamic behaviors such as shifting and ambiguous task intent. Furthermore, the system architecture lacks a physical separation mechanism for the primary and secondary eye detection links, and does not propose an asynchronous link fusion judgment scheme. In the process of multi-source visual perception data fusion, the lack of a clear hierarchical process structure can easily lead to priority conflicts between key target recognition and environmental background contour recognition, thereby interfering with the output stability of the decision module. At the same time, the method lacks a response mechanism for changes in physiological characteristics after user visual fatigue or prolonged fixation. It cannot provide real-time feedback on the user's current perceptual state through physiological micro-features such as pupil tension, fixation inertia, or eye movement frequency. Therefore, when faced with situations such as cognitive overload or visual sluggishness, the system cannot adopt protective strategies such as delayed response, path buffering, or forced task closure, posing a potential risk of interactive fatigue.

[0004] Furthermore, because its recognition mechanism fails to integrate the synergistic coupling relationship between eye rhythm, ROI space jump trend and pupil tension, and lacks the ability to dynamically model the coupling of state vectors, it cannot achieve deep self-matching of the process during task path selection. The system can only react based on a fixed interaction model and lacks an adaptive process scheduling strategy, thus limiting its scalability and robustness in complex multimodal task scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time eye recognition and visual detection process control method, thereby solving some of the drawbacks and shortcomings pointed out in the background art.

[0006] The present invention addresses the aforementioned technical problems by employing the following technical solution: a real-time eye recognition and visual detection process control method, comprising: applying temporal edge perturbation stimulation in the visual field to induce a spontaneous explicit response of the user's dominant eye; collecting the perceptual response differences of the user's two eyes to the perturbation stimulation in non-fixation areas to determine the current dominance state of the dominant eye; based on the micro-pressure difference distribution in different quadrants of the user's eye socket, inferring the visual attention center in real time, and performing spatial geometric deformation on the region of interest (ROI) in the visual detection process accordingly, so that the detection area dynamically adapts to the user's fixation shift trend; Based on the dominance state of the dominant eye and the determination result of the visual attention center, two physically independent visual detection links are constructed: a feature target strong discrimination process driven by the dominant eye and an environment contour weak confirmation process driven by the secondary eye. The links are executed asynchronously, and their respective detection results are synchronized to the upper-level control module for fusion decision-making. When frequent alternation of dominance between primary and secondary eyes is detected, a logical thermal impedance adjustment mechanism is introduced based on the above visual detection link to delay the switching response of the visual process and set a buffer time window to absorb the processing fluctuations caused by the state switching. Based on the statistical information of the primary eye recognition frequency, the user's current task behavior pattern is inferred, and the visual detection process path is dynamically matched by constructing a mapping relationship between eye dominance frequency and user intent type.

[0007] Furthermore, the temporal lateral perturbation stimulus includes a periodic reverse pattern sequence with rotational micromotion properties, used to induce phase locking of the dominant eye in motion pattern perception; during the dominant eye identification process, short-term interference intervention is used to observe the eye-specific recovery path, so as to determine the eye with visual recovery tendency as the dominant eye.

[0008] Furthermore, the task division of the detection chain between the primary and secondary vision eyes is based on adjusting the link perception intensity weights according to the user's past task preferences, and is driven by behavioral experience for dynamic reconstruction. When signal gaps appear in both chains at the same time, the system constructs a virtual visual chain, merges the historical gaze path data of both eyes to generate a simulated detection stream, and temporarily fills the gaps in the primary and secondary chains.

[0009] Furthermore, the thermal impedance adjustment mechanism includes multi-dimensional situational awareness inputs, including eye movement frequency, gaze micro-drift amplitude, and binocular pupil defocus, which are used to dynamically adjust the impedance curve shape in real time. When the thermal impedance mechanism is activated for more than a threshold time and the visual target is not achieved, the system switches to a forced monocular dominance state to suppress the detection chain of the other eye to achieve task closure.

[0010] Furthermore, the core of the real-time visual process adaptive control method based on eye switching and pupil dynamic response lies in: constructing a task type state vector by combining eye switching rhythm, ROI jump direction, and pupil tension response to achieve deep self-matching of the process; simultaneously, allowing users to explicitly invoke predefined process templates via eye-tracking commands in unpredictable states to manually intervene in and cover the visual detection process; to achieve the above objectives, the ternary feature set is mapped to a dynamic state representation through a nonlinear operator, and temporal weights and an asymmetric coupling mechanism are introduced, so that the state representation not only captures the current visual intent but also predicts potential behavioral trends; the task type state vector is defined as: in: This is a task type state vector, used to represent the user's state over time. The overall visual intent state; Let be the eye-specific rhythm switching function, representing the time interval... The instantaneous frequency difference of the switching between primary and secondary visual dominance; This is the ROI jump direction operator, representing the dynamic offset trend of the region of interest in two-dimensional space, and is taken in the form of a vector gradient. The pupillary tension response function is defined as the dynamic change of the minute amplitude tension of pupil constriction / dilation over time. Here are the feature weight coefficients, where Used to non-linearly amplify the sensitivity of the eye to changing rhythms and states. and The contributions of spatial attention shift and pupillary physiological feedback were measured separately. This is a time-series decay factor used to ensure that recent features have a higher weight in state calculation than distant features. The time decay constant; This is an asymmetric coupling operator, indicating that a non-linear interaction is used when combining the three types of features, rather than a simple weighted superposition. The integration process characterizes the dynamic cumulativeity of the state vector, that is, the evolution of features from the past to the present moment jointly determines the current state; The task type state vector is calculated using this formula. It can simultaneously integrate the temporal rhythm of eye switching, the spatial dynamics of ROI, and the physiological signals of the pupil in a single expression; the state vector is used as input into the process path matching strategy, enabling the system to adaptively select the detection path according to the user's implicit intention; in unpredictable environments, users can force the state vector result generated by the formula to be overridden through eye movement commands, directly waking up the target process template, realizing the complementary fusion of adaptive and manual intervention. Constructing task type state vectors In the process, the key physiological and behavioral features in the user's visual intent are integrated into a dynamic continuous expression; considering that the state is not only affected by the instantaneous signal, but also by the cumulative effect of the historical feature evolution trajectory, the state function is first constructed using an integral structure. Starting from the three core features, we first define The eye-switching rhythm reflects the alternating dominance between the dominant and secondary eyes and is an important time-series parameter for assessing user attention stability. Because users' eye-switching can exhibit sudden intensifications or reversals in certain scenarios, this parameter is elevated to a non-linear form to enhance the system's sensitivity to such rhythmic abrupt changes. ,in The weights that can amplify the impact of fluctuating behavior on the state vector; secondly... , representing the offset trend of the ROI jump direction in two-dimensional space, its gradient expression can capture the continuous jumping behavior of the region of interest in the horizontal or vertical direction, reflecting the target center of gravity shift characteristics of the user in the visual search process, therefore a coefficient is introduced. To adjust its contribution to the ability to predict state offset direction; Third item Characterizing pupillary tension response is actually a low-frequency and continuous physiological reflex signal, which is more unconscious than eye movements, therefore... The amplitude root distribution of its micro-tension is extracted to avoid interference from high-frequency physiological noise on state expression, while introducing... As a physiological weighting coefficient that regulates the degree of its participation; The three items mentioned above are heterogeneous at both the physiological and behavioral levels, therefore they cannot be simply summed up. Instead, they are determined through... Represented as an asymmetric coupling operator, it indicates that the interaction between features has a directional and priority scheduling mechanism (such as in eye-tracking dominant states). Priority action, pupil-dominant state (weight increased significantly) To reflect the dominant role of temporal recency in current behavior, a temporal decay term is introduced. ,in It is a positive real number that controls the decay rate. This represents the historical distance from the current time; therefore, the closer the behavioral feature is to the current time, the greater its weight, achieving dynamic weighted memory of recent behavior. Finally, the aforementioned weighted coupled features are used as the integrand function for the time period from the past to the current moment. By performing integration, we obtain the continuously evolving state vector. It includes both the accumulated behavioral history over time and the ability to map and predict the current user's visual intent.

[0011] Furthermore, a feature interaction tensor is introduced during the construction of the state vector, and the time delay difference between the eye switching frequency and the ROI jump direction is embedded as a dynamic parameter in the state generation process to capture the coupling relationship between user response and intent; a tension persistence inertia index is introduced into the pupil tension response to measure the micro-tension trend during fixation and predict the critical state before attention drift.

[0012] Furthermore, a behavior coherence threshold is set before the state vector is constructed. When there is amplitude conflict or frequency decoupling in the ternary features, the system pauses the process matching to avoid erroneous decisions. The system feeds back the jump direction trend in the state vector to the ROI geometric mapping unit, so that the subsequent ROI deformation structure is pre-adjusted in the main direction of the matching vector.

[0013] Furthermore, the process path matching strategy is based on the formation of vector mutation trajectories from multi-period state vectors to identify user task switching trends; when the state vectors exhibit a long-term stable state, the system enters a process steady-state fault-tolerant window, which allows short-term matching deviations to not trigger process jumps.

[0014] Furthermore, the triggering rule of the eye-tracking command adopts a method based on simultaneously encoding the process template index and activation mode type for each action sequence, which is used for high-density multi-task scheduling; when the eye-tracking command sequence is triggered, the consistency score between the current state vector and the command template bound state vector is compared first. If the similarity is insufficient, the system delays the execution of the template to prevent erroneous triggering.

[0015] The real-time eye recognition and visual detection process control method of this invention has significant beneficial effects. It induces the dominant eye's explicit response by applying specific perturbations in the visual field and accurately identifies the dominant eye's dominance state by combining the perceptual differences in non-fixated areas. Simultaneously, it introduces micro-pressure differences in the orbital quadrants to achieve real-time inference of the visual attention center, enabling the region of interest (ROI) to dynamically deform in space to adapt to fixation shift trends, thereby improving the matching degree of the detection process to user visual behavior. By constructing physically independent links for parallel execution and fusion decision-making through a strong discrimination process driven by the dominant eye and a weak confirmation process driven by the secondary eye, the system's visual processing stability and sensitivity in complex scenarios are significantly enhanced. By introducing a logical thermal impedance mechanism to delay frequent switching responses and setting a buffer time window, the processing fluctuations caused by the alternation of dominant and secondary eyes are suppressed. Furthermore, the system combines the dominant eye recognition frequency to infer user behavior patterns to dynamically match the detection path, thereby enabling the system to have task adaptive capabilities. In addition, the system integrates rotational micro-motion perturbation graphics in the dominant eye recognition to achieve motion pattern perception locking, and uses short-term interference intervention to identify and recover the path, enhancing the recognition accuracy of the true dominant eye. The system employs a dynamic reconstruction of the detection chain based on user behavior preferences, constructs a virtual visual chain to fill information gaps when the double chain fails, adjusts thermal impedance morphology through multi-dimensional situational awareness input, and forces monocular control closure procedures during task lag, ensuring stable operation even in situations with high behavioral uncertainty. Simultaneously, it constructs task-type state vectors using eye-switching rhythm, ROI jump direction, and pupillary tension response, combining asymmetric coupling and temporal weighting mechanisms to ensure the state representation has behavioral prediction capabilities. Furthermore, it enhances user intervention control by covering the process path with eye-tracking commands. The state vector construction process incorporates feature interaction tensors and tension persistence. The inertia index characterizes the coupling trend between intention and response. When behavioral features conflict, it avoids erroneous decisions by pausing matching through behavioral coherence threshold and feeds back the jump trend to the ROI structure for pre-adjustment, enabling the system to have forward-looking adaptability. Multi-cycle state vector tracking forms vector mutation trajectories to identify task switching trends, and a steady-state fault-tolerant window is set to shield short-term erroneous offsets. In eye-tracking command control, a process template coding mechanism and consistency scoring strategy are introduced to avoid false triggering. The whole method is organically integrated in terms of real-time performance, accuracy, stability, and user controllability, comprehensively improving the human-computer visual interaction capability under complex tasks. Attached Figure Description

[0016] Figure 1 This is a functional relationship diagram of the real-time eye recognition and visual detection process of the present invention.

[0017] Figure 2 This is a simplified diagram showing the relationship between the primary and secondary vision recognition and detection process in this invention.

[0018] Figure 3 This is a flowchart illustrating the adaptive visual detection process driven by the multimodal state vector of this invention.

[0019] Figure 4 This is a simplified flowchart of the dominant eye determination and multi-link detection in Embodiment 1 of the present invention.

[0020] Figure 5 This is a flowchart of the state vector-driven multi-target visual detection process of the AR system in Embodiment 2 of the present invention. Detailed Implementation

[0021] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0022] Combined with appendix Figure 1This invention provides a real-time eye recognition and visual detection process control method. First, a temporal edge perturbation stimulus with specific directionality and positional control is applied to the visual field in front of the user. This perturbation stimulus is located on the outer visual boundary outside the user's fixation area, preferably in the temporal region where the dominant eye typically has stronger perceptual sensitivity. It is applied instantaneously in the form of optical flash, image shift, or contrast perturbation to induce an unconscious overt response from the user's dominant eye, such as slight eyeball micro-adjustment, abrupt changes in pupil constriction amplitude, or instantaneous visual tracking. The system simultaneously acquires the micro-response differences of the user's two eyes to this perturbation stimulus in the non-fixation area using a high-frequency camera module. Preferably, the fixation micro-drift angle, pupil diameter change rate, and instantaneous potential fluctuation of the orbicularis oculi muscle of the left and right eyes at the moment of perturbation are extracted. A multi-channel difference comparison algorithm is used to normalize and analyze the response parameters to determine the current dominant eye's dominance state. After identifying the dominant eye, the system further acquires the surface micro-pressure distribution signals of the four quadrants (upper outer, upper inner, lower outer, and lower inner) in the user's eye socket region. These signals are perceived in real time through a micro-pressure-sensitive array deployed inside the smart glasses or head-mounted device. The system calculates the current visual attention center position using a spatial pressure difference vector field. This attention center position is used as a dynamic center reference and input to the Region of Interest (ROI) geometric mapping module in the visual detection process. The ROI mapping module performs spatial geometric deformation operations on the original detection area based on the spatial vector offset relationship between the current gaze center and the predetermined detection target. Preferably, it adjusts the ROI boundary contour through nonlinear scaling, affine tilting, bending clipping, etc., so that it dynamically and adaptively follows the user's gaze offset trend. This allows the visual detection process to adjust the detection focus and detection range in real time according to the user's subtle behavioral intentions without the user needing to explicitly gaze or operate.

[0023] After completing the identification of the dominant eye's state and the inference of the visual attention center, the system uses this judgment result as the basis for process assignment, initiating two physically independent visual detection links. These links consist of a strong feature target discrimination process dominated by the dominant eye and a weak environmental contour confirmation process dominated by the secondary eye. The former focuses on high-resolution recognition of the fine structure, key features, shape boundaries, and color contrast of the main target region in the current visual task, preferably employing high-frequency image sampling and high-confidence threshold strategies for the fine-grained judgment process. The latter, based on the secondary eye's perception capabilities, performs low-interference confirmation of background contours, spatial boundaries, and low-contrast targets in a wide-angle visual field, adapting to... For assisted understanding tasks in complex or dynamic scenarios, the two links are physically deployed with independent image channels and processing nodes. Logically, their respective processes are started and stopped by the gaze states of the primary and secondary eyes, ensuring task separation rather than competitive execution. The links run in parallel asynchronously to avoid process blockage or image resource conflicts caused by switching of primary and secondary eye states or attentional drift. After each link completes its discrimination task, the two links synchronize their detection results to the upper-level vision control module. This control module introduces a multi-source information fusion strategy to dynamically match, coordinate conflicts, and adjust confidence weights of the results from the two paths, ultimately forming a unified visual decision output.

[0024] In a continuously operating visual inspection pipeline, when the system detects frequent alternations in dominance between the dominant and secondary eyes within a short period—that is, the dominance states of the dominant and secondary eyes frequently switch within multiple time slices, and the switching threshold exceeds a set steady-state range—the system determines that it is currently in an unstable eye-discrimination state. Subsequently, based on the deployed strong dominant eye discrimination pipeline and weak secondary eye confirmation pipeline structure, a logical thermal impedance adjustment mechanism is introduced to control the switching delay of the visual inspection process and stabilize the process state. This mechanism draws on the principle of physical thermal resistance, limiting the immediate triggering of process reconstruction after a change in eye dominance by setting a process switching cooling interval. Specifically, the system sets a buffer time window for each dominant / secondary eye state switch. Within this time window, the current process does not immediately terminate or reconstruct, but continues to maintain the previous dominant state. The detection path absorbs processing fluctuations caused by state switching, thereby avoiding frequent process interruptions, resource switching conflicts, and false triggering. Simultaneously, the system continuously counts the number of recognitions and the duration of dominance of the dominant eye per unit time, generating a dominant eye recognition frequency curve. This curve is then correlated with the user's behavior data. By analyzing the distribution of this frequency feature sequence, the system infers the user's current task behavior mode, such as fine recognition mode, rapid inspection mode, or fuzzy search mode. Based on this, a mapping relationship between eye-dominance frequency and user intent type is constructed. This relationship is preferably modeled using a non-linear dynamic mapping strategy, enabling the system to perceive the user's task requirements based on eye-control characteristics, thereby dynamically matching the most suitable visual detection process path and executing actions such as process adjustment, parameter switching, and resource redirection.

[0025] Combined with appendix Figure 2The temporal lateral perturbation stimulus not only includes spatial lateral offset features but is also specially designed as a periodically opposite pattern sequence with rotational micro-motion attributes. This pattern sequence consists of two regions on either side of the visual field, each presenting a dynamic pattern dominated by micro-rotation. Their rotation directions and period parameters are set opposite, and their phase initiation is asynchronous, thus creating competitive motion perception input in the visual system. Utilizing the dominant eye bias-locking mechanism in the human eye's perception of contradictory motion, the system guides the user's dominant eye to generate a stronger phase-sensory match with one side of the pattern sequence, i.e., a phase-locking response occurs. This response can manifest as a brief directional fixation or a micro-saccade tending towards the locked side. The system detects the occurrence of this response. Frequency and bias intensity determine the initial candidates for the dominant eye. Furthermore, a short-term interference intervention mechanism is introduced into the dominant eye recognition process. That is, at the critical stage of recognition, interference perturbation is added to the current visual input, such as background inversion, color jitter, or contrast threshold jump. After causing visual interference, the recovery path of the user's eye is observed in a short period of time. That is, it is evaluated which eye recovers to the previous visual rhythm, gaze direction, or micro-movement state faster after the perturbation is removed. The eye with the recovery advantage is determined as the current dominant eye. The system cross-validates the dominant eye recognition result based on the dual signals of visual recovery tendency and rotation lock response. It is suitable for scenarios where the difference between dominant and sub-eye abilities is small or there is state drift, and the dominance is quickly determined.

[0026] The detection chain tasks driven by the primary and secondary vision are not only structurally independent, but their task division also has the ability to adaptively adjust according to user behavior habits. That is, the perception intensity weights between the detection chains are not fixed, but dynamically adjusted based on the user's eye-movement behavior data in past tasks. The system tracks the user's eye-specific usage preferences in specific task types (such as target search, image recognition, or scene inspection) over a long period of time, extracting the behavioral tendencies shown in the execution of tasks between the primary and secondary vision. For example, parameters such as the average dwell time of the primary vision when judging high-contrast target features and the micro-jump frequency of the secondary vision in spatial edge perception are used as empirical basis for adjusting the perception weights of the links. This allows the functional distribution of the detection chains to dynamically self-fit with the user's historical behavior path, thereby forming a task division mechanism driven by behavioral experience, and periodically reconstructing the response priority and resource allocation within the links. The system employs a matching strategy to achieve deep adaptation of the visual detection architecture to individual user differences. Furthermore, in special circumstances, such as when both the primary and secondary vision-driven detection chains experience input signal gaps (i.e., neither obtains effective image features or gaze feedback within a short period, for example, due to visual distortion caused by blinking, occlusion, or complex dynamic backgrounds), the system activates a virtual visual chain construction mechanism. By recalling historical data of binocular gaze paths over a recent period, the system extracts the gaze focus evolution trajectory and ROI traversal order, combines it with the background semantic information of the current scene to simulate the path, and generates a simulated detection stream as a temporary substitute. This simulated stream is injected into the detection control process to maintain perceptual continuity. This virtual visual chain continuously provides stable detection outputs until the real detection chain recovers, and achieves smooth switching by comparing the confidence level with the real chain results, effectively filling the visual gap problem caused by the lack of primary and secondary chain signals.

[0027] The thermal impedance adjustment mechanism introduced in response to the frequent alternation of dominant and sub-eye dominance employs a nonlinear dynamic adjustment strategy to stabilize the switching response of the visual detection process. It dynamically controls the transition between dominant and sub-eye dominance by constructing a response-impedance relationship curve. The shape of this impedance curve is not statically set, but is generated in real time by multi-dimensional situational awareness input, which includes three key indicators: eye movement frequency, gaze micro-drift amplitude, and binocular pupil defocus. Eye movement frequency reflects the user's current visual exploration rhythm. When the frequency increases, the system determines that the user is in a rapid search or alert state, and correspondingly reduces the slope of the impedance curve to improve switching flexibility. The gaze micro-drift amplitude is used to judge gaze stability. When the amplitude is in the high fluctuation range, it indicates that the user's visual intention has not yet stabilized, and the system increases the impedance value to delay the process switching. As a physiological indicator of binocular visual fusion ability, when this indicator is abnormally high, it means that the perceptual difference between the dominant and secondary eyes has increased. To prevent false triggering, the system raises the impedance curve as a whole and extends the buffer zone to ensure stability. In addition, the system also has a thermal impedance activation time monitoring. When the impedance mechanism is continuously activated for more than a set threshold time and the target of the current visual detection task has not been achieved, such as unsuccessful target recognition or unclosed response, the system determines that it is currently in a high-blocking and low-efficiency state, and then triggers the forced switching logic to switch the visual process from binocular alternating drive mode to monocular dominance mode. The eye with the higher previous dominance rate is selected as the dominant eye, and the execution authority of the detection chain driven by the other eye is forcibly suspended or suppressed, so that the task is completed independently in the dominant eye chain.

[0028] Combined with appendix Figure 3 This approach maps multimodal physiological signals and behavioral features generated by users during visual interaction into predictive state vectors, which drive the deep adaptive evolution of the visual detection process. A dynamic state vector model integrating three key features (eye-switching rhythm, ROI jump direction, and pupillary tension response) is constructed to achieve process path matching and control. Simultaneously, it allows users to explicitly invoke predefined process templates through active eye-tracking commands in unexpected scenarios for process coverage and manual intervention.

[0029] The task type state vector is defined using the following mathematical expression: in, The task type state vector reflects the user's state at the current moment. The overall visual intent is expressed. Function terms. This represents the difference in switching frequency between the dominant and secondary visual eyes, used to quantify the fluctuations in the dominant rhythm of visual attention between the two eyes; to enhance the system's response to sudden switching patterns, this function introduces a power exponent. ,pass Strengthen the weighting amplification of nonlinear rhythm fluctuations; second term It is the ROI jump direction gradient operator, used to capture the offset trend of the user's gaze region of interest in two-dimensional space, and weighted by the jump behavior in this space. This enables directional prediction of visual search trajectories; the third item Representing pupillary tension response, as a physiological indicator of unconscious attentional regulation, it is expressed in square root form. The energy amplitude of micro-tension is extracted to suppress high-frequency noise and highlight low-frequency stability, with weighting coefficients. Adjust its level of participation. The above three characteristics are expressed through symbols. This represents an asymmetric coupling operation, where a nonlinear interaction mechanism is introduced during feature combination, resulting in nonlinear weighted superposition, thereby achieving dynamic switching between priority scheduling and state-dominant factors. Furthermore, the time weighting factor... This reflects the temporal decay characteristics of behavioral memory, among which As a positive real number, it controls the rapid decay of the impact of long-term behavior, ensuring that the state vector is more dependent on recent behavior signals. Ultimately, by adjusting the time interval... The dynamic integral of all integrand feature terms generates the state vector. This enables the temporal continuity modeling of user visual intent.

[0030] This state vector not only integrates heterogeneous data from physiological and behavioral levels, but also possesses the ability to predict potential user behavioral trends. It can be directly used as input to the visual process path matching module, enabling the system to achieve optimal matching and adaptive adjustment of process strategies without explicit control commands. In unpredictable and complex scenarios, such as when the system's decoding of the state vector becomes uncertain, users can actively trigger preset process templates through rapid eye-tracking commands (such as binocular convergence or continuous fixation) to override the current state-driven outcome.

[0031] To achieve a deeper and more forward-looking model of user visual intent, a feature interaction tensor mechanism is introduced into the construction of task type state vectors. Its core lies in capturing the temporal correlation and behavioral response coupling patterns between different visual physiological features. Specifically, by constructing a time delay difference parameter between the eye-switching frequency and the ROI jump direction, this parameter is dynamically embedded into the state vector generation process to identify the lead-lag behavior structure in user attention control. This involves determining whether there is a synchronous shift between the dominant eye-controlled switching and the ROI migration. For example, if the dominant eye switching behavior frequently lags behind the ROI jump, it indicates instability in the user's visual execution path. The system records the trend of this difference over time through a tensor structure and integrates it into the state representation, thus forming the state vector. It provides temporal-domain interaction supplements, thereby significantly improving its ability to interpret complex visual intentions; in addition, it introduces a tension persistence inertia index in the pupil tension response section. This index measures the temporal inertia of pupillary microtension during sustained fixation. Its definition considers the consistency of the rate of change of pupillary tension per unit time with the direction of continuity. If the index remains highly stable within a certain period, it indicates that the user's attention has not yet drifted. However, if a critical abrupt change in trend occurs, the system can predict the upcoming shift in attentional focus. Therefore, by... Fusion to In this project, the state representation is made capable of perceiving the critical state before the transfer of visual attention, thereby completing the adjustment and preparation of the detection path in advance in the process control, avoiding the lag response and perception gap of the visual process, and improving the overall detection efficiency and task closure rate.

[0032] To ensure high behavioral consistency and policy reliability in the state vector-driven visual detection process, a behavioral coherence threshold determination mechanism is introduced before constructing the task-type state vector. This mechanism detects whether there are amplitude conflicts or frequency decoupling among the three feature groups (eye-switching rhythm, ROI jump direction, and pupillary tension response). Amplitude conflicts refer to situations where the features exhibit inconsistent directions or cancel each other out in terms of numerical trends within the same time period. For example, the eye-switching rhythm tends to be stable while the ROI direction jumps drastically, or pupillary tension continuously increases while the fixation area tends to be still. These reflect an incoordination in the user's visual behavior. Frequency decoupling refers to a lack of phase synchronization in the change rhythms of the three features. For example, one feature is in a high-frequency oscillation state while other features remain in a low-frequency stable state. The system establishes behavioral coherence thresholds by setting cross-feature frequency difference thresholds and co-directionality indicators. The coherence determination model pauses the state vector matching process and subsequent visual detection path scheduling if the current behavior does not meet the coherence threshold condition. This avoids path decision errors or abnormal activation due to signal conflicts. The matching process is restarted only after the features are re-coupled to ensure system stability. Simultaneously, after the state vector is constructed, the system feeds back the trend information about the ROI jump direction to the ROI geometric mapping unit in real time. This trend information indicates the future spatial migration direction of the user's region of interest in the form of the main vector direction. Based on this, the mapping unit pre-adjusts the structure of the region of interest (ROI). Specifically, it performs nonlinear geometric deformation along the dominant direction of the state vector on the basis of the original ROI boundary, so that the next detection region has directional adaptation capability, thereby completing the reconstruction of the detection region in advance before the user's gaze is about to drift.

[0033] To improve the adaptive accuracy of the visual inspection workflow and its dynamic perception of user task states, a path matching strategy based on multi-period state vectors is proposed. Its core mechanism lies in the analysis of task type state vectors generated over consecutive time periods. Perform time-series stacking and trajectory mapping to construct a vector-variant trajectory sequence. By analyzing the trend of the trajectory in the high-dimensional state space, the system determines whether the user's visual behavior is in a stable execution phase or whether a potential task intent shift has occurred. Specifically, when the state vector exhibits behavioral characteristics such as gradient slippage, abrupt changes in direction, or magnitude jumps over multiple cycles, the system identifies this as a task switching trend trigger condition, matches a preset process template, and prepares to execute a process path jump. When the state vector sequence maintains structural stability, consistent direction, and low magnitude fluctuations over a long period, the system determines that the user is in a continuous focused state and enters the process steady-state fault-tolerant window mode. In this mode, a tolerance mechanism is implemented for short-term state vector deviations, i.e., a floating threshold range is set. When the state vector deviates locally but does not exceed this tolerance range, the system does not immediately respond to the process switching command but maintains the original process to continue execution, in order to prevent erroneous process jumps caused by non-intent disturbances such as pupil fluctuations and micro-eye movements.

[0034] To meet the flexible workflow scheduling requirements in high-density, multi-tasking environments, an eye-tracking instruction encoding triggering mechanism is proposed. Each set of eye-tracking action sequences not only serves as a trigger control signal but also carries the index information of the workflow template and the encoded content of the corresponding activation mode. Specifically, each set of eye-tracking instructions consists of several temporally related eye-tracking micro-behaviors, including rapid saccades, fixed-point gaze, and short-range jumps. Each combination mode is mapped to a specific visual detection workflow template number according to a preset encoding rule and further bound to the activation type (such as immediate switching, delayed warm-up, and parallel wake-up) to form a high-density, distinguishable, and polymorphic control instruction set, enabling users to achieve reusable scheduling and dynamic switching of workflows solely through eye movements. After the user issues an eye-tracking instruction sequence, the system does not immediately execute the template invocation operation but first processes the current task type state vector. The template reference state vector bound to this instruction Similarity comparison is performed, and a comprehensive scoring mechanism such as inner product, cosine of the angle, or feature-weighted Euclidean distance is used to measure the semantic consistency between the two. If the score is lower than the set threshold, it indicates that there is a deviation between the current visual intent state and the tuned template. The system will then delay the execution of the template to prevent the activation of the erroneous process caused by user accidental touch or non-intent eye movement. The process switching will be formally triggered after the continuous recognition scores tend to converge or reach stable consistency.

[0035] Example 1: Combined with appendix Figure 4In this embodiment, User A is a technical engineer operating a high-precision remote robotic arm. Daily, User A needs to perform multi-point visual positioning, recognition, and feedback through a visual interface. Their dominant eye determines the signal path of the core target recognition link. Before the system starts, it is not clear whether User A's dominant eye is their left or right eye. This recognition needs to be completed within 1.5 seconds without disrupting their workflow. At this point, the system enters the dominant eye determination sub-process, using a perturbation stimulus mechanism for rapid identification.

[0036] The system displays a holographic interface in front of the user (covering the temporal edge at an angle of 35° to 55°), simultaneously injecting two sequences of rotating micro-motion stimuli. A set of rotating sequences with opposite periodic characteristics is projected onto the left and right edges respectively; for example, the left image rotates clockwise at a low speed of 0.3Hz, and the right image rotates counterclockwise at a medium speed of 0.5Hz. The image type consists of a black and white spiral texture superimposed with pseudo-random flashing dots, and the frequency difference creates a phase-resistance condition. The stimulation duration is 800ms. During the acquisition period, a binocular eye tracker records the relative response density and reflection frequency (unit: spike / sec) of both eyes to the stimuli in the non-fixational center area.

[0037] The data sampling results are as follows: The average frequency of micromotion response of the left eye to the stimulus pattern on the side within 0–800 ms was 12.4 spike / sec, with an average reaction delay of 96 ms. The average frequency of response of the right eye to the stimulus pattern in the same period was 5.7 spike / sec, and the average reaction delay was 143 ms. Furthermore, the micro-motion peak spectrum of the left eye is phase-locked with the rotation period of the stimulation pattern on the left, with a phase shift of less than 0.07π. The right eye, however, showed phase-locking failure, with a mismatch between the peak response frequency and the stimulus frequency, and a shift greater than 0.31π.

[0038] The system employs a phase consistency function. Phase-locked loop (PLL) scoring is performed, among which... This corresponds to the phase of the dominant frequency of eye movement micro-movements. This corresponds to the rotation phase of the perturbation pattern. Left eye. right eye Combining the response delay difference ΔT = 143ms – 96ms = 47ms and the response frequency difference ΔF = 12.4 – 5.7 = 6.7 spike / sec, a perceived advantage index is constructed: Based on the score comparison, the left eye's dominance is significantly higher than that of the right eye, and it is tentatively determined to be the dominant eye.

[0039] To further verify whether the dominant eye's manifestation is stable, the system introduced a short-term interference stimulus (e.g., an interference bar flashing in the middle of the screen, with the interference phase conflicting with the dominant side pattern) within the next 300ms. This interference was removed immediately after lasting 80ms, and the system continued to monitor the reconstructed response trajectories of both eyes within an 800ms recovery window.

[0040] The recorded results show that the left eye rapidly resumed phase-locked behavior within 240ms after the interference was removed, and the phase stabilized within 0.09π; while the right eye did not show a stable phase-locked trend and exhibited phase oscillations. Considering the recovery time difference ΔTr = 330ms (right) – 240ms (left) = 90ms, the system further confirmed that the left eye had a stronger visual recovery capability based on the stability of the recovery path, ultimately identifying the left eye as the dominant eye and binding the left eye detection link to the main path flow of the feature target.

[0041] The system has determined that the left eye is the dominant visual eye through a temporal perturbation stimulation mechanism, and has constructed a strong feature target discrimination link dominated by the dominant visual eye (left) and a weak environmental contour confirmation link dominated by the secondary visual eye (right) in the visual process control module. As the complexity of the task increases, for example, when the user needs to perform rapid welding point confirmation and surrounding environment heat source avoidance judgment at the same time, the system begins to introduce a task perception preference adjustment mechanism to perform personalized dynamic reconstruction of the dual-chain task division.

[0042] In the initial stage of task execution, based on A's past 42 task operation records, the system found that when performing tasks involving the localization of high-precision small targets, A relied more on the dominant eye. This was manifested in the dominant eye's eye movement frequency reaching 0.87Hz, micro-movement span less than 0.9°, and ROI fixation time significantly higher than the secondary eye. However, in wide-angle environment scanning tasks, the secondary eye showed increased involvement, with average pupillary tension fluctuations less than the dominant eye. Based on this data, the system set the perception intensity weights according to the empirical vector during the task initialization phase as follows: As the task progresses to the middle stage, the user needs to simultaneously judge a rapidly moving end of the operating arm and a newly emerging background interference signal. During this process, because the user gazes at the left side of the screen while simultaneously and rapidly and frequently scanning the right side, the distribution of eye movements in both eyes becomes abnormally sparse. This causes the system to identify that both the main and secondary visual chains have entered a signal void region, and neither eye can stably capture effective target features for a period lasting 280ms. At this point, the system triggers a virtual visual chain construction mechanism, fitting binocular fixation path data from the past 6 seconds (a total of 62 frames sampled), constructing a spatiotemporal fixation trajectory using Bezier curves, and then generating a simulated detection stream by combining it with a historical saliency heatmap of the ROI. This stream is used to perform virtual contour discrimination of the predicted region, temporarily filling in the gaps in the main and secondary visual chains before the real signal recovers. According to calculations, the virtual stream achieves an average hit rate of 76.5% on the real target region, effectively preventing process interruption or recognition failure due to visual chain breaks during this period.

[0043] Towards the end of the task, due to prolonged high eye-tracking load, the user's pupil defocus fluctuations increased, the average focal shift of the left eye increased to 0.43 mm, the fixation micro-drift amplitude rose to 2.1°, and the eye movement frequency decreased to 0.42 Hz. The system identified the current situation as being at the fatigue threshold, and the thermal impedance adjustment mechanism was immediately activated. This mechanism integrates the above three types of situational data to generate a thermal impedance curve and performs impedance state assessment based on the following nonlinear function: in This represents the current eye movement frequency. To observe the micro-drift variance, The pupil defocus difference is set by an empirical coefficient. Calculate the current The system value has exceeded the system threshold of 4.5 and the duration has reached 380ms, exceeding the set tolerance threshold of 300ms. The system therefore determines that forced intervention is required to stabilize the process. Thus, it enters a forced monocular dominance state, immediately freezes the weak confirmation chain of the right eye, retains only the dominant eye detection channel, and forcibly closes the current process.

[0044] Example 2: Combined with appendix Figure 5 In this embodiment, User B is an aviation equipment maintenance personnel who needs to visually identify and input commands for multiple micro-defects on the wing using a head-mounted augmented reality display system. The system constructs a task type state vector based on real-time eye-tracking data, gaze behavior, and pupil changes to regulate the detection process path. To adapt to the dynamics of individual visual behavior, the system adopts the following state function: Specifically, the data was sampled based on user B's average behavior within a 5-minute operation window: The eye-switching rhythm function is The unit is Hz, which represents the difference in switching frequency between the primary and secondary eyes; The gradient of the ROI transition direction vector is The unit is deg / s, which represents the spatial jumping trend of the region of interest; The pupillary tension response function is The unit is mm² / s², which represents the micro-tension fluctuation of the pupil. The constant parameter is set as follows: Simplify the formula, substitute the components, and perform integration: Substitute the example: 1. Regarding the first item, Expand and approximately integrate; 2. In the second item, Substituting this into the equation yields the exponentially modulated cosine integral. 3. In the third item, It can be numerically integrated.

[0045] The result calculated after fitting using sample points is: The approximate value of the first integral is ≈10.62. The approximate value of the second integral is ≈ 4.41 The approximate value of the third integral is ≈ 8.35. Attenuation factor is The final state vector value is obtained as follows: At this point, the system determines that the state vector is in the mid-to-high-level task attention matching region (the threshold range is set to [0, 2.5] for normal, and above 2.5 for high-order active intervention state), and therefore starts the high-resolution ROI fine recognition process template according to the set strategy.

[0046] During execution, a sudden mechanical interference caused screen tremors. The system, through its eye-tracking command detection module, identified three consecutive rapid left-right-left saccades by the user. This sequence was mapped to an emergency zoom-in search template in the user-defined template. The system compared the cosine similarity of this command state vector with the currently calculated state vector. The matching value was 0.71 (above the threshold of 0.65), confirming the command's validity. The system immediately terminated the current process and loaded a large-scale contour search process, expanding the ROI detection domain to 1.8 times the original area.

[0047] The system has identified the dominant eye (left eye) and entered the complex target recognition state. Immediately following this, it initiates refined modeling of the state vector and optimization of process control. First, the system constructs the task type state vector. Previously, a feature interaction tensor mechanism was added, which is based on the eye switching frequency. and ROI change direction Time delay difference Embedded as a dynamic parameter in the state generation process, this captures the coupling relationship between user responses and intentions. Specifically, if user A's eye-switching frequency increases from 0.8Hz to 1.2Hz over a certain period, and the peak change trend of the ROI direction shows a time lag of approximately 120ms, this delay value is introduced into a simplified interaction tensor term: in Normalization coefficients (e.g.) Substitute into the calculation: This tensor value is then combined with the original feature input to construct the state vector, significantly improving the capture of behavioral synchronization.

[0048] Secondly, in the pupillary tension response process, a tension persistence inertia index is introduced. Used to measure the trend stability of micro-tension during fixation. For example, user A measured pupillary tension values ​​of 1.05, 1.08, and 1.10 (units arbitrary) three times consecutively during a stable fixation phase, with each time interval being 50 ms. The formula (unit tension change per second) can be used: Once the threshold of 0.3 is exceeded, the system determines that a critical drift is about to occur, and therefore adjusts the detection strategy in advance, reducing the ROI and increasing the detection refresh frequency.

[0049] Next, the system sets a behavioral coherence threshold before performing state vector matching to detect whether the ternary features maintain coherence. For example, currently... However, if the eye switching frequency increases dramatically to 2.0Hz, while the ROI jump slows down to 0.05deg / s, and the amplitude and direction of the three are not consistent (high-frequency switching but no response from the ROI), the system considers it an amplitude conflict. At this time, the process path matching is paused, and the system waits for the features to recover before proceeding to the next matching step.

[0050] Finally, the system feeds back the trend vector in the state vector regarding the ROI transition direction, such as a main direction of +30° (shifted to the right), to the ROI geometric mapping unit in real time to adjust the structure of the region of interest for the next detection cycle. Setting the current ROI as a square region centered ±5°, the system pre-adjusts it to be 2° to the right of the center and its width increased to ±6° to cover the user's gaze migration range in advance, thereby improving the predictability and detection accuracy of the workflow response.

[0051] Following the previous stage where the system completed the main eye recognition and state vector construction based on the behavior interaction mechanism, user A then enters a process of recognizing multiple micro-operation targets. For example, moving from task point P1 to P2 and then discovering a new temporary target P3, the system activates a process path matching strategy based on multi-cycle state vectors, generating a state vector sequence over 3 consecutive seconds (recorded once every 0.5 seconds, for a total of 6 sampling points): , , , , , Its high-dimensional representation shows that the direction and amplitude are basically consistent, with a standard deviation of only ~0.075, indicating that the user's visual intent is in a highly stable state. Therefore, the system triggers the steady-state fault-tolerant window mechanism, setting the time window to 2 seconds. Even if slight state fluctuations (such as...) are received within the window, the system will be able to handle the situation. It will not immediately jump to a different process, but will continue to execute the current fine path, reducing false jumps caused by temporary blurred eye movements.

[0052] Following this, user A's eye movement accidentally triggered a rapid left-right-fixed-point level 3 action. The system interpreted this as an activation instruction for a panoramic review template, which was mapped to number 7. The activation method was a parallel warm-up. Therefore, an eye movement instruction mapping encoding sequence was constructed, in which template index 7 and activation type encoding were combined to form 7a. Before actually activating the template, the system extracted the average value of the current state vector (3.0) and the template-bound state vector. Calculate the similarity between the two (using cosine similarity): If the similarity is higher than the set threshold of 0.95, it is decided to immediately preheat the template in parallel, allowing users to launch new lookaround scan tasks simultaneously without interrupting the current process; if the similarity is low (e.g. The system delays the activation of the template until the status values ​​converge again to prevent accidental operation. In this example, the high similarity allows the new template to intervene immediately, expand the ROI in real time, and start scanning the look-around target P3 required for operation and maintenance. At the same time, the original path continues to execute without process conflicts.

[0053] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A real-time eye recognition and visual detection process control method, characterized in that... include: Temporal edge perturbation stimulation is applied in the visual field to induce the spontaneous explicit response of the user's dominant eye. The perceptual response differences of the user's two eyes to the perturbation stimulation in non-fixation areas are collected to determine the current dominance state of the dominant eye. Based on the micro-pressure difference distribution in different quadrants of the user's eye socket, the visual attention center of gravity is inferred in real time, and spatial geometric deformation is performed on the region of interest (ROI) in the visual detection process to dynamically adapt the detection area to the user's fixation shift trend. Based on the dominance state of the dominant eye and the determination result of the visual attention center, two physically independent visual detection links are constructed: a feature target strong discrimination process driven by the dominant eye and an environment contour weak confirmation process driven by the secondary eye. The links are executed asynchronously, and their respective detection results are synchronized to the upper-level control module for fusion decision-making. When frequent alternation of dominance between primary and secondary eyes is detected, a logical thermal impedance adjustment mechanism is introduced based on the above visual detection link to delay the switching response of the visual process and set a buffer time window to absorb the processing fluctuations caused by the state switching. Based on the statistical information of the primary eye recognition frequency, the user's current task behavior pattern is inferred, and the visual detection process path is dynamically matched by constructing a mapping relationship between eye dominance frequency and user intent type.

2. The real-time eye recognition and visual detection process control method according to claim 1, characterized in that... The temporal lateral perturbation stimulus includes a periodic reverse pattern sequence with rotational micromotion properties, used to induce phase locking of the dominant eye in motion pattern perception; during the dominant eye identification process, short-term interference intervention is used to observe the eye-specific recovery path in order to identify the eye with visual recovery tendency as the dominant eye.

3. The real-time eye recognition and visual detection process control method according to claim 1, characterized in that... The task division of the detection chain between the primary and secondary vision is based on the user's past task preferences, which adjusts the link perception intensity weight and adopts behavioral experience-driven dynamic reconstruction. When signal voids appear in both chains at the same time, the system constructs a virtual vision chain, merges the historical gaze path data of both eyes to generate a simulated detection stream, and temporarily fills the gaps in the primary and secondary chains.

4. The real-time eye recognition and visual detection process control method according to claim 1, characterized in that... The thermal impedance adjustment mechanism includes multi-dimensional situational awareness inputs, including eye movement frequency, gaze micro-drift amplitude, and binocular pupil defocus, which are used to dynamically adjust the impedance curve shape in real time. When the thermal impedance mechanism is activated for more than a threshold time and the visual target is not achieved, the system switches to a forced monocular dominance state to suppress the detection chain of the other eye to achieve task closure.

5. The real-time eye recognition and visual detection process control method according to claim 1, characterized in that... The mapping relationship of the user intent type is to use eye switching rhythm, ROI jump direction, and pupil tension response as a ternary feature group to construct a task type state vector to achieve process depth self-matching; in the process path matching strategy, users are allowed to forcibly evoke the process template through eye movement commands to deal with manual intervention in unpredictable states.

6. The real-time eye recognition and visual detection process control method according to claim 5, characterized in that... The state vector is constructed by introducing a feature interaction tensor, and the time delay difference between the eye switching frequency and the ROI jump direction is embedded as a dynamic parameter in the state generation process to capture the coupling relationship between user reaction and intention; the pupil tension response introduces a tension persistence inertia index to measure the micro-tension trend during fixation and predict the critical state before attention drift.

7. The real-time eye recognition and visual detection process control method according to claim 6, characterized in that... Before constructing the state vector, a behavior coherence threshold is set. When there is amplitude conflict or frequency decoupling in the ternary features, the system pauses the matching process to avoid incorrect decisions. The system feeds back the jump direction trend in the state vector to the ROI geometric mapping unit, so that the subsequent ROI deformation structure is pre-adjusted towards the main direction of the matching vector.

8. The real-time eye recognition and visual detection process control method according to claim 7, characterized in that... The process path matching strategy is based on multi-cycle state vectors to form vector mutation trajectories, which are used to identify user task switching trends. When the state vector exhibits a stable state over a long period of time, the system enters the steady-state fault-tolerant window of the process, which allows short-term matching deviations to not trigger process jumps.

9. The real-time eye recognition and visual detection process control method according to claim 8, characterized in that... The triggering rule of the eye-tracking command adopts a process template index and activation mode type that are simultaneously encoded for each action sequence, which is used for high-density multi-task scheduling. When the eye-tracking command sequence is triggered, the consistency score between the current state vector and the command template binding state vector is compared first. If the similarity is insufficient, the system delays the execution of the template to prevent false triggering.

Images