A low-perception health guidance method based on behavior rhythm resonance of a desk scene

Abstract

The application discloses a low-perception health guidance method based on behavior rhythm resonance of an office table scene, and is applied to a system comprising an office table and office equipment carrying a visual sensor. The method comprises the following steps: acquiring a user visual data stream through the visual sensor, and establishing a reference health posture model locally; identifying a repetitive micro-motion cycle of the user, and analyzing characteristic changes to establish a personal behavior rhythm model; combining the two models, and predicting the fatigue risk of a specific muscle group in real time through a time sequence prediction algorithm; when the risk exceeds a threshold value, generating a multi-mode guidance instruction set synchronized with the current action cycle of the user, so as to guide the user to perform micro-motions for relaxing target muscle groups. The application realizes non-interfering, personalized and accurately targeted preventive health intervention, and significantly improves the compliance and acceptance of the intervention.

Description

Technical Field

[0001] This invention relates to the fields of human-computer interaction and health assistance technology. Specifically, it relates to a user behavior rhythm analysis and low-perception health guidance method based on computer vision, which is particularly suitable for office scenarios. By analyzing the user's micro-movement cycle, it predicts the risk of muscle fatigue and generates synchronous low-perception guidance signals. Background Technology

[0002] With the increasing prevalence of office automation and sedentary work, musculoskeletal strain among professionals is becoming increasingly prominent. Existing healthy office assistance technologies mainly include software with timed reminders and smart office furniture with real-time posture correction or preset programs, but all of them have limitations.

[0003] Software solutions that provide timed reminders intervene in user behavior by popping up notifications at fixed times. The main drawback of this type of solution is that the preset intervention timing is completely out of sync with the user's real-time physiological load, work content, and concentration state. Forced interruptions can disrupt the flow state of work, resulting in a poor user experience and often low user compliance.

[0004] Real-time posture correction solutions based on sensor feedback typically detect poor posture by deploying sensors in seats or wearable devices and providing feedback through vibration or alarms. This type of solution is essentially a reactive approach, only issuing warnings after poor posture has already occurred, and cannot prevent the accumulation of muscle fatigue. Furthermore, the alarm signals themselves constitute significant interference, and the singular feedback mode can easily lead users to habitually ignore them.

[0005] Smart office furniture with pre-programmed settings, such as height-adjustable desks that automatically adjust their height according to a schedule, offers physical posture changes, but its adjustment patterns are fixed and lack intelligence. Its operating rhythm cannot adapt to the user's actual work pace and fatigue cycles, making it difficult to provide timely intervention at the moments when the user needs it most but is least sensitive to it.

[0006] In summary, existing technologies generally design health interventions as behaviors that require explicit user perception and response to external commands, failing to achieve deep integration with the user's natural workflow. Therefore, there is an urgent need for a new human-computer interaction method that can deeply understand and adapt to individual work rhythms, predict fatigue risks, and provide precise and synchronized guidance with extremely low perceptibility. Summary of the Invention

[0007] To address the problems of existing technologies, the main objective of this invention is to provide a low-perception health guidance method based on behavioral rhythm resonance in a desk setting. This method aims to solve the problems of abruptness, inefficiency, and lack of foresight in existing intervention methods. By establishing a personal behavioral rhythm and health posture model, it can predict the fatigue risk of specific muscle groups and generate multi-mode guidance instructions that are synchronized with the user's current action cycle and have an intensity below the perception threshold. This allows for non-intrusive, personalized, and precisely targeted pre-emptive preventive health intervention without the need for dedicated hardware.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A low-awareness health guidance method based on behavioral rhythm resonance in a desk setting, applied to a system including a desk and office equipment, wherein the office equipment is equipped with a visual sensor, includes the following steps: S1: Acquire the user's visual data stream through the visual sensor mounted on the office equipment, and establish a baseline health posture model of the user based on the visual data stream in the local device; S2: Based on the visual data stream, identify the user's head tilting motion cycle in front of the desk as a repetitive micro-motion cycle, and analyze the characteristic changes of the continuous cycle in the local device to establish a personal behavioral rhythm model of the user. S3: Based on the personal behavioral rhythm model and the baseline health posture model, predict the risk probability of muscle fatigue and the associated target muscle groups of the user in real time through a time-series prediction algorithm. S4: When the risk probability exceeds a preset threshold, a multi-mode guidance instruction set synchronized with the user's current action cycle is generated based on the associated target muscle group; the guidance strategy is configured to guide the user to perform reverse or relaxing micro-movements that can soothe the target muscle group.

[0009] As a preferred embodiment, step S1 includes: The visual data stream is processed by pose feature extraction to obtain a multi-frame pose data sequence describing the spatial relationships of the main parts of the user's body; wherein, the pose data sequence includes coordinate data representing the position of the body parts; During the first time period in which the user does not receive any guidance instructions, the first statistical distribution of each joint angle and the second statistical distribution of the torso tilt angle are calculated based on the posture data sequence. The central values ​​and normal fluctuation ranges of the first and second statistical distributions are defined as the baseline healthy posture model.

[0010] As a preferred embodiment, the posture data sequence includes head joint coordinates; step S2 includes: Based on the head joint coordinates in the visual data stream, the continuous change in head spatial position from below the first threshold to above the second threshold is identified as the head pitching motion cycle.

[0011] As a preferred embodiment, step S2, establishing a personal behavioral rhythm model, includes: Calculate the duration of N consecutive head pitch cycles to obtain a duration sequence; where N is a preset number greater than or equal to 3. Extract the cervical spine joint angle sequence as a function of time within each cycle; The mean and variance of the duration sequence, and the dominant frequency component of the angle sequence in the frequency domain, are calculated to form the personal behavior rhythm model.

[0012] As a preferred embodiment, step S3 includes: The current feature vector output by the personal behavior rhythm model, and the difference vector between the current joint angle calculated from the posture data sequence and the corresponding center value in the benchmark health posture model, are input into a neural network model. The neural network model outputs the probability value of muscle fatigue occurring within the next M time windows, and outputs a classification label identifying the target muscle group.

[0013] As a preferred embodiment, the generation of the classification label for the target muscle group is also based on a biomechanical mapping table, which predefines the correspondence between different joint angle combinations and the main force-bearing muscle groups; the system queries the mapping table according to the current joint angle combination, verifies the queried muscle group with the classification label output by the neural network model, and jointly determines the associated target muscle group.

[0014] As a preferred embodiment, step S4 includes: Based on the associated target muscle group, a preset guidance strategy library is queried, which stores guidance signal combinations and synchronization phase parameters corresponding to different target muscle groups; Based on the query results, the set of boot instructions, which includes display control instructions and audio control instructions, is generated.

[0015] As a preferred embodiment, the display control command is configured to control the graphical interface elements to generate a linear displacement lasting T seconds, wherein the spatial direction of the linear displacement is preset according to the direction of the relaxation movement required by the target muscle group; the audio control command is configured to generate a beam of narrowband noise with a center frequency of F Hz.

[0016] As a preferred embodiment, the multi-mode boot instruction set is executed by controlling the display unit and audio output unit of the office equipment: The display unit drives the graphical interface elements to move according to the display control instructions in the guidance instruction set; The audio output unit outputs the corresponding acoustic signal according to the audio control commands in the guidance command set.

[0017] As a preferred embodiment, the method further includes step S5: During the second time period after the guidance strategy is executed, the user's subsequent visual data stream is collected through the visual sensor; Based on the subsequent visual data stream, the change in joint angle corresponding to the associated target muscle group is calculated; If the change is less than a preset effect threshold, then when generating a guidance strategy for the same target muscle group, the intensity or duration parameter of the guidance signal in the guidance instruction set is enhanced.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) In this invention, by establishing a baseline health posture model and an individual behavior rhythm model, the system completes the quantitative modeling of individual differences in posture and work mode locally, so that subsequent guidance has the effect of personalized adaptation and natural synchronization. This is based on the user's own standards and rhythms to make decisions, thereby significantly improving the individual compliance and acceptance of the intervention.

[0019] (2) In this invention, the risk prediction algorithm in real time constitutes the transformation of the solution from passive response to active prevention. By analyzing the continuous and dynamic deviation of personal behavioral rhythms relative to the health benchmark, the risk trend can be identified before the physical symptoms of muscle fatigue or strain appear, resulting in a prospective intervention effect. The health management node is moved from post-correction to pre-prevention, which reduces the probability of cumulative damage from the root.

[0020] (3) In this invention, a multi-mode guidance instruction set synchronized with the current action cycle is generated, and the abstract prediction result is transformed into a specific, low-perception physical guidance. By precisely synchronizing the triggering of the guidance instruction with the user's own micro-action cycle, and designing the guidance form as a micro-movement to soothe the target muscle group, not only is the user's perceptual load and sense of interruption greatly reduced, but the guidance behavior is also integrated into the user's current workflow almost naturally. Moreover, it achieves a leap from generalized reminders for the whole body or posture to precise biomechanical regulation for specific muscle groups, and the accuracy and efficiency of the intervention are substantially improved. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a method according to an embodiment of the present invention.

[0022] Figure 2This is a schematic diagram of one application scenario of the method according to an embodiment of the present invention.

[0023] Figure 3 This is a second schematic diagram illustrating the application scenario of the method according to an embodiment of the present invention.

[0024] Attached image labels: 10, office desk; 20, office equipment; 30, auxiliary camera. Detailed Implementation

[0025] To better illustrate the objectives, technical solutions, and advantages of the present invention, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0026] like Figures 1 to 3 The diagram illustrates a low-awareness health guidance method based on behavioral rhythm resonance in a desk 10 scenario, according to an embodiment of the present invention. This method is applied to a system comprising a desk 10, office equipment 20, and an auxiliary camera 30, wherein the office equipment 20 is equipped with a visual sensor. The method includes the following steps: S1. The visual data stream of the user is acquired through the visual sensor and auxiliary camera 30 mounted on the office equipment 20, and a baseline health posture model of the user is established on the local device based on the visual data stream.

[0027] Specifically, the visual sensor may be a camera integrated above the display of the office device 20. After acquiring the visual data stream, a lightweight human posture estimation algorithm (e.g., a posture estimation model based on the MobileNetV2 backbone network) is run on a local device (such as a computer) to extract the two-dimensional or three-dimensional coordinate sequence of the user's main body joints from each frame of the image, forming a machine-understandable representation of the visual data stream. An auxiliary camera 30 is used to supplement the acquisition of the visual data stream. To establish a baseline healthy posture model, the system prompts the user to adjust to a comfortable posture during the initial calibration phase (e.g., the first 5 minutes of the user's first use), and acquires multiple frames of posture data during this period. Subsequently, the algorithm calculates the mean and standard deviation of the angles of each key joint (e.g., cervical lordosis angle, lumbar flexion angle, elbow angle) during this period, using the mean as the user's health baseline value and the standard deviation range as the normal physiological fluctuation range, together constituting their personalized baseline healthy posture model.

[0028] It's worth noting that the lightweight pose estimation algorithm uses a model tailored from the OpenPose architecture. Its output of 11 joints includes: nose, left eye, right eye, left ear, right ear, left shoulder, right shoulder, left elbow, right elbow, left wrist, and right wrist. The intrinsic parameter matrix K obtained through camera calibration and the known user sitting height range (e.g., setting the initial height of the hip joint from the table to H = 50 ± 10 cm) serve as prior constraints for the PnP algorithm, solving for the 3D coordinates of the user joints in the camera coordinate system.

[0029] S2. Based on the visual data stream, identify the repetitive micro-movement cycles of the user in front of the desk 10, and analyze the characteristic changes of the continuous cycles in the local device to establish a personal behavioral rhythm model of the user.

[0030] Specifically, when identifying repetitive micro-movement cycles, the system continuously analyzes the motion trajectories of key points in the visual data stream. Taking the common head-tilt cycle as an example, the algorithm tracks the vertical coordinates of key points such as the tip of the nose or the center of the forehead. When the value is below an adaptive threshold (e.g., the historical average position minus an offset), it is marked as the starting point of head-down movement; when it rises above another threshold, it is marked as the ending point of head-up movement, thus segmenting a single cycle. To establish a personal behavioral rhythm model, the system continuously records the duration (cycle T) of, for example, 20 consecutive cycles (N=20) during the user's normal work period and extracts the curve of cervical spine angle changing over time within each cycle. By calculating the mean, variance, and autocorrelation characteristics of cycle T, and extracting the dominant frequency from the cervical spine angle curve using Fourier transform, these time-domain and frequency-domain features are vectorized, thus constructing a personal behavioral rhythm model that quantifies the user's inherent work rhythm.

[0031] For each identified single movement cycle (such as the i-th head pitch cycle), its duration T_i is extracted. The cervical lordosis angle data sequence θ_i(t) sampled at equal intervals within this cycle is subjected to a Hanning window followed by a Fast Fourier Transform (FFT), and the frequency corresponding to the largest amplitude value in the amplitude spectrum is taken as the dominant frequency component f_i of that cycle. The feature vector of N consecutive cycles can be represented as: F = [mean(T), var(T), mean(f), var(f), ...], where mean and var represent the mean and variance, respectively.

[0032] S3. Based on the personal behavioral rhythm model and the baseline health posture model, the user's risk probability of muscle fatigue and the associated target muscle groups are predicted in real time using a time-series prediction algorithm.

[0033] Specifically, during real-time prediction, the system concatenates the latest feature vector (e.g., features from the previous 5 cycles) output by the individual's behavioral rhythm model with the difference vector between the joint angles calculated from the current posture and the baseline health posture values ​​in the benchmark health posture model. This concatenated vector is then input into a pre-trained temporal classification model (e.g., a one-dimensional convolutional neural network combined with a long short-term memory network). The model outputs two core results: first, a probability value (a scalar between 0 and 1) indicating the risk of muscle fatigue within the next 5 minutes; and second, a classification label for a target muscle group (e.g., the upper trapezius or sternocleidomastoid). This classification is based on the model's learned association between posture features and muscle group biomechanics; for example, a sustained forward head tilt is more likely to be associated with fatigue in the posterior neck muscles.

[0034] S4. When the risk probability exceeds a preset threshold, a multi-mode guidance instruction set synchronized with the user's current action cycle is generated based on the associated target muscle group; the guidance strategy is configured to guide the user to perform reverse or relaxing micro-movements that can soothe the target muscle group.

[0035] Specifically, when the risk probability exceeds a preset threshold (e.g., 0.7), the system queries a pre-defined biomechanical guidance strategy database based on the target muscle group determined in S3. This database defines reverse or relaxing micro-movements and corresponding low-perception guidance methods for different muscle groups. For example, for the upper trapezius, the strategy is to induce a shoulder-sinking movement. Subsequently, the system accurately calculates the phase of the user's current movement cycle and generates a guidance instruction set containing multi-modal instructions at a specific phase of the next cycle (such as the instant the head-up movement begins). For example: 1) Visual instruction: controls the operating system to slowly move the current focus window diagonally downwards at a speed of 2 pixels per second, inducing the user to unconsciously sink their shoulders to follow the content; 2) Auditory instruction: generates a pink noise pulse that starts synchronously with the head-up movement, lasts for 1 second, and has a center frequency of 150Hz, to enhance the feeling of relaxation through auditory cues. The intensity of the instructions is pre-calibrated to ensure that it is below the user's perception threshold.

[0036] S5. During the second time period after the execution of the guidance strategy, the user's subsequent visual data stream is collected through the visual sensor; based on the subsequent visual data stream, the change in joint angle corresponding to the associated target muscle group is calculated; if the change is less than a preset effect threshold, when generating a guidance strategy for the same target muscle group, the intensity or duration parameter of the guidance signal in the guidance instruction set is enhanced.

[0037] Specifically, within a short window (e.g., 30 seconds) after the completion of the guidance instruction set, the system re-collects user posture data via visual sensors and recalculates the major joint angles associated with the target muscle group. This angle is compared with the baseline angle recorded before guidance to obtain the angle change. If this change is less than a preset effect threshold (e.g., a desired angle improvement of 50%), the guidance effect is deemed insufficient. The system does not immediately perform a second strong intervention, but instead marks the next guidance decision module targeting the same muscle group in the algorithm with a need for enhancement. When the muscle group is identified as high-risk and requires guidance again, the system will, within a sub-perceptual intensity range, prioritize selecting a previously unused guidance pattern combination from the strategy library, or, without exceeding the perception threshold, slightly increase the duration parameter of the guidance signal (e.g., visual slowing duration) by 20% to achieve adaptive optimization.

[0038] It is worth noting that, through locally running lightweight posture estimation algorithms (such as MobileNetV2-based models) and initial calibration, the system establishes a baseline healthy posture model entirely based on the user's own physiological characteristics. Furthermore, by tracking the trajectories of key points (such as the vertical coordinates of the nose tip) to identify micro-rhythms such as the head pitch cycle, and performing time-frequency domain feature analysis on their duration and angle curves, a quantitative personal behavioral rhythm model is constructed. This technical approach ensures that all subsequent interventions deeply align with the user's individual differences and inherent work rhythm, fundamentally guaranteeing a high degree of personalization and natural synchronization in guidance, and avoiding the incompatibility of generic solutions.

[0039] This system fuses individual behavioral rhythm characteristics with real-time posture deviations relative to a health baseline and inputs them into a temporal prediction model (such as a one-dimensional CNN combined with LSTM) for joint analysis. This enables the system to identify subtle, continuous trend changes that deviate from healthy patterns, thus predicting the risk probability of specific muscle groups (such as the upper trapezius) before physiological symptoms of muscle fatigue or strain appear. This mechanism substantially advances the health intervention point from traditional post-event correction to pre-event prevention, achieving a forward-looking upgrade of the intervention paradigm.

[0040] When the predicted risk exceeds the limit, the system does not issue an abrupt alarm. Instead, it generates a multi-mode guidance instruction set that is precisely synchronized with the user's current action cycle based on a pre-set biomechanical guidance strategy database. For example, by controlling the focus window to slowly move in a specific direction at a sub-perceptual speed (e.g., 2 pixels per second), combined with a synchronously activated specific spectrum sound (e.g., a 150Hz pink noise pulse), the system induces the user to unconsciously perform reverse or relaxing micro-movements (e.g., shoulder shrugs) of the target muscle group through a synergistic approach of visual and auditory channels. This process effectively integrates biomechanical regulation into the user's task flow, achieving precise targeted regulation while minimizing interference with the main task and ensuring a smooth user experience.

[0041] By collecting posture data again and calculating joint angle changes after guidance, the system can objectively evaluate the effectiveness of a single guidance session. If the effect is not as expected, its optimization mechanism does not simply amplify the signal, but adjusts the strategy within the sub-perceptual intensity range, such as switching guidance mode combinations or fine-tuning the duration parameter (e.g., increasing it by 20%). This closed-loop design enables the system to adaptively learn and optimize guidance strategies for specific users, continuously improving the effectiveness of interventions and long-term user compliance, ensuring the method's sustained effectiveness and robustness.

[0042] In summary, this method, through a coherent technical chain of individualized precise modeling, temporal risk prediction, rhythmic synchronous multimodal guidance, and closed-loop adaptive optimization, systematically produces a comprehensive technical effect that is highly personalized, proactively preventive, has low perceived interference, and possesses self-optimization capabilities. It provides a solution for achieving efficient and humane proactive health intervention in office settings.

[0043] Specifically, in this embodiment, the pose feature extraction process is completed on the local device by running an optimized neural network model. This model employs a lightweight encoder-decoder structure. The encoder is a small MobileNetV3 network used to extract high-level semantic features from the input image frames. The decoder consists of a series of deconvolutional layers used to reconstruct the feature maps and output heatmaps of 11 human body joints (such as the nose, left and right shoulders, left and right elbows, etc.). The system obtains the pixel coordinates of each joint by finding the peak points of the heatmaps, and uses the camera's intrinsic parameter matrix and the known user sitting height range as prior constraints. The PnP algorithm is used to estimate the three-dimensional spatial coordinates of the joints, ultimately outputting the pose data sequence describing the spatial relationships of the user's main body parts. All processing is completed locally on the device; the original image frames are released immediately after processing in memory to ensure user privacy.

[0044] The neural network model consists of a one-dimensional convolutional layer (kernel size 3, number of filters 32), a max-pooling layer, an LSTM layer with 64 memory units, and two fully connected output layers. The model's training data comes from publicly available datasets (such as a synchronized dataset of surface electromyography (sEMG) signals and joint angles under specific movements), with time periods where the sEMG signal amplitude exceeds a threshold labeled as "fatigue." The cross-entropy loss function is used, and the Adam optimizer is employed.

[0045] Specifically, in this embodiment, the biomechanical mapping table is a pre-stored, structured lookup table. Its construction is based on publicly available biomechanical research findings and laboratory pre-experimental data. Each row in the table records a joint angle combination pattern and its associated primary muscle group. For example, a combination key value might be {cervical lordosis angle: [25 degrees, 35 degrees], lumbar flexion angle: <10 degrees, right shoulder abduction angle: [0 degrees, 15 degrees]}, and its associated target muscle group is labeled as the right upper trapezius and right levator scapulae. During real-time queries, the system matches the calculated current user joint angle combination (the numerical range of each primary joint angle) with the keys in this table. The matching algorithm uses similarity calculation based on interval overlap. When the row with the highest matching degree is found, the muscle group information recorded in that row is extracted and cross-validated with the classification labels output by the neural network model. If both point to the same muscle group or muscle group, then the muscle group is ultimately determined to be the associated target muscle group. Part of the contents of the biomechanical mapping table are shown in the following table:

[0046] Specifically, in this embodiment, the pre-set guidance strategy library is a relational database table or a JSON-formatted configuration file. Each strategy record contains key fields including: target muscle group ID, guidance signal combination (e.g., visual-linear displacement_auditory-narrowband noise), synchronization phase parameter (e.g., triggered at the 'tilt start' phase of the head pitch cycle, with a delay of 0 milliseconds), and the specific parameter index of each guidance signal. For example, for the sternocleidomastoid muscle, the guidance signal combination of a strategy record might be visual guidance-horizontal slow movement_auditory guidance-low-frequency pulse, with its synchronization phase parameter set to trigger at the midpoint of the 'head rotates to the left' cycle. Based on the determined target muscle group ID, the system quickly retrieves all applicable strategy records through a database query interface or key-value lookup. Then, combining this with the real-time phase information output by the current user's rhythm model, it selects the record with the highest matching degree and generates the final executable display and audio control instruction set based on its stored parameter index.

[0047] Specifically, in this embodiment, the linear displacement lasting T seconds in the display control command has a duration parameter T determined through pre-experimentation to ensure a smooth displacement process and a value below the conscious perception threshold. For example, the value of T is typically set to 3.0 to 5.0 seconds, and the total displacement is controlled between 1% and 2.5% of the screen width. The system searches for the corresponding displacement vector from a predefined direction-target mapping table based on the desired relaxation direction for the target muscle group. For example, to relax the right trapezius muscle, the vector direction might be downward to the left (inducing the user to slightly turn left and lower their right shoulder). The narrowband noise with a center frequency of F Hz in the audio control command is selected according to the principle of non-interference in auditory perception, typically avoiding the main frequency band of human voice (300Hz-3400Hz), and can be selected around 125Hz or 8000Hz, with a bandwidth controlled within 1 / 3 octave. The sound pressure level is strictly set below the ambient noise masking threshold to ensure its presence but not explicit identification as an external prompt. All parameters are calibrated and stored in the strategy library before the product leaves the factory.

[0048] refer to Figure 3 As another preferred implementation of the guidance, the multi-mode guidance instruction set can also control the display unit of the office equipment to render and drive a virtual human skeleton model to perform a demonstration of soothing movements.

[0049] The system pre-configures a simplified 3D skeletal model of the human upper body, which includes key joint nodes corresponding to pose estimation. During initial calibration, the system adaptively scales the skeletal proportions of this virtual model based on the user's baseline healthy posture model, making it roughly match the user's actual body shape.

[0050] When a specific target muscle group needs to be guided, the system queries the guidance strategy library for the corresponding standard joint movement sequence of "reverse or relaxation micro-movements". For example, for the "right upper trapezius", the standard movement sequence is defined as: the virtual model's head turns slightly to the left (<10 degrees) while the right shoulder joint moves downward (about 2-3 cm). The system converts this standard movement sequence into driving data for each joint of the virtual model.

[0051] The system calculates the phase of the user's current action cycle and triggers a guiding command at the appropriate starting point of the next cycle (such as the instant the head begins to lift). The display unit renders a semi-transparent virtual model in a non-core visual area of ​​the screen (such as the lower right corner). This model then demonstrates the aforementioned soothing action in a slow, smooth animation (lasting for, for example, 3-5 seconds).

[0052] To ensure the guidance is conducted at a low level of awareness, the virtual model has a simple appearance (such as wireframe or monochrome rendering), smooth animation speed, and high overall transparency (such as setting the transparency to 70%), and its presentation occurs in sync with the user's own movement rhythm. Users do not need to pay conscious attention; their surrounding vision and motion perception will subconsciously perceive and may imitate the movement pattern, thereby achieving the purpose of soothing guidance.

[0053] The following is an example of the adaptive interaction experienced by a user on a typical workday morning when using an intelligent office system built based on a low-awareness health guidance method for behavioral rhythm resonance in a desk 10 scenario, as described in this invention: 1. Initial Calibration and Learning. The system starts automatically after the user turns on their computer. A screen prompt appears: "Health posture calibration is in progress. Please sit comfortably as usual." For the next 5 minutes, the user browses emails normally. Step S1 uses a lightweight posture estimation algorithm via camera to record multiple posture frames in the user's natural state, calculating their personalized baseline health posture model. For example, the baseline value for the cervical spine forward tilt angle is 18 degrees, with a normal fluctuation range of ±3 degrees.

[0054] 2. Rhythm Modeling and Risk Prediction. A user begins processing a complex Excel report, entering a focused cycle of looking down at the document, looking up at the screen, and typing. Step S2 works silently in the background: by tracking the vertical movement of the user's brow point, it accurately identifies each head-down-up movement cycle as an independent head-tilt cycle. The system continuously analyzed 20 such cycles, finding an average cycle length of 12 seconds and a stable pattern in the cervical spine angle curve, thus establishing a personalized behavioral rhythm model for the user.

[0055] 3. Initial predictive intervention. After working continuously for half an hour, the temporal prediction model (LSTM) in step S3 issued an alert: it fused the user's latest rhythm characteristics (the cycle has been slightly extended to 13.5 seconds) with the current real-time posture (the cervical spine forward tilt angle is maintained at 25 degrees) and predicted that the probability of fatigue of the upper trapezius muscle on the right side within 5 minutes has reached 0.72, which exceeds the preset threshold of 0.7.

[0056] Synchronous guidance occurs: Step S4 responds immediately. It queries the strategy library and determines that the best relaxation strategy for the right upper trapezius muscle is to induce shoulder drop and a slight leftward turn of the head. The system accurately predicts that the user's next head-up movement will begin approximately 1.2 seconds later.

[0057] Seamless guidance: In the instant the user subconsciously looks up, two things happen simultaneously: 1) The Excel window they are working on begins to slowly pan to the lower left of the screen at a rate of 2 pixels per second; 2) A very faint 150Hz pink noise pulse, similar to background white noise, sounds in the headphones, lasting for 1 second before disappearing. The user is completely unaware that this is system guidance; simply to more comfortably see the data moving to the lower left, they unconsciously lower their right shoulder slightly and turn their head about 5 degrees to the left. A gentle micro-movement targeting the muscles is completed unconsciously.

[0058] 4. Effect Evaluation and Adaptive Optimization Thirty seconds after the guided session ended, step (S5) initiated an effectiveness evaluation. It remeasured the user's cervical spine and shoulder angles using a camera and found that the joint angle associated with the right trapezius muscle had only improved by 40% of the expected effect (below the 50% threshold), thus determining that the guided session was insufficient. The step silently tagged the right upper trapezius muscle as needing optimization in the background.

[0059] 5. Adaptive Reboot When the user resumed work, the system again predicted a high risk for the same muscle group. This time, the system (S4, based on the optimized decision from S5) generated a slightly different set of guidance instructions: the direction of the window easing remained the same, but the duration was extended from 3 seconds to 3.6 seconds; simultaneously, instead of using audio signals, it attempted pure visual guidance. This time, the angle improvement reached 65%, and the guidance was successful. The system recorded that the extended visual approach was more effective for this user's muscle group, for use in optimizing future strategy selections.

[0060] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application of the technical solution and the constraints involved. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0061] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0062] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0063] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0064] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0065] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A low-awareness health guidance method based on behavioral rhythmic resonance in a desk setting, applied to a system including a desk and office equipment, wherein the office equipment is equipped with a visual sensor, characterized in that, Includes the following steps: S1: Acquire the user's visual data stream through the visual sensor mounted on the office equipment, and establish a baseline health posture model of the user based on the visual data stream in the local device; S2: Based on the visual data stream, identify the user's head tilting motion cycle in front of the desk as a repetitive micro-motion cycle, and analyze the characteristic changes of the continuous cycle in the local device to establish a personal behavioral rhythm model of the user. S3: Based on the personal behavioral rhythm model and the baseline health posture model, predict the risk probability of muscle fatigue and the associated target muscle groups of the user in real time through a time-series prediction algorithm. S4: When the risk probability exceeds a preset threshold, a multi-mode guidance instruction set synchronized with the user's current action cycle is generated based on the associated target muscle group. The guidance strategy is configured to guide the user to perform reverse or relaxing micro-movements that can soothe the target muscle group.

2. The low-perception health guidance method based on behavioral rhythm resonance in an office desk setting, as described in claim 1, is characterized in that... Step S1 includes: The visual data stream is processed by pose feature extraction to obtain a multi-frame pose data sequence describing the spatial relationships of the main parts of the user's body; wherein, the pose data sequence includes coordinate data representing the position of the body parts; During the first time period in which the user does not receive any guidance instructions, the first statistical distribution of each joint angle and the second statistical distribution of the torso tilt angle are calculated based on the posture data sequence. The central values ​​and normal fluctuation ranges of the first and second statistical distributions are defined as the baseline healthy posture model.

3. The low-perception health guidance method based on behavioral rhythm resonance in an office desk setting, as described in claim 2, is characterized in that... The posture data sequence includes head joint coordinates; step S2 includes: Based on the head joint coordinates in the visual data stream, the continuous change in head spatial position from below the first threshold to above the second threshold is identified as the head pitching motion cycle.

4. The low-perception health guidance method based on behavioral rhythm resonance in a desk setting according to claim 3, characterized in that, In step S2, establishing a personal behavioral rhythm model includes: Calculate the duration of N consecutive head pitch cycles to obtain a duration sequence; where N is a preset number greater than or equal to 3. Extract the cervical spine joint angle sequence as a function of time within each cycle; The mean and variance of the duration sequence, and the dominant frequency component of the angle sequence in the frequency domain, are calculated to form the personal behavior rhythm model.

5. The low-perception health guidance method based on behavioral rhythm resonance in a desk setting according to claim 1, characterized in that, Step S3 includes: The current feature vector output by the personal behavior rhythm model, and the difference vector between the current joint angle calculated from the posture data sequence and the corresponding center value in the benchmark health posture model, are input into a neural network model. The neural network model outputs the probability value of muscle fatigue occurring within the next M time windows, and outputs a classification label identifying the target muscle group.

6. The low-perception health guidance method based on behavioral rhythm resonance in a desk setting according to claim 5, characterized in that, The generation of the classification labels for the target muscle groups is also based on a biomechanical mapping table, which predefines the correspondence between different joint angle combinations and the main force-bearing muscle groups. The system queries the mapping table according to the current joint angle combination, verifies the obtained muscle groups with the classification labels output by the neural network model, and jointly determines the associated target muscle groups.

7. The low-perception health guidance method based on behavioral rhythm resonance in an office desk setting according to claim 1, characterized in that, Step S4 includes: Based on the associated target muscle group, a preset guidance strategy library is queried, which stores guidance signal combinations and synchronization phase parameters corresponding to different target muscle groups; Based on the query results, the set of boot instructions, which includes display control instructions and audio control instructions, is generated.

8. A low-perception health guidance method based on behavioral rhythm resonance in a desk setting, as described in claim 7, is characterized in that... The display control command is configured to control the graphical interface elements to generate a linear displacement lasting T seconds, the spatial direction of which is preset according to the direction of the relaxation movement required by the target muscle group; the audio control command is configured to generate a beam of narrowband noise with a center frequency of F Hz.

9. A low-perception health guidance method based on behavioral rhythm resonance in a desk setting, as described in claim 1, is characterized in that... The multi-mode boot instruction set is executed by controlling the display unit and audio output unit of the office equipment: The display unit drives the graphical interface elements to move according to the display control instructions in the guidance instruction set; The audio output unit outputs the corresponding acoustic signal according to the audio control commands in the guidance command set.

10. A low-perception health guidance method based on behavioral rhythm resonance in a desk setting, as described in claim 1, is characterized in that... The method further includes step S5: During the second time period after the guidance strategy is executed, the user's subsequent visual data stream is collected through the visual sensor; Based on the subsequent visual data stream, the change in joint angle corresponding to the associated target muscle group is calculated; If the change is less than a preset effect threshold, then when generating a guidance strategy for the same target muscle group, the intensity or duration parameter of the guidance signal in the guidance instruction set is enhanced.

Images