A sofa posture self-adaptive adjusting system based on multi-modal sensor fusion

Abstract

The present application relates to the technical field of adaptive control system, specifically to a sofa posture adaptive adjustment system based on multi-modal sensor fusion, comprising: a baseline modeling module, a monitoring and diagnosis module and an intervention decision module. The present application establishes a personalized dynamic characteristic parameter baseline for the user through the baseline modeling module, and accurately distinguishes between unconscious posture deviation and conscious task-oriented action by the micro-intention recognition program unit in the monitoring and diagnosis module, so that the intervention decision module only starts the adaptive deviation suppression adjustment when the former is identified. The present application solves the technical problem that the existing control system cannot distinguish the cause of state change, resulting in frequent false triggering of adjustment actions, and realizes personalized and intelligent closed-loop control under the avoidance of invalid interference.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of adaptive control systems, in particular to a sofa posture adaptive adjustment system based on multi-modal sensor fusion. BACKGROUND

[0002] Program control systems have been applied to daily equipment such as furniture, but the control logic is usually fixed, lacking the ability to adapt to complex dynamic environments.

[0003] The control systems in the prior art mostly rely on preset fixed thresholds or simple timing logic to trigger adjustment actions. Such systems cannot establish personalized steady-state reference models for different controlled objects, so their adjustment targets lack pertinence. More importantly, existing technologies usually cannot distinguish whether the state change of the controlled object is caused by systematic state drift or by transient disturbance caused by the operator's short-term task instruction. This indiscriminate response logic often leads to false triggering at inappropriate times, interfering with normal operation. In addition, the parameters of its adjustment process are usually fixed and cannot be adjusted dynamically according to feedback, making it difficult for the system to achieve truly unobtrusive and refined closed-loop control.

[0004] Therefore, those skilled in the art urgently need a new program control system. The system needs to be able to autonomously learn and establish personalized dynamic reference baselines for controlled objects, and accurately distinguish between system state drift and task-oriented disturbance by analyzing the timing patterns of signals, so as to perform precise and adaptive closed-loop feedback adjustment at the right time to avoid unnecessary interference.

[0005] To this end, a sofa posture adaptive adjustment system based on multi-modal sensor fusion is proposed. SUMMARY

[0006] The purpose of the present application is to provide a sofa posture adaptive adjustment system based on multi-modal sensor fusion, which establishes personalized dynamic characteristic baselines and uses micro-intention recognition programs to distinguish between unconscious posture deviations and conscious task-oriented actions, achieving precise and gradual closed-loop feedback adjustment only for unconscious posture deviations while avoiding unnecessary interference to the user, thereby improving the adaptive ability and intelligent level of the system.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0008] A sofa posture adaptive adjustment system based on multi-modal sensor fusion, comprising:

[0009] A baseline modeling module: a steady-state reference coordinate representing a user in a preset working state is collected by a multi-modal sensor array, and a personalized dynamic characteristic parameter baseline is generated based on the steady-state reference coordinate;

[0010] A monitoring and diagnosis module: a micro-intention recognition program unit is run, a time sequence change pattern of the steady-state reference coordinate is analyzed, a real-time state change of the user is determined, the real-time state change includes two types, including an unconscious attitude deviation caused by a physiological state change and a conscious attitude change actively generated by the user to complete a short-term task, and the real-time attitude change of the user is continuously compared with the dynamic characteristic parameter baseline to detect whether there is a deviation;

[0011] An intervention decision module: when the following two conditions are met simultaneously: first, it is detected that the real-time state change of the user has deviated from the dynamic characteristic parameter baseline; second, it is determined that the type of the deviation is an unconscious attitude deviation; a control instruction is generated and sent to an intelligent actuator to drive the intelligent actuator to perform deviation suppression adjustment to restore the state of the user to within the state range defined by the dynamic characteristic parameter baseline.

[0012] Preferably, the baseline modeling module specifically comprises:

[0013] A calibration program is started, the multi-modal sensor array continuously collects and records multi-dimensional signals generated by the user in a stable reference attitude within a preset duration, and a set composed of the multi-dimensional signals is defined as the steady-state reference coordinate; parameters of each multi-dimensional signal in the steady-state reference coordinate are statistically processed, the mean and standard deviation of each are calculated, and 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 a normal fluctuation interval around the mean.

[0014] Preferably, the calibration program is a user interaction process, which issues an instruction to the user to enter a calibration state and guides the user to maintain the stable reference attitude on a sofa, and the multi-dimensional signals are synchronously collected during the entire period when the user maintains the reference attitude.

[0015] The multi-dimensional signals include a body pressure distribution pattern collected by a pressure sensor matrix, an electromyographic signal collected by a surface electromyographic sensor, and a thermal signal collected by a passive infrared sensor array.

[0016] Preferably, the micro-intention recognition program unit comprises:

[0017] A signal preprocessing subunit for receiving original multi-dimensional signals with time stamps from the multi-modal sensor array, and performing filtering, denoising and time synchronization processing on the multi-dimensional signals to generate time sequence data in a unified format;

[0018] a dynamic feature extraction subunit, configured to analyze the time series data within a sliding time window and extract a set of dynamic feature vectors quantifying the rate, amplitude and direction of signal change;

[0019] a pattern classification subunit, configured to input the dynamic feature vectors into a pre-trained classification model and output a classification result from the model, which classifies the current real-time posture change as the unconscious posture deviation and / or the conscious posture change.

[0020] Preferably, the pattern classification subunit is specifically configured to:

[0021] form a feature vector sequence from a plurality of the dynamic feature vectors;

[0022] 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 deviation and / or conscious posture change;

[0023] the recurrent neural network model identifies a first type of time series pattern representing slow and continuous change and a second type of time series pattern representing fast and instantaneous change by processing the feature vector sequence, and classifies them as unconscious posture deviation and conscious posture change, respectively.

[0024] Preferably, the deviation suppression adjustment is a pre-set, multi-stage, gradual adjustment process, and the intervention decision module generates a sequence of control instructions with fixed amplitude and rate parameters according to the detected deviation degree, and the intelligent actuator slowly and continuously changes the geometric shape of the sofa body at a speed lower than the user's normal perception threshold according to the instruction sequence until the user's real-time state returns to the range defined by the dynamic feature parameter baseline.

[0025] Preferably, the user's normal perception threshold is a set of parameters dynamically determined by a feedback learning mechanism, specifically:

[0026] 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 posture change signal;

[0027] When the user generates the posture change signal, it indicates that the parameters of this adjustment have exceeded the user's normal perception threshold, and the system will automatically adjust the amplitude and rate parameters to a lower level in subsequent adjustments;

[0028] When the user does not generate the sudden posture change signal, it indicates that the current adjustment parameter is within the normal perception threshold of the user, and the system will maintain and / or fine-tune the current parameter.

[0029] Preferably, the intervention decision module is further configured to: when the monitoring and diagnosis module determines that the real-time posture change of the user is the conscious posture change, temporarily suspend the deviation suppression adjustment function and enter an observation and learning mode; in the observation and learning mode, if the conscious posture change lasts for more than a preset threshold, the system automatically collects data in the posture and updates the dynamic characteristic parameter baseline.

[0030] Compared with the prior art, the present application has the following beneficial effects:

[0031] 1. The present application overcomes the fundamental defect of the prior art that the adjustment target is not accurate by establishing a unique dynamic characteristic parameter baseline for each user. The system no longer forces the user to adapt to a fixed, theoretical standard posture, but uses the user's own optimal state as the adjustment reference system. This highly personalized closed-loop control makes the adjustment instruction more targeted and effective, accurately adapts to the physical characteristics and usage habits of different users, and significantly improves the accuracy and practical application value of the adjustment system.

[0032] 2. The present application solves the technical problem of frequent false triggering of adjustment actions caused by the inability to distinguish the reasons for state changes by introducing a micro-intention recognition program module. The system can accurately identify unconscious posture deviations that require intervention by analyzing the timing pattern of sensor signals, and can actively ignore normal, task-oriented actions of the user. This conditional intervention with logical judgment ability changes the adjustment system from a simple mechanical executor to an intelligent agent that understands the context, greatly improving user experience and system practicality.

[0033] 3. The deviation suppression adjustment method and dynamic learning mechanism of the user perception threshold proposed in the present application are different from the fixed and easily perceived adjustment method. The adjustment process of the present application is gradual and slow, and the adjustment parameter can be self-adapted according to user feedback, always striving to remain below the normal perception threshold of the user. This adjustment method does not interrupt the user's work or rest, and optimizes the posture in a nearly unconscious state of the user, achieving the unification of efficient intervention and extreme comfort experience, significantly improving the friendliness of the system and the long-term acceptance of the user. BRIEF DESCRIPTION OF DRAWINGS

[0034] Fig. 1 A system flowchart of a sofa posture self-adaptive adjustment system based on multi-modal sensor fusion is provided for the present application.

[0035] Fig. 2 A system architecture diagram of a sofa posture adaptive adjustment system based on multi-modal sensor fusion is proposed for the present application.

[0036] Fig. 3 A flowchart of a sofa posture adaptive adjustment system based on multi-modal sensor fusion is proposed for the present application. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0038] Embodiment one

[0039] Please refer to Figs. 1-3 The present application provides a sofa posture adaptive adjustment system based on multi-modal sensor fusion, and the technical solutions are as follows:

[0040] A sofa posture adaptive adjustment system based on multi-modal sensor fusion, as shown in Figs. 1-3 , comprises:

[0041] Baseline modeling module: acquire a steady-state reference coordinate representing a user in a preset working state through a multi-modal sensor array, and generate a personalized dynamic characteristic parameter baseline based on the steady-state reference coordinate;

[0042] Monitoring and diagnosis module: run a micro-intention recognition program unit, analyze the time sequence change pattern of the steady-state reference coordinate, determine the real-time state change of the user, and the real-time state change includes two types, including unconscious posture deviation caused by physiological state change and conscious posture change actively generated by the user to complete a short-term task; and continuously compare the real-time posture change of the user with the dynamic characteristic parameter baseline to detect whether there is a deviation;

[0043] Intervention decision module: when the following two conditions are met simultaneously: first, it is detected that the real-time state change of the user has deviated from the dynamic characteristic parameter baseline; second, it is determined that the type of the deviation is unconscious posture deviation; generate and send a control instruction to an intelligent actuator to drive the intelligent actuator to perform deviation suppression adjustment, and restore the state of the user to within the state range defined by the dynamic characteristic parameter baseline.

[0044] Further, the baseline modeling module specifically comprises:

[0045] The calibration procedure is started, and the multi-modal sensor array continuously acquires and records the multi-dimensional signals generated by the user in a stable reference posture for a preset duration, and defines the set of multi-dimensional signals as the steady-state reference coordinates; the parameters of each multi-dimensional signal in the steady-state reference coordinates are statistically processed to calculate the mean and standard deviation, and 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 interval around the mean.

[0046] The baseline modeling module further includes an ergonomic optimization unit for actively performing a series of minor exploratory sofa posture adjustments after the calibration procedure generates a preliminary dynamic characteristic parameter baseline, and real-time monitoring of the feedback of the multi-dimensional signals to find an optimal reference coordinate that allows the user's core muscle group to maintain a stable state with the lowest energy consumption, and the optimal reference coordinate is taken as the final calibration result of the dynamic characteristic parameter baseline.

[0047] The present application improves baseline modeling from passive learning to active optimization. It is no longer limited to learning the user's subjective perception of comfortable posture, but objectively finds a more scientific and healthy posture for the user's core muscle group with the lowest energy consumption through exploratory adjustment and real-time physiological signal feedback. This makes the final adjustment target not only personalized, but also optimized, which fundamentally guides the user to a less tiring sitting posture and improves the health value of the system.

[0048] Further, the calibration procedure is a user interaction process that instructs the user to enter the calibration state and guides the user to maintain the stable reference posture on the sofa. The multi-dimensional signals are synchronously acquired during the entire period when the user maintains the reference posture.

[0049] The multi-dimensional signals include body pressure distribution patterns acquired by a pressure sensor matrix, electromyographic signals acquired by surface electromyographic sensors, and thermal signals acquired by a passive infrared sensor array.

[0050] The calibration procedure is started and guided through a mobile application in communication with the system. The application guides the user to adjust to a reference posture with both feet flat, back against the sofa backrest, straight and relaxed through its graphical user interface and / or voice prompts.

[0051] In the acquisition process, the multi-dimensional signal is not directly recorded as the original data stream, but the key characteristic values are extracted to form the steady reference coordinate. Specifically, for the body pressure distribution pattern, the two-dimensional coordinates of the pressure center are extracted; for the electromyographic signal, the root mean square value in a specific frequency band is extracted; and for the thermal signal, the effective trigger number per unit time is extracted.

[0052] The preset duration period can be set to 10 to 30 seconds to ensure that the acquired characteristic values can reflect the stable state of the user rather than instantaneous fluctuations.

[0053] In generating the dynamic characteristic parameter baseline, the tolerance range can be specifically set to 1.5 to 2.5 times the standard deviation of the characteristic value added or subtracted from the mean value of the characteristic value, to construct a dynamic monitoring interval that can accommodate normal micro-movements and identify abnormal deviations.

[0054] The present application standardizes the calibration process of the user through the graphical or voice guidance of the mobile application, ensures the effectiveness and consistency of the initial reference posture, and fundamentally avoids the problem of baseline misalignment caused by the user's random sitting posture, which affects the accuracy of all subsequent judgments. The scheme clearly specifies the process of refining a set of low-dimensional, stable, and information-rich key characteristic values from high-dimensional and complex original sensing signals. This feature extraction step not only significantly reduces the computational complexity of subsequent data processing, enabling the system to run more efficiently, but also enhances the robustness of the model by focusing on core information, making it less susceptible to irrelevant signal noise. The dynamic baseline established by statistical methods, which includes a clear tolerance range, overcomes the defect of being too strict when comparing with a single static target point, which is prone to false positives due to normal human micro-movements. It constructs a "healthy and comfortable domain" that can tolerate reasonable fluctuations, making the subsequent monitoring and diagnosis function more practical and accurate, and providing a solid and reliable data foundation for the intelligent and self-adaptive adjustment capability of the entire system.

[0055] Further, the micro-intention recognition program unit includes:

[0056] The signal preprocessing sub-unit is configured to receive the original multi-dimensional signals with time stamps from the multi-modal sensor array, and perform filtering, denoising, and time synchronization processing on the multi-dimensional signals to generate time series data in a unified format.

[0057] The dynamic feature extraction sub-unit is configured 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.

[0058] a pattern classification subunit configured to input the dynamic feature vector into a pre-trained classification model and output a classification result from the model, the classification result determining the current real-time posture change as the unconscious posture deviation and / or the conscious posture change.

[0059] Further, the pattern classification subunit is specifically configured to:

[0060] construct a feature vector sequence from a plurality of the dynamic feature vectors;

[0061] the classification model is a recurrent neural network model pre-trained on a dataset containing a large number of labeled samples, wherein the samples are labeled as unconscious posture deviation and / or conscious posture change;

[0062] the recurrent neural network model identifies a first type of time series pattern representing slow and continuous change and a second type of time series pattern representing fast and instantaneous change by processing the feature vector sequence, and classifies them as unconscious posture deviation and conscious posture change, respectively.

[0063] In the signal preprocessing subunit, the filtering and denoising process can specifically include: using a 20-500Hz band-pass filter for the electromyographic signal, and using a notch filter capable of eliminating power frequency interference for all signals.

[0064] In the dynamic feature extraction subunit, the length of the sliding time window can be set to 2-5 seconds with a 50% overlap rate to ensure continuity of analysis. The dynamic feature vector can specifically include: displacement velocity and acceleration of the pressure center calculated from the pressure distribution pattern, root mean square value and average power frequency calculated from the electromyographic signal, and trigger frequency per unit time calculated from the thermal signal.

[0065] In this embodiment, the classification model in the pattern classification subunit is preferably a long short-term memory network model, which is better at capturing long-term dependencies in time series. The dataset for training can be constructed by inviting multiple testers to perform script tasks containing pre-set intentional actions and long-term sitting while synchronously recording sensor data and manually labeling.

[0066] The application can remove environmental noise and artifact interference in the original signal to the greatest extent through targeted filtering preprocessing, ensure the signal-to-noise ratio and effectiveness of the input data, and provide a reliable data basis for subsequent analysis. The scheme creatively converts the abstract problem of identifying the human "intention" into an engineering problem of time series analysis on a specific and quantifiable dynamic feature vector. This feature engineering method makes the target of the classification task clearer and more accurate, and is a key step to realize high-precision recognition. Moreover, the long short-term memory network model is preferred, which is particularly good at processing long-term dependencies, so that it can effectively capture and distinguish the two patterns with significant differences in the time dimension, i.e. "slow and continuous" unconscious deviation and "fast and instantaneous" conscious change. Finally, the clear training data set construction method provides a clear and feasible path for the implementation of the pre-trained model, ensures the high accuracy and reproducibility of the classification model, and makes the intelligent judgment of the whole system no longer a fuzzy "black box", but a reliable technical implementation with clear basis and reliable performance.

[0067] Further, the monitoring and diagnosis module further comprises a multi-source information fusion arbitration unit, which is located before the mode classification subunit, and is used for giving different confidence weights to signal features from different types of sensors, and using an evidence theory algorithm to fuse and process possible multi-source conflicting and / or fuzzy signal information, so as to output a unified state judgment result with the highest confidence.

[0068] The application solves the problem of information conflict or ambiguity of different sensors in a complex scene by fusing and arbitrating multi-source sensing information, and significantly improves the accuracy, robustness and reliability of the final state judgment of the system.

[0069] Further, the deviation suppression adjustment is a preset, multi-stage and gradual adjustment process, the intervention decision module generates a control instruction sequence with fixed amplitude and rate parameters according to the detected deviation degree, and the intelligent actuator changes the geometric shape of the sofa body slowly and continuously at a speed lower than the normal perception threshold of the user according to the instruction sequence until the real-time state of the user returns to the range defined by the dynamic feature parameter baseline.

[0070] Further, the user normal perception threshold is a parameter set dynamically determined through a feedback learning mechanism, specifically:

[0071] After performing a deviation suppression adjustment, the multidimensional signals of the multi-modal sensor array are immediately monitored and analyzed to determine whether the user has generated a sudden posture change signal;

[0072] When the user generates the gesture change signal, it indicates that the current adjustment parameter exceeds the user's normal perception threshold, and the system will automatically reduce the amplitude and rate parameters in the subsequent adjustment;

[0073] When the user does not generate the sudden gesture change signal, it indicates that the current adjustment parameter is within the user's normal perception threshold, and the system will maintain and / or fine-tune the current parameter.

[0074] The multi-stage gradual adjustment process can be realized by a proportional-integral control algorithm to replace fixed parameters. The intervention decision module takes the detected deviation as the input error, and continuously generates control instructions related to the current size and historical accumulation of the error through the PI algorithm, so that the amplitude and rate of adjustment dynamically and smoothly adapt to changes in the deviation, thereby avoiding the abruptness caused by fixed or segmented parameters.

[0075] The sudden gesture change signal can be quantitatively defined as: within a short time window (e.g. 2 seconds) after performing an adjustment, the displacement speed of the pressure center coordinates monitored by the pressure sensor matrix, or the change rate of the root mean square value of the electromyographic signal monitored by the surface electromyographic sensor, exceeds a preset alert threshold.

[0076] In the feedback learning mechanism, the adjustment of the parameters can adopt an adaptive step strategy. Specifically, when it is judged that the user generates the sudden gesture change signal, the system can multiply the current amplitude and rate parameters by a decay coefficient less than 1 (e.g. 0.8); when it is judged that the user does not generate the sudden gesture change signal, the system can multiply by a gain coefficient slightly greater than 1 (e.g. 1.05) to tentatively find the parameter boundary of the optimal adjustment efficiency without causing the user to be aware.

[0077] The application replaces the rigidity and abruptness caused by fixed parameters or simple segmented adjustment by introducing a proportional-integral control algorithm, dynamically and smoothly associating the amplitude and rate of adjustment with the degree of deviation of the user's posture and its duration, so that the intensity of each adjustment exactly matches the actual demand, and the process is more natural and humanized. The "sudden posture change" signal as the feedback learning trigger condition is clearly quantified and defined, providing an objective and reliable basis for the system to determine whether the user has noticed the adjustment behavior, avoiding the performance instability problem caused by ambiguous judgment criteria in the learning process. An adaptive step strategy with decay and gain coefficients is adopted, giving the system the ability to optimize itself. It enables the system to automatically converge to the unique "inconspicuous adjustment" parameter boundary of each user like a human expert through continuous benign exploration and user non-sensory feedback, and finally realizes the ideal state of minimizing user disturbance while maximizing adjustment effect.

[0078] Further, the intervention decision module is further configured to: when the monitoring and diagnosis module determines that the real-time posture change of the user is the conscious posture change, temporarily suspend the deviation suppression adjustment function, and enter an observation learning mode; in the observation learning mode, if the conscious posture change lasts for more than a preset threshold, the system automatically collects data in this posture and updates and iterates the dynamic characteristic parameter baseline.

[0079] The preset threshold can be a duration, for example, 10 to 20 minutes, to ensure that the new posture is a stable posture that the user intends to maintain for a long time.

[0080] The update and iteration of the dynamic characteristic parameter baseline can specifically adopt a weighted average algorithm. The algorithm takes the newly collected data representing the new stable posture as a new instantaneous baseline, and performs weighted fusion with the original dynamic characteristic parameter baseline, wherein the weight of the original baseline is larger (for example, 0.9-0.95), and the weight of the new instantaneous baseline is smaller (for example, 0.05-0.1). In this way, the baseline is updated smoothly and gradually to ensure the stability of the system adjustment.

[0081] After the baseline update is completed, or if the user recovers to the posture range defined by the original baseline within the threshold time, the observation learning mode ends, and the system automatically reactivates the deviation suppression adjustment function.

[0082] The intervention decision module further comprises a bias type diagnosis and targeted adjustment strategy unit, which is used to further subdivide the deviation into specific posture problem types such as lumbar curvature collapse, pelvic retroversion and / or body asymmetric side inclination, based on specific changes in the stress distribution pattern after determining that the deviation type is unconscious posture deviation, and to call a specific intelligent actuator combination and action sequence corresponding to the specific posture problem type from a preset adjustment strategy library to achieve targeted and efficient adjustment of different posture problems.

[0083] The present application improves the adjustment from general recovery to precise targeted intervention, significantly improves the efficiency, accuracy and final ergonomic effect of a single adjustment action by diagnosing the specific root cause of the posture problem and "treating the disease".

[0084] The present application can accurately learn the user's persistent new preferred posture and smoothly and stably iterate its dynamic baseline through time threshold filtering and weighted average updating algorithm, effectively avoiding false learning of temporary postures and adjustment instability caused by baseline mutation. It greatly improves the harmony of human-computer interaction and the level of individualization of the system, so that it can truly and continuously meet the dynamic needs of the user.

[0085] The present application overcomes the defects of inaccurate adjustment target and poor adaptability caused by general models by establishing a personalized dynamic feature parameter baseline for the user, making the adjustment more targeted and effective. More importantly, the system can accurately distinguish between unconscious posture deviation that needs intervention and user's normal task-oriented action through the micro-intention diagram recognition program, and only execute adjustment when it is determined as the former. This conditional intervention logic completely avoids ineffective adjustment and operation interference caused by misjudgment of user's intention, greatly improving the intelligent level, practicality and user experience of the system.

[0086] Embodiment two

[0087] The present embodiment provides a specific application in the scene of home health care monitoring. The user can be an elderly person who needs to sit on a sofa for a long time, has difficulty in movement or has weak muscle strength.

[0088] The user's family members or caregivers start a one-time calibration program for the user through a mobile application that communicates with the system. Under the guidance of the calibration program, the user is adjusted to a reference posture that is comfortable and well supported according to his or her physical condition. The system then collects data and establishes a personalized dynamic feature parameter baseline. This dynamic feature parameter baseline is not only the basis for subsequent adjustment, but also can serve as a digital archive reflecting the user's long-term posture health.

[0089] In daily use, the monitoring and diagnosis module of the system works continuously. When the system's built-in micro-intention recognition program unit based on recurrent neural network detects that the user slowly produces an unconscious posture deviation due to fatigue or muscle weakness, the intervention decision module will start immediately; in this embodiment, the unconscious posture deviation can be body slumping or tilting.

[0090] The deviation suppression adjustment adopted by the system is a very mild intervention. The system will slowly and gradually increase the convexity of the waist support or slightly adjust the seat surface inclination at a speed that the user can hardly perceive, giving the user a subconscious support guide to help him or her recover to a healthier posture without disturbance. The amplitude and rate of the whole process will also be dynamically optimized through a feedback learning mechanism: the system will monitor the user's response to the adjustment, and if the user produces a sudden action, it will automatically lower the adjustment parameters, so as to ensure that the adjustment action is always within the user's most comfortable and imperceptible interval.

[0091] In addition, the micro-intention recognition program unit in this embodiment is also trained to recognize a conscious posture change, i.e. the user's intention to stand up. This intention can be accurately recognized through the timing features such as the user's body center of gravity moving forward and the specific activation pattern of the leg and waist and abdominal muscle groups.

[0092] When the intention to stand up is recognized, the intervention decision module will execute a preset standing-up assistance instruction sequence: it will temporarily suspend the deviation suppression adjustment function, and instead control the intelligent actuator to smoothly and slightly raise and tilt the seat surface, providing a gentle boost for the user's standing-up action, thereby significantly reducing the difficulty and risk of the user's standing up.

[0093] If the user's physical condition changes and needs to maintain a new, different stable posture from the initial baseline for a long time, the system will also enter an observation and learning mode, intelligently identify this long-term preference, and automatically and smoothly update its dynamic characteristic parameter baseline, ensuring that the adaptive adjustment function of the system always meets the user's latest needs without frequent manual reconfiguration.

[0094] This embodiment provides an intelligent, proactive and humanized solution for the family health care monitoring scenario by applying the technical solutions of the present application, which not only prevents secondary injury caused by bad posture, but also actively provides safety assistance through intention recognition, demonstrating high practical value.

[0095] Although embodiments of the present application 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 present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A sofa posture self-adaptive adjustment system based on multi-modal sensor fusion, characterized in that, The application relates to a dynamic characteristic parameter baseline modeling method and system for a user in a preset working state. The baseline modeling module comprises: starting a calibration program, continuously collecting and recording multidimensional signals generated by the user in a stable reference posture within a preset duration, and defining a set composed of the multidimensional signals as the stable reference coordinates; statistically processing parameters of each multidimensional signal in the stable reference coordinates, calculating the mean value and standard deviation of each multidimensional signal, and combining the mean value with a tolerance range set based on the standard deviation to generate the dynamic characteristic parameter baseline, wherein the tolerance range defines a normal fluctuation interval around the mean value; The monitoring and diagnosis module comprises: a micro-intention recognition program unit for analyzing the time sequence change mode of the stable reference coordinates, determining the real-time state change of the user, and continuously comparing the real-time posture change of the user with the dynamic characteristic parameter baseline to detect whether there is a deviation. The micro-intention recognition program unit comprises: a signal preprocessing subunit for receiving original multidimensional signals with time stamps from the multimodal sensor array, filtering, denoising and time synchronizing the multidimensional signals, and generating time sequence data in a unified format; a dynamic feature extraction subunit for analyzing the time sequence data within a sliding time window and extracting a group of dynamic feature vectors capable of quantifying the signal change rate, amplitude and direction; a pattern classification subunit for inputting the dynamic feature vectors into a pre-trained classification model and outputting a classification result from the model, which determines the current real-time posture change as the unconscious posture deviation and / or the conscious posture change; an intervention decision module for generating and sending a control instruction to an intelligent actuator to drive the intelligent actuator to perform deviation suppression adjustment and restore the state of the user to within the state range defined by the dynamic characteristic parameter baseline when the following two conditions are met simultaneously: first, the real-time state change of the user has deviated from the dynamic characteristic parameter baseline; and second, the type of the deviation is determined to be the unconscious posture deviation. The calibration program is a user interaction process for issuing an instruction to the user to enter a calibration state and guiding the user to maintain the stable reference posture on a sofa, and the multidimensional signals are synchronously collected during the whole period when the user maintains the reference posture.

2. The sofa posture self-adaptive adjustment system based on multi-modal sensor fusion according to claim 1, characterized in that: The multidimensional signals comprise body pressure distribution patterns collected by a pressure sensor matrix, electromyographic signals collected by surface electromyographic sensors and thermal signals collected by a passive infrared sensor array. The micro-intention recognition program unit comprises:

3. The sofa posture self-adaptive adjustment system based on multi-modal sensor fusion of claim 1, characterized in that, ​ a signal preprocessing subunit configured to receive raw multi-dimensional signals with timestamps from the multi-modal sensor array, and to filter, denoise and time-synchronize the multi-dimensional signals to generate time-series data in a unified format; a dynamic feature extraction subunit configured to analyze the time-series data within a sliding time window, and to extract a set of dynamic feature vectors quantifying the rate, amplitude and direction of signal changes; a pattern classification subunit configured to input the dynamic feature vectors into a pre-trained classification model, and to output a classification result from the model, which classifies the current real-time posture change as the unconscious posture deviation and / or the conscious posture change.

4. The sofa posture self-adaptive adjustment system based on multi-modal sensor fusion of claim 1, characterized in that, the pattern classification subunit is specifically configured to: construct a feature vector sequence from a plurality of the dynamic feature vectors; the classification model is a recurrent neural network model pre-trained on a dataset containing a large number of labeled samples, wherein the samples are labeled as unconscious posture deviation and / or conscious posture change; the recurrent neural network model identifies a first type of time-series pattern representing slow and continuous changes, and a second type of time-series pattern representing fast and instantaneous changes, and classifies them as unconscious posture deviation and conscious posture change, respectively.

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

6. The sofa posture self-adaptive adjustment system based on multi-modal sensor fusion according to claim 5, characterized in that, the user's normal perception threshold is a set of parameters dynamically determined through a feedback learning mechanism, and specifically: immediately after performing a deviation suppression adjustment, the multi-modal sensor array is monitored and analyzed to determine whether the user has generated a sudden posture change signal; when the user generates the posture change signal, it indicates that the parameters of this adjustment have exceeded the user's normal perception threshold, and the system will automatically reduce the amplitude and rate parameters in subsequent adjustments; when the user does not generate the sudden posture change signal, it indicates that the parameters of this adjustment are within the user's normal perception threshold, and the system will maintain and / or fine-tune the current parameters.

7. The sofa posture self-adaptive adjustment system based on multi-modal sensor fusion of claim 1, wherein, the intervention decision module is further configured to temporarily suspend the deviation suppression adjustment function and enter an observation and learning mode when the monitoring and diagnosis module determines that the user's real-time posture change is the conscious posture change; in the observation and learning mode, if the conscious posture change lasts longer than a pre-set threshold, the system automatically collects data under this posture and updates and iterates the dynamic feature parameter baseline.

Images