Sofa posture self-adaptive adjusting system based on multi-modal sensing fusion

Abstract

The invention relates to the technical field of adaptive control systems, in particular to a sofa posture adaptive adjustment system based on multi-modal sensing fusion, which comprises a baseline modeling module, a monitoring diagnosis module and an intervention decision module. According to the invention, a personalized dynamic characteristic parameter baseline is established for a user through the baseline modeling module, and unconscious attitude deviation and conscious task-oriented actions are accurately distinguished by a micro intention recognition program unit in the monitoring and diagnosis module; and the intervention decision-making module starts adaptive deviation suppression adjustment only when the former is identified. According to the invention, the technical problem that the existing control system cannot distinguish the state change reasons to cause frequent false triggering of the adjusting action is solved, and personalized and intelligent closed-loop control under the condition of avoiding invalid interference is realized.

Description

Technical Field

[0001] This invention relates to the field of adaptive control system technology, specifically to a sofa posture adaptive adjustment system based on multimodal sensing fusion. Background Technology

[0002] Programmable control systems have been applied to everyday equipment such as furniture, but their control logic is usually relatively fixed and lacks the ability to adapt to complex dynamic environments.

[0003] Most existing control systems rely on preset fixed thresholds or simple timing logic to trigger adjustment actions. These systems cannot establish personalized steady-state reference models for different controlled objects, thus their adjustment targets lack specificity. More importantly, existing technologies typically cannot distinguish whether changes in the state of the controlled object stem from systemic state drift or transient disturbances caused by short-term operator commands. This indiscriminate response logic often leads to false triggering at inappropriate times, interfering with normal operation. Furthermore, the parameters of their adjustment process are usually fixed and cannot be dynamically adjusted based on feedback, making it difficult for the system to achieve truly seamless and precise closed-loop control.

[0004] Therefore, those skilled in the art urgently need a novel program control system. This system needs to be able to autonomously learn and establish a personalized dynamic reference baseline for the controlled object, and accurately distinguish between system state drift and task-oriented disturbances by analyzing the timing patterns of signals, thereby executing precise and adaptive closed-loop feedback regulation at the right time to avoid ineffective interference.

[0005] To address this, a sofa posture adaptive adjustment system based on multimodal sensing fusion is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a sofa posture adaptive adjustment system based on multimodal sensor fusion. By establishing a personalized dynamic feature baseline and using a micro-intention recognition program to distinguish between unconscious posture deviations and conscious task-oriented actions, the system can achieve precise and gradual closed-loop feedback adjustment only for unconscious posture deviations without causing unnecessary interference to the user, thereby improving the system's adaptive capability and intelligence level.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A sofa posture adaptive adjustment system based on multimodal sensing fusion includes: Baseline modeling module: Acquires steady-state reference coordinates representing the user's preset working state through a multimodal sensor array, and generates personalized dynamic feature parameter baselines based on the steady-state reference coordinates; Monitoring and Diagnostic Module: Runs the micro-intent recognition program unit, analyzes the temporal change pattern of the steady-state reference coordinate, and determines the user's real-time state changes. Real-time state changes include two types: unconscious posture deviations caused by changes in physiological state and conscious posture changes actively generated by the user to complete short-term tasks. The module continuously compares the user's real-time posture changes with the dynamic feature parameter baseline to detect whether there are any deviations. Intervention Decision Module: When the following two conditions are met simultaneously: First, it is detected that the user's real-time state change has deviated from the dynamic feature parameter baseline; Second, it is determined that the type of deviation is unintentional posture deviation; Control commands are generated and sent to the intelligent actuator to drive the intelligent actuator to perform deviation suppression adjustment and restore the user's state to the state range defined by the dynamic feature parameter baseline.

[0008] Preferably, the baseline modeling module specifically comprises: The calibration procedure is initiated. The multimodal sensor array continuously acquires and records multidimensional signals generated by the user under a stable reference attitude within a preset duration. The set of multidimensional signals is defined as the steady-state reference coordinates. Statistical processing is performed on the parameters of each multidimensional signal in the steady-state reference coordinates to calculate their mean and standard deviation. The mean is combined with a tolerance range set based on the standard deviation to generate the dynamic characteristic parameter baseline, wherein the tolerance range defines the normal fluctuation range around the mean.

[0009] Preferably, the calibration procedure is a user interaction process that issues a command to the user to enter the calibration state and guides the user to maintain the stable reference posture on the sofa. During the entire time period in which the user maintains the reference posture, the multi-dimensional signals are collected synchronously. The multidimensional signals include body pressure distribution patterns acquired by a pressure sensor matrix, electromyographic signals acquired by a surface electromyography sensor, and thermal signals acquired by a passive infrared sensor array.

[0010] Preferably, the micro-intent recognition program unit includes: The signal preprocessing subunit is used to receive the original multi-dimensional signal with timestamps from the multi-modal sensor array, and to perform filtering, noise reduction and time synchronization processing on the multi-dimensional signal to generate time series data in a unified format. The dynamic feature extraction subunit is used to analyze the time series data within a sliding time window and extract a set of dynamic feature vectors that can quantify the rate, amplitude and direction of signal change. The pattern classification subunit is used to input the dynamic feature vector into a pre-trained classification model, and the model outputs a classification result, which determines the current real-time posture change as the unconscious posture deviation and / or conscious posture change.

[0011] Preferably, the pattern classification subunit is specifically used for: The feature vector sequence will be composed of multiple dynamic feature vectors; The classification model is a recurrent neural network model, which is pre-trained on a dataset containing a large number of labeled samples, wherein the samples are labeled as unconscious posture deviations and / or conscious posture changes. The recurrent neural network model processes the feature vector sequence to identify a first type of temporal pattern that represents slow, continuous changes and a second type of temporal pattern that represents rapid, instantaneous changes, and classifies them as unconscious posture deviation and conscious posture change, respectively.

[0012] Preferably, the deviation suppression adjustment is a preset, multi-stage, gradual adjustment process. The intervention decision module generates a set of control command sequences with fixed amplitude and rate parameters based on the detected degree of deviation. The intelligent actuator slowly and continuously changes the geometry of the sofa body at a speed lower than the user's normal perception threshold according to the command sequence until the user's real-time state returns to the range defined by the dynamic feature parameter baseline.

[0013] Preferably, the user's normal perception threshold is a set of parameters dynamically determined through a feedback learning mechanism, specifically: After performing a deviation suppression adjustment, the multi-dimensional signals of the multi-modal sensor array are immediately monitored and analyzed to determine whether the user has generated a sudden attitude change signal. When the user generates the posture change signal, it indicates that the parameters adjusted this time exceed the user's normal perception threshold, and the system will automatically reduce the amplitude and rate parameters in subsequent adjustments. If the user does not generate the sudden change in posture signal, it indicates that the parameter adjusted this time is within the user's normal perception threshold, and the system will maintain and / or fine-tune the current parameter.

[0014] Preferably, the intervention decision module is further configured to: when the monitoring and diagnosis module determines that the user's real-time posture change is a conscious posture change, temporarily suspend the deviation suppression and adjustment function and enter an observation and learning mode; in the observation and learning mode, if the duration of the conscious posture change exceeds a preset threshold, the system automatically collects data under that posture and updates and iterates the dynamic feature parameter baseline.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention overcomes the fundamental flaw of inaccurate adjustment targets in existing technologies by establishing a unique dynamic characteristic parameter baseline for each user. The system no longer forces users to adapt to a fixed, theoretically standard posture, but instead uses the user's own optimal state as the adjustment reference. This highly personalized closed-loop control makes adjustment commands more targeted and effective, accurately adapting to the different body characteristics and usage habits of various users, significantly improving the accuracy and practical application value of the adjustment system.

[0016] 2. This invention solves the technical problem of frequent false triggering of adjustment actions due to the inability to distinguish the causes of state changes by introducing a micro-intent recognition program module. The system can accurately identify unintentional posture deviations that truly require intervention by analyzing the temporal patterns of sensor signals, and can proactively ignore normal, task-oriented actions of the user. This conditional intervention with logical judgment capabilities transforms the adjustment system from a simple mechanical actuator into an intelligent agent that understands context, greatly improving the user experience and the system's practicality.

[0017] 3. The deviation suppression and adjustment method and the dynamic learning mechanism for user perception threshold proposed in this invention differ from fixed and easily perceptible adjustment methods. The adjustment process of this invention is gradual and slow, and its adjustment parameters can be adaptively adjusted according to user feedback, always striving to keep them below the user's normal perception threshold. This adjustment method will not interrupt the user's work or rest, and completes posture optimization in a near-unconscious state, achieving a unity of efficient intervention and ultimate comfort experience, significantly improving the system's user-friendliness and long-term user acceptance. Attached Figure Description

[0018] Figure 1 This is a system flowchart of a sofa posture adaptive adjustment system based on multimodal sensing fusion proposed in this invention; Figure 2 This is a system architecture diagram of a sofa posture adaptive adjustment system based on multimodal sensing fusion proposed in this invention; Figure 3 This is a flowchart illustrating a sofa posture adaptive adjustment system based on multimodal sensing fusion proposed in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 Please see Figures 1 to 3 This invention provides a sofa posture adaptive adjustment system based on multimodal sensor fusion, the technical solution of which is as follows: A sofa posture adaptive adjustment system based on multimodal sensing fusion, such as Figures 1-3 As shown, it includes: Baseline modeling module: Acquires steady-state reference coordinates representing the user's preset working state through a multimodal sensor array, and generates personalized dynamic feature parameter baselines based on the steady-state reference coordinates; Monitoring and Diagnostic Module: Runs the micro-intent recognition program unit, analyzes the temporal change pattern of the steady-state reference coordinate, and determines the user's real-time state changes. Real-time state changes include two types: unconscious posture deviations caused by changes in physiological state and conscious posture changes actively generated by the user to complete short-term tasks. The module continuously compares the user's real-time posture changes with the dynamic feature parameter baseline to detect whether there are any deviations. Intervention Decision Module: When the following two conditions are met simultaneously: First, it is detected that the user's real-time state change has deviated from the dynamic feature parameter baseline; Second, it is determined that the type of deviation is unintentional posture deviation; Control commands are generated and sent to the intelligent actuator to drive the intelligent actuator to perform deviation suppression adjustment and restore the user's state to the state range defined by the dynamic feature parameter baseline.

[0021] Furthermore, the baseline modeling module specifically comprises: The calibration procedure is initiated. The multimodal sensor array continuously acquires and records multidimensional signals generated by the user under a stable reference attitude within a preset duration. The set of multidimensional signals is defined as the steady-state reference coordinates. Statistical processing is performed on the parameters of each multidimensional signal in the steady-state reference coordinates to calculate their mean and standard deviation. The mean is combined with a tolerance range set based on the standard deviation to generate the dynamic characteristic parameter baseline, wherein the tolerance range defines the normal fluctuation range around the mean.

[0022] The baseline modeling module also includes an ergonomic optimization unit, which actively performs a series of minor, exploratory sofa shape adjustments after the calibration program generates the initial dynamic feature parameter baseline, and monitors the feedback of the multi-dimensional signals in real time to find an optimal reference coordinate that allows the user's core muscle groups to maintain a stable state with the lowest energy consumption, and uses the optimal reference coordinate as the final calibration result of the dynamic feature parameter baseline.

[0023] This invention elevates baseline modeling from passive learning to active optimization. It moves beyond simply learning the user's subjectively perceived comfortable posture; instead, through exploratory adjustments and real-time physiological signal feedback, it objectively finds a more scientifically sound healthy posture for the user, one with the lowest core muscle energy consumption. This makes the final adjustment goal not only personalized but also optimized, fundamentally guiding the user to a less fatigue-prone sitting posture and enhancing the system's health value.

[0024] Furthermore, the calibration procedure is a user interaction process that issues a command to the user to enter the calibration state and guides the user to maintain the stable reference posture on the sofa. During the entire time period in which the user maintains the reference posture, the multi-dimensional signals are collected synchronously. The multidimensional signals include body pressure distribution patterns acquired by a pressure sensor matrix, electromyographic signals acquired by a surface electromyography sensor, and thermal signals acquired by a passive infrared sensor array.

[0025] The calibration procedure is initiated and guided via a mobile application that communicates with the system. This application guides the user to a reference posture—with feet flat on the ground, back against the sofa back, upright and relaxed—through its graphical user interface and / or voice prompts.

[0026] During the acquisition process, the multi-dimensional signals are not directly recorded as raw data streams, but key feature values ​​are extracted to construct the steady-state reference coordinates. Specifically, for the body pressure distribution pattern, the two-dimensional coordinates of its pressure center are extracted; for the electromyographic signal, its root mean square value within a specific frequency band is extracted; and for the thermal signal, the number of effective triggers per unit time is extracted.

[0027] The preset duration can be set to 10 to 30 seconds to ensure that the collected feature values ​​reflect the user's stable state rather than instantaneous fluctuations.

[0028] When generating the dynamic characteristic parameter baseline, the tolerance range can be specifically set to the mean of each characteristic value plus or minus 1.5 to 2.5 times the standard deviation of that characteristic value, so as to construct a dynamic monitoring interval that can accommodate normal micro-movements and identify abnormal deviations.

[0029] This invention standardizes the user calibration process through graphical or voice guidance via a mobile application, ensuring the validity and consistency of the initial reference posture. This fundamentally avoids the problem of baseline inaccuracies caused by users' arbitrary sitting postures, which in turn affect the accuracy of all subsequent judgments. The solution clearly defines the specific process of extracting a set of low-dimensional, stable, and information-rich key feature values ​​from high-dimensional, complex raw sensor signals. This feature extraction step not only significantly reduces the computational complexity of subsequent data processing, enabling the system to operate more efficiently, but also improves the robustness of the model by focusing on core information, making it less susceptible to interference from irrelevant signal noise. The dynamic baseline established through statistical methods, containing a clearly defined tolerance range, overcomes the shortcomings of using a single static target point for comparison, which is overly stringent and prone to misjudgment due to normal human body micro-movements. It constructs a "healthy comfort domain" that can tolerate reasonable fluctuations, making subsequent monitoring and diagnostic functions more practical and accurate, and providing a solid and reliable data foundation for the intelligent and adaptive adjustment capabilities of the entire system.

[0030] Furthermore, the micro-intent recognition program unit includes: The signal preprocessing subunit is used to receive the original multi-dimensional signal with timestamps from the multi-modal sensor array, and to perform filtering, noise reduction and time synchronization processing on the multi-dimensional signal to generate time series data in a unified format. The dynamic feature extraction subunit is used to analyze the time series data within a sliding time window and extract a set of dynamic feature vectors that can quantify the rate, amplitude and direction of signal change. The pattern classification subunit is used to input the dynamic feature vector into a pre-trained classification model, and the model outputs a classification result, which determines the current real-time posture change as the unconscious posture deviation and / or conscious posture change.

[0031] Furthermore, the pattern classification subunit is specifically used for: The feature vector sequence will be composed of multiple dynamic feature vectors; The classification model is a recurrent neural network model, which is pre-trained on a dataset containing a large number of labeled samples, wherein the samples are labeled as unconscious posture deviations and / or conscious posture changes. The recurrent neural network model processes the feature vector sequence to identify a first type of temporal pattern that represents slow, continuous changes and a second type of temporal pattern that represents rapid, instantaneous changes, and classifies them as unconscious posture deviation and conscious posture change, respectively.

[0032] In the signal preprocessing subunit, the filtering and noise reduction processing may specifically include: using a 20-500Hz bandpass filter for the electromyography signal, and using a notch filter capable of eliminating power frequency interference for all signals.

[0033] In the dynamic feature extraction subunit, the length of the sliding time window can be set to 2 to 5 seconds, with a 50% overlap rate to ensure the continuity of the analysis. The dynamic feature vector may specifically include: the displacement velocity and acceleration of the pressure center calculated from the pressure distribution pattern, the root mean square value and average power frequency calculated from electromyographic signals, and the trigger frequency per unit time calculated from thermal signals.

[0034] In this embodiment, the classification model in the pattern classification subunit is preferably a Long Short-Term Memory (LSTM) network model, as it is better at capturing long-term dependencies in time series. The training dataset can be constructed by inviting multiple testers to simultaneously record sensor data and manually label it while performing a script task involving pre-set intentional actions and prolonged sitting.

[0035] This invention, through targeted filtering preprocessing, effectively removes environmental noise and artifact interference from the original signal to the greatest extent, ensuring the signal-to-noise ratio and validity of the input data and providing a reliable data foundation for all subsequent analyses. The solution creatively transforms the abstract problem of recognizing human "intention" into an engineering problem of temporal analysis of a set of specific, quantifiable dynamic feature vectors. This feature engineering method makes the classification task's objective clearer and more precise, a key step in achieving high-precision recognition. Furthermore, it selects a Long Short-Term Memory (LSTM) network model, particularly adept at handling long-term dependencies, enabling it to fundamentally and effectively capture and distinguish between "slow, continuous" unconscious deviations and "rapid, instantaneous" conscious changes—two patterns with significant differences in the time dimension. Finally, a clear method for constructing the training dataset provides a clear and feasible path for the implementation of the pre-trained model, ensuring high accuracy and reproducibility of the classification model. This transforms the system's intelligent judgment from a vague "black box" into a well-documented and reliable technical implementation.

[0036] Furthermore, the monitoring and diagnosis module also includes a multi-source information fusion arbitration unit, which is located before the pattern classification subunit. This unit is used to assign different confidence weights to signal features from different types of sensors, and to use evidence theory algorithms to fuse potentially conflicting and / or ambiguous signal information, thereby outputting a unified state judgment result with the highest confidence.

[0037] This invention solves the problem of information conflict or ambiguity that may occur between different sensors in complex scenarios by fusing and arbitrating multi-source sensor information, and significantly improves the accuracy, robustness and reliability of the final state judgment of the system.

[0038] Furthermore, the deviation suppression adjustment is a preset, multi-stage, gradual adjustment process. The intervention decision module generates a set of control command sequences with fixed amplitude and rate parameters based on the detected degree of deviation. The intelligent actuator, according to the command sequence, slowly and continuously changes the geometry of the sofa body at a speed lower than the user's normal perception threshold until the user's real-time state returns to the range defined by the dynamic feature parameter baseline.

[0039] Furthermore, the user's normal perception threshold is a set of parameters dynamically determined through a feedback learning mechanism, specifically: After performing a deviation suppression adjustment, the multi-dimensional signals of the multi-modal sensor array are immediately monitored and analyzed to determine whether the user has generated a sudden attitude change signal. When the user generates the posture change signal, it indicates that the parameters adjusted this time exceed the user's normal perception threshold, and the system will automatically reduce the amplitude and rate parameters in subsequent adjustments. If the user does not generate the sudden change in posture signal, it indicates that the parameter adjusted this time is within the user's normal perception threshold, and the system will maintain and / or fine-tune the current parameter.

[0040] The multi-stage gradual adjustment process can be implemented using a proportional-integral (PI) control algorithm instead of fixed parameters. The intervention decision module uses the detected deviation as input error and continuously generates control commands related to the current magnitude and historical accumulation of the error through a PI algorithm. This allows the adjustment amplitude and rate to dynamically and smoothly adapt to changes in the deviation, thus avoiding the abruptness caused by fixed or segmented parameters.

[0041] The sudden change in posture signal can be quantified as follows: within a short time window (e.g., within 2 seconds) after an adjustment is performed, the displacement velocity of the pressure center coordinate monitored by the pressure sensor matrix, or the rate of change of the root mean square value of the electromyography signal monitored by the surface electromyography sensor, exceeds a preset alert threshold.

[0042] In the feedback learning mechanism, the parameter adjustment can employ an adaptive step-size strategy. Specifically, when it is determined that the user has generated a sudden attitude change signal, the system can multiply the current amplitude and rate parameters by an attenuation coefficient less than 1 (e.g., 0.8); when it is determined that the user has not generated a sudden attitude change signal, the system can multiply by a gain coefficient slightly greater than 1 (e.g., 1.05) to tentatively find the parameter boundaries of optimal adjustment efficiency without alerting the user.

[0043] This invention introduces a proportional-integral control algorithm to dynamically and smoothly correlate the amplitude and rate of adjustment with the degree and duration of user posture deviation. This completely replaces the stiffness and abruptness of fixed parameters or simple segmented adjustments, ensuring that each adjustment precisely matches the actual needs, making the process more natural and human-centered. A clear quantitative definition is provided for the "sudden posture change" signal, which serves as a trigger for feedback learning. This provides an objective and reliable basis for the system to determine whether its adjustment behavior is perceived by the user, avoiding performance instability caused by ambiguous judgment criteria during the learning process. An adaptive step size strategy with attenuation and gain coefficients endows the system with self-optimization capabilities. Like a human expert, the system can intelligently and efficiently converge automatically to the unique "imperceptible adjustment" parameter boundaries for each user through continuous positive trial and error and imperceptible user feedback. Ultimately, this achieves the ideal state of maximizing adjustment effect while minimizing user interference.

[0044] Furthermore, the intervention decision module is configured to: when the monitoring and diagnosis module determines that the user's real-time posture change is a conscious posture change, temporarily suspend the deviation suppression and adjustment function and enter an observation and learning mode; in the observation and learning mode, if the duration of the conscious posture change exceeds a preset threshold, the system automatically collects data under that posture and updates and iterates the dynamic feature parameter baseline.

[0045] The preset threshold can be a duration, such as 10 to 20 minutes, to ensure that the new posture is a stable posture that the user intends to maintain for a long period of time.

[0046] The updating and iteration of the dynamic characteristic parameter baseline can specifically employ a weighted average algorithm. This algorithm uses the newly acquired data representing the new stable attitude as a new instantaneous baseline and fuses it with the original dynamic characteristic parameter baseline using a weighted average. The original baseline has a larger weight (e.g., 0.9-0.95), while the new instantaneous baseline has a smaller weight (e.g., 0.05-0.1). This method achieves smooth and gradual baseline updates to ensure the system's adjustment stability.

[0047] Once the baseline update is complete, or if the user recovers to the attitude range defined by the original baseline ahead of time within the threshold period, the observation and learning mode ends, and the system will automatically reactivate the deviation suppression adjustment function.

[0048] The intervention decision module also includes a deviation type diagnosis and targeted adjustment strategy unit. After determining that the deviation type is an unconscious posture deviation, it further subdivides the deviation into specific posture problem types such as lumbar curvature collapse, pelvic tilt and / or body asymmetric lateral tilt based on specific changes in the pressure distribution pattern. According to the specific posture problem type, it calls a corresponding specific intelligent actuator combination and action sequence from a preset adjustment strategy library to achieve targeted and efficient adjustment of different posture problems.

[0049] This invention elevates adjustment from general restoration to precise targeted intervention. By diagnosing the specific root cause of posture problems and "treating the symptoms," it significantly improves the efficiency, accuracy, and final ergonomic effect of a single adjustment movement.

[0050] This invention, through time threshold filtering and a weighted average update algorithm, enables the system to accurately learn the user's persistent new preferred postures and smoothly and stably iterate its dynamic baseline. This effectively avoids erroneous learning of temporary postures and instability caused by baseline abrupt changes. It significantly improves the harmony of human-computer interaction and the system's personalization level, allowing it to truly and continuously meet the user's dynamic needs.

[0051] This invention overcomes the shortcomings of general models, such as inaccurate adjustment targets and poor adaptability, by establishing personalized dynamic characteristic parameter baselines for users, making adjustments more targeted and effective. More importantly, the system, through a micro-intention mapping recognition program, can accurately distinguish between unconscious postural deviations requiring intervention and normal task-oriented actions of the user, and only performs adjustments when the former is determined. This conditional intervention logic completely avoids ineffective adjustments and operational interference caused by misjudging user intentions, greatly improving the system's intelligence, practicality, and user experience.

[0052] Example 2 This embodiment provides a specific application in a home-based elderly care monitoring scenario. The user can be an elderly person who needs to sit on a sofa for extended periods, has limited mobility, or has weak muscles.

[0053] A family member or caregiver initiates a one-time calibration procedure for the user via a mobile application that communicates with the system. Guided by the calibration procedure, the user is adjusted to a comfortable and well-supported reference posture that conforms to their physical condition. The system then collects data and establishes a personalized baseline of dynamic characteristic parameters. This baseline not only serves as the basis for subsequent adjustments but also as a digital profile reflecting the user's long-term postural health.

[0054] During daily use, the system's monitoring and diagnostic module operates continuously. When the system's built-in micro-intent recognition program unit, based on a recurrent neural network, detects that the user is slowly and unconsciously deviating from their posture due to fatigue or insufficient muscle strength, the intervention decision module will be activated immediately; in this embodiment, unconscious posture deviation can be a sliding or tilting of the body.

[0055] The deviation suppression adjustment employed in this system is an extremely gentle intervention. The system slowly and gradually increases the lumbar support protrusion or slightly adjusts the seat tilt angle at a rate almost imperceptible to the user, providing subconscious support and guidance to help them return to a healthier posture without disturbance. The amplitude and rate of the entire process are dynamically optimized through a feedback learning mechanism: the system monitors the user's response to the adjustment, and if the user makes a sudden movement, it automatically lowers the adjustment parameters, thus ensuring that the adjustment movements always remain within the user's most comfortable and imperceptible range.

[0056] Furthermore, the micro-intention recognition program unit in this embodiment is also trained to recognize a conscious posture change—that is, the user's intention to stand up. This intention can be accurately recognized through temporal features such as the user's forward shift of body weight and specific activation patterns of the leg, waist, and abdominal muscle groups.

[0057] Once the intention to stand up is recognized, the intervention decision module will execute a preset sequence of standing assistance instructions: it will temporarily suspend the deviation suppression adjustment function and instead control the intelligent actuator to smoothly and slightly raise and tilt the seat forward, providing a gentle assistance to the user's standing action, thereby significantly reducing the difficulty and risk of the user standing up.

[0058] If a user's physical condition changes and requires them to maintain a new, stable posture different from the initial baseline for an extended period, the system will enter observation and learning mode. It will intelligently identify this long-term preference and automatically and smoothly update its dynamic characteristic parameter baseline, ensuring that the system's adaptive adjustment function always keeps in line with the user's latest needs without the need for frequent manual resets.

[0059] This embodiment provides an intelligent, proactive, and humanized solution for home-based health and wellness monitoring by applying the technical solution of the present invention. It can not only prevent secondary injuries caused by poor posture, but also proactively provide safety assistance through intent recognition, demonstrating extremely high practical value.

[0060] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A sofa posture adaptive adjustment system based on multimodal sensing fusion, characterized in that, include: Baseline modeling module: Acquires steady-state reference coordinates representing the user's preset working state through a multimodal sensor array, and generates personalized dynamic feature parameter baselines based on the steady-state reference coordinates; Monitoring and Diagnostic Module: Runs the micro-intent recognition program unit, analyzes the temporal change pattern of the steady-state reference coordinate, and determines the user's real-time state changes. Real-time state changes include two types: unconscious posture deviations caused by changes in physiological state and conscious posture changes actively generated by the user to complete short-term tasks. The module continuously compares the user's real-time posture changes with the dynamic feature parameter baseline to detect whether there are any deviations. Intervention decision module: When the following two conditions are met simultaneously: First, it is detected that the user's real-time state change has deviated from the baseline of the dynamic feature parameters; Second, the type of deviation is determined to be an unintentional posture deviation; Control commands are generated and sent to the intelligent actuator to drive it to perform deviation suppression adjustment, restoring the user's state to within the state range defined by the dynamic characteristic parameter baseline.

2. The sofa posture adaptive adjustment system based on multimodal sensing fusion according to claim 1, characterized in that, The baseline modeling module is specifically as follows: The calibration procedure is initiated. The multimodal sensor array continuously acquires and records multidimensional signals generated by the user under a stable reference attitude within a preset duration. The set of multidimensional signals is defined as the steady-state reference coordinates. Statistical processing is performed on the parameters of each multidimensional signal in the steady-state reference coordinates to calculate their mean and standard deviation. The mean is combined with a tolerance range set based on the standard deviation to generate the dynamic characteristic parameter baseline, wherein the tolerance range defines the normal fluctuation range around the mean.

3. The sofa posture adaptive adjustment system based on multimodal sensing fusion according to claim 2, characterized in that: The calibration procedure is a user interaction process that issues a command to the user to enter the calibration state and guides the user to maintain the stable reference posture on the sofa. During the entire time period in which the user maintains the reference posture, the multi-dimensional signals are collected synchronously. The multidimensional signals include body pressure distribution patterns acquired by a pressure sensor matrix, electromyographic signals acquired by a surface electromyography sensor, and thermal signals acquired by a passive infrared sensor array.

4. The sofa posture adaptive adjustment system based on multimodal sensing fusion according to claim 1, characterized in that, The micro-intent recognition program unit includes: The signal preprocessing subunit is used to receive the original multi-dimensional signal with timestamps from the multi-modal sensor array, and to perform filtering, noise reduction and time synchronization processing on the multi-dimensional signal to generate time series data in a unified format. The dynamic feature extraction subunit is used to analyze the time series data within a sliding time window and extract a set of dynamic feature vectors that can quantify the rate, amplitude and direction of signal change. The pattern classification subunit is used to input the dynamic feature vector into a pre-trained classification model, and the model outputs a classification result, which determines the current real-time posture change as the unconscious posture deviation and / or conscious posture change.

5. A sofa posture adaptive adjustment system based on multimodal sensing fusion according to claim 4, characterized in that, The pattern classification subunit is specifically used for: The feature vector sequence will be composed of multiple dynamic feature vectors; The classification model is a recurrent neural network model, which is pre-trained on a dataset containing a large number of labeled samples, wherein the samples are labeled as unconscious posture deviations and / or conscious posture changes. The recurrent neural network model processes the feature vector sequence to identify a first type of temporal pattern that represents slow, continuous changes and a second type of temporal pattern that represents rapid, instantaneous changes, and classifies them as unconscious posture deviation and conscious posture change, respectively.

6. The sofa posture adaptive adjustment system based on multimodal sensing fusion according to claim 1, characterized in that: The deviation suppression adjustment is a preset, multi-stage, gradual adjustment process. The intervention decision module generates a set of control command sequences with fixed amplitude and rate parameters based on the detected degree of deviation. The intelligent actuator, according to the command sequence, slowly and continuously changes the geometry of the sofa body at a speed lower than the user's normal perception threshold until the user's real-time state returns to the range defined by the dynamic feature parameter baseline.

7. A sofa posture adaptive adjustment system based on multimodal sensing fusion according to claim 6, characterized in that, The user's normal perception threshold is a set of parameters dynamically determined through a feedback learning mechanism, specifically: After performing a deviation suppression adjustment, the multi-dimensional signals of the multi-modal sensor array are immediately monitored and analyzed to determine whether the user has generated a sudden attitude change signal. When the user generates the posture change signal, it indicates that the parameters adjusted this time exceed the user's normal perception threshold, and the system will automatically reduce the amplitude and rate parameters in subsequent adjustments. If the user does not generate the sudden change in posture signal, it indicates that the parameter adjusted this time is within the user's normal perception threshold, and the system will maintain and / or fine-tune the current parameter.

8. A sofa posture adaptive adjustment system based on multimodal sensing fusion according to claim 1, characterized in that, The intervention decision module is further configured to: when the monitoring and diagnosis module determines that the user's real-time posture change is a conscious posture change, temporarily suspend the deviation suppression and adjustment function and enter an observation and learning mode; in the observation and learning mode, if the duration of the conscious posture change exceeds a preset threshold, the system automatically collects data under that posture and updates and iterates the dynamic feature parameter baseline.

Images