Community public chair system with health monitoring and emotion healing functions and suitable for old people

Abstract

The invention relates to the technical field of public facilities, particularly discloses a community public chair system with health monitoring and emotion healing functions and aims to solve the problems that existing community public chairs are single in function and lack of systematic recognition and active intervention on physiological and psychological states of elderly users. The system comprises a multi-mode sensing array integrated on a seat, an environment regulation and control and healing execution unit and a cloud collaborative analysis and decision platform. The sensing array collects physiological and behavior signals of a user; the cloud platform fuses and analyzes the data and generates a personalized chemotherapy healing strategy; the execution unit executes intervention such as environment adjustment and sensory stimulation according to the strategy. According to the invention, real-time non-inductive monitoring and personalized active healing of the state of the elderly user are realized, and the health service capability of community communal facilities is improved.

Description

Technical Field

[0001] This invention belongs to the field of public facilities technology, specifically relating to a community elderly-friendly public seating system with health monitoring and emotional healing functions. Background Technology

[0002] Against the backdrop of a deepening global population aging, improving the health, well-being, and quality of life of the elderly has become a crucial issue in social public management and services. Among these, community public spaces, as the core carriers of daily activities and social interaction for the elderly, play a fundamental supporting role in promoting their physical and mental health and strengthening social connections. Currently, the design and upgrading of community public facilities are gradually shifting from meeting basic functions to focusing on user experience and comprehensive benefits.

[0003] Community public seating serves as a frequently used rest area and potential social space for the elderly, and its design optimization is a key aspect of building an age-friendly community environment. Existing design practices mainly focus on basic attributes such as the physical safety and ergonomic comfort of the seating, aiming to provide stable and reliable outdoor rest support for the elderly.

[0004] Current technologies for designing community public seating largely rely on experiential judgment and limited functionality, lacking a systematic identification and scientific quantification of the multi-layered and diverse needs of elderly users, including their physiological, psychological, and social needs. This makes it difficult for traditional seating designs to effectively address the deeper healing needs of the elderly in areas such as health monitoring, emotion regulation, and social interaction. Although research has confirmed the positive effects of the natural environment and outdoor activities on the mental health of the elderly, the practice of systematically integrating these healing mechanisms into the micro-facilities of public seating through intelligent and low-psychological-resistance methods remains lacking. Therefore, how to transcend traditional design paradigms and construct a smart community public seating system that can naturally integrate into the daily lives of the elderly and combines health monitoring with proactive emotional healing has become an urgent technical challenge to be solved in improving the quality of community elderly care services and achieving preventive health interventions. Summary of the Invention

[0005] The purpose of this invention is to provide a community-based public seating system for the elderly with health monitoring and emotional healing functions, in order to solve the technical contradiction that existing community public seating has limited functions and lacks the ability to systematically identify, quantify, and proactively intervene in the physiological and psychological states of elderly users.

[0006] The technical solution of this invention is a community elderly-friendly public seating system with health monitoring and emotional healing functions. The system consists of a multimodal sensing array integrated into the public seating body, an environmental control and healing execution unit deployed inside and around the seating, and a cloud-based collaborative analysis and decision-making platform. The three components interact and coordinate commands through a local edge computing gateway and a wireless communication network, forming a closed-loop service system from physiological and psychological signal acquisition and multi-dimensional state assessment to personalized proactive intervention.

[0007] The public seating features an ergonomically designed curved backrest and widened seat, with flexible pressure distribution sensing pads embedded within the seat and backrest. These pads consist of a matrix of miniature piezoresistive sensing units, each sampling the pressure value of the contact surface in real time at a sampling frequency of 10 times per second. The inner surface of the armrests integrates non-contact photoplethysmography (PPG) sensors and skin conductivity contact electrodes. The PPG sensors utilize a specific wavelength of light-emitting diodes and a photodetector arranged opposite each other to extract pulse wave signals by detecting periodic changes in capillary blood volume in the fingertips or palms. The skin conductivity contact electrodes are made of inert metal and measure changes in the conductivity of the palm skin surface in a constant-pressure mode. A miniature wide-angle camera facing the user is discreetly installed on the upper edge of the seat back. This camera captures the user's facial video stream at a rate of 1 frame per second and has a built-in privacy filtering chip. After completing the face region detection and feature extraction locally, it immediately discards the original video data and only uploads the geometric feature vectors of the extracted eye and mouth regions.

[0008] The data from the multimodal sensing array is initially fused by an edge computing gateway integrated within the seat armrest. The edge computing gateway incorporates a signal preprocessing module, a feature extraction engine, and a local lightweight state assessment model. The signal preprocessing module performs noise reduction and normalization on the data from the pressure distribution sensing pad, generating a user posture pressure center trajectory map and a static pressure distribution heatmap. The feature extraction engine performs time-domain and frequency-domain analysis on the photoplethysmography (PPG) signal, calculating the heart rate and the ratio of low-frequency power to high-frequency power in heart rate variability. Simultaneously, the feature extraction engine performs event-related potential (ERP) analysis on the skin conductance signal, extracting the baseline value of skin conductance level and the event-related fluctuation amplitude. For facial feature vectors acquired from the camera, the feature extraction engine calculates the rate of change in eye opening and closing and the sequence of changes in the corner of the mouth's upward angle between consecutive frames. The edge computing gateway then uploads the processed multimodal feature data, including the pressure center coordinate sequence, heart rate variability ratio, skin conductance response amplitude, eye movement, and corner of the mouth movement, to a cloud-based collaborative analysis and decision-making platform in real time via a 4G or 5G wireless communication module.

[0009] The cloud-based collaborative analysis and decision-making platform consists of a user state fusion analysis module, a healing strategy generation engine, and a system resource scheduler. The user state fusion analysis module receives feature data streams from an edge computing gateway and accesses real-time temperature, humidity, and light intensity data from a community weather station. This module runs a multi-task learning model based on a deep neural network. This model takes time-series feature data as input and outputs three evaluation indices in parallel: a physiological fatigue index, an emotional valence index, and a social interaction willingness index. The physiological fatigue index quantifies decreased heart rate variability, increased center of stress oscillation, and duration of maintaining a specific sitting posture. The emotional valence index is jointly inferred based on mouth corner movement, eye movement, and skin conductance response patterns; its value ranges from -1 to +1, with negative values ​​representing negative emotional tendencies and positive values ​​representing positive emotional tendencies. The social interaction willingness index is estimated by analyzing the user's head orientation deviation angle relative to the community path and the duration of solitude while seated.

[0010] The healing strategy generation engine, based on three evaluation indices output in real time by the user status fusion analysis module, combined with current time, weather data, and the user's historical preference profile, matches and generates personalized composite healing instruction sequences from a pre-set healing strategy library. The healing strategy library includes 12 basic healing strategies across three categories: environmental regulation, sensory stimulation, and social facilitation. Environmental regulation strategies include localized microclimate adjustment of the seat and background soundscape generation. Sensory stimulation strategies include directional aroma release, seat rhythm guidance, and soothing lighting. Social facilitation strategies include age-friendly interactive content push and nearby seat-linked prompts.

[0011] The environmental regulation and healing execution unit specifically includes multiple execution components integrated within the seat. A semiconductor cooling and heating element and a micro-fan are installed beneath the seat surface, forming a local temperature control system for implementing microclimate regulation within environmental regulation strategies. A full-band micro-speaker array and several sets of multi-color LED beads are embedded in the top of the seat back, used for background soundscape generation and soothing light illumination strategies, respectively. Micro-ultrasonic atomizers are located at the front of the armrests on both sides of the seat, connected to replaceable plant essential oil capsules for implementing directional aroma release strategies. A low-power linear motor is installed at the bottom of the seat, generating gentle vertical rhythms within a preset safe amplitude and frequency range for implementing seat rhythm guidance strategies. A low-power e-ink display screen is mounted on the side of the armrests for pushing text and image information. In addition, each seat is equipped with a near-field sensing module based on Bluetooth Low Energy beacons to detect the occupancy status of neighboring seats.

[0012] The system resource scheduler is responsible for coordinating the execution sequence and resource allocation of healing strategies. When the healing strategy generation engine outputs a composite healing instruction, the system resource scheduler first checks the current status and energy consumption of each execution component. Then, based on preset priority rules and coordination logic, it decomposes the instruction into specific control commands and sends them to the corresponding execution components. Priority rules stipulate that strategies involving user physiological safety and emergency emotional soothing have the highest execution priority. Coordination logic requires, for example, that when performing soothing light and color irradiation, a background soundscape with a matching frequency be simultaneously activated to enhance sensory synergy.

[0013] Furthermore, the multi-task learning model in the user state fusion analysis module is updated using a periodic online incremental learning mechanism. This mechanism periodically extracts anonymized user data after desensitization and the final labeled healing effect feedback data from the cloud database as new training samples to fine-tune the model parameters, thereby enabling the model to adapt to the common characteristics and seasonal variation patterns of elderly groups in different communities.

[0014] Furthermore, the nearby seat linkage prompt function in the aforementioned social facilitation strategy operates as follows: When the system assesses that a user's social interaction willingness index remains above a threshold of 0.5 for more than 5 minutes, and the near-field perception module detects another user within 3 meters of their side seat, the system resource scheduler will simultaneously push a lightweight interactive prompt, such as a friendly greeting or a simple suggestion for a common topic, to the e-ink displays of both seats. This process does not involve the exchange or display of any personally identifiable information.

[0015] Furthermore, the healing strategy generation engine incorporates a long-term benefit optimization mechanism based on reinforcement learning when matching strategies. This mechanism treats each healing intervention as a decision action, using the improvement in the user's physiological fatigue index and emotional valence index over a subsequent period as a reward signal. Through a policy gradient algorithm, it continuously optimizes the mapping relationship from state to strategy, aiming to find the healing strategy combination sequence that can produce the most lasting positive impact.

[0016] Furthermore, the system includes a visual data dashboard for community managers. Presented via a web interface, this dashboard displays, in heatmap form, the real-time and historical usage density of all public seating within the community, the trend of changes in the regional average emotional valence index, and statistics on the frequency of use of various therapeutic strategies. This provides data support for the optimized layout of community public spaces and the precise allocation of elderly care service resources.

[0017] Compared with the prior art, the advantages and positive effects of the present invention are as follows: 1. This invention deeply integrates multimodal physiological and psychological perception technology, edge computing, and cloud-based intelligent analysis into public seating, a common facility, to construct a non-invasive, low-psychological-resistance continuous health monitoring interface. The system can non-invasively collect multi-dimensional signals such as heart rate variability, skin conductance, facial micro-expressions, and behavioral postures, and uses deep learning models for fusion analysis. This enables real-time quantitative assessment of elderly users' physiological fatigue, emotional state, and social willingness, fundamentally addressing the technological gap of traditional seating's inability to recognize users' deep-seated states, and providing a precise data foundation for preventative health interventions.

[0018] 2. This invention proposes and implements a closed-loop healing system from state recognition to proactive intervention. The system does not passively provide rest; instead, based on real-time assessment results, it dynamically generates and executes personalized, multi-faceted healing strategies that integrate environmental regulation, sensory stimulation, and social facilitation. This proactive intervention mechanism can provide immediate feedback to the elderly, such as local microclimate regulation and soothing multi-sensory stimulation, effectively transforming public seating from static resting points into dynamic health-promoting nodes with emotion regulation and social catalysis functions, significantly enhancing the health value and proactive service of community public facilities.

[0019] 3. This invention ensures the synergy, safety, and long-term effectiveness of therapeutic interventions through a system resource scheduler and reinforcement learning optimization mechanism. The system can intelligently coordinate the orderly operation of multiple execution components, avoiding strategy conflicts and prioritizing user safety. Simultaneously, the reinforcement learning mechanism enables the system to autonomously learn optimal intervention strategies from long-term interactions, allowing the therapeutic effect to continuously optimize over time. This results in an intelligent service capability that adapts to different users and environments, realizing a paradigm shift in public facility services from standardization to personalization.

[0020] This invention achieves advanced functionality while strictly adhering to privacy protection and ethical design principles. By processing sensitive video data locally using edge computing, employing near-field anonymized interactive prompts, and ensuring all data analysis is based on de-identified information, a balance is struck between providing precise services and protecting user privacy. Furthermore, the data dashboard for administrators provides macro-level decision-making support for the scientific planning and efficient management of community elderly care resources, enabling effective linkage between micro-level individual services and macro-level community governance. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall technical solution architecture of the community elderly-friendly public seating system with health monitoring and emotional healing functions proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the multi-dimensional state evaluation model based on multimodal perception and deep neural networks in this invention; Figure 3 This is a logical flowchart of the process from state assessment to the generation of personalized healing strategies in this invention. Figure 4 This is a schematic diagram of the system interaction and data flow of multiple healing units, such as environmental regulation, sensory stimulation and social promotion, working together in this invention; Detailed Implementation Example 1

[0022] This invention provides a community-based public seating system for the elderly with health monitoring and emotional therapy functions. Please refer to the appendix. Figure 1 The system's overall architecture consists of three core layers: a multimodal sensing array and environmental control and therapeutic execution unit integrated into the public seating itself; an edge computing gateway serving as a local computing node; and a cloud-based collaborative analysis and decision-making platform located in a remote data center. These three layers continuously interact and coordinate commands through a wireless communication network, forming a complete closed-loop service system encompassing physiological and psychological signal acquisition, multi-dimensional quantitative assessment of states, and personalized proactive intervention. The system aims to transform community public seating from passive rest facilities into intelligent service nodes capable of proactively sensing, understanding, and responding to the health and emotional needs of elderly users.

[0023] The public seat serves as the physical carrier for all sensing and execution components of the system, and its structural design strictly adheres to ergonomic principles for the elderly. The seat is made of a composite of high-density, moisture-proof, and UV-resistant engineering plastics and anti-corrosion wood. The backrest is designed to conform to the natural curve of the human spine, with the radius of curvature optimized based on extensive data analysis of elderly individuals' vital signs to ensure even support for the lower back. The seat width is 20% wider than standard public seats, reaching 60 centimeters, providing more stable sitting support and space for the elderly to stand up. The seat and backrest upholstery are made of breathable, wear-resistant outdoor-grade fabric, beneath which is embedded a flexible pressure distribution sensor pad, one of the core sensing components.

[0024] The flexible pressure distribution sensing pad is composite-pressed from multi-layer flexible circuit boards and an elastic silicone substrate. The core of the pad consists of a matrix of miniature piezoresistive sensing units, each measuring 5 mm by 5 mm, with a center-to-center spacing of 10 mm. The entire seat area contains 600 sensing units (30 x 20), while the backrest area contains 375 units (25 x 15). Each miniature piezoresistive sensing unit samples the pressure exerted on the skin at its location in real time at a fixed sampling frequency of 10 times per second, with a pressure measurement range of 0 to 200 kPa and a resolution of 0.1 kPa. All sensing units are addressed and read data via a row and column scanning circuit. The scanning circuit converts the original analog resistance signal into a digital signal, which is then transmitted to the edge computing gateway integrated within the seat armrest via a dedicated internal serial peripheral interface bus.

[0025] The armrests of the seat are constructed of solid wood covered with soft polyurethane foam. Inside the armrests, in the area where the user's hands naturally rest, are integrated two sets of key physiological signal sensors. The first set is a non-contact photoplethysmography (PPG) sensor. This sensor module is encapsulated in a slightly recessed area inside the armrest, covered by a highly transparent acrylic cover. Inside the module are a pair of light-emitting diodes (LEDs) of a specific wavelength and a high-sensitivity photodetector arranged opposite each other. The LEDs emit green light with a wavelength of 530 nanometers, which is sensitive to the absorption characteristics of hemoglobin in blood. When the user places their fingers or palms naturally in the sensing area, the light emitted by the LEDs penetrates the skin tissue, is partially absorbed by the blood in the capillaries, and the reflected light is received by the photodetector. With the periodic beating of the heart, the blood volume in the capillaries undergoes minute periodic changes, causing a corresponding modulation of the reflected light intensity. The photodetector converts the light intensity signal into a weak current signal, which, after passing through a built-in preamplifier and analog-to-digital converter, generates the original pulse wave voltage sequence signal.

[0026] The second group consists of skin conductivity contact electrodes. These electrodes are made of a pair of circular contact pads, 15 mm in diameter, made of an inert metal material. The surfaces are specially treated to ensure good biocompatibility and conductivity. The two electrode pads are embedded 20 mm apart into the inner surface of the armrest, with the skin-contacting portion slightly raised. The measurement circuit uses a constant voltage mode, applying a constant 0.5 volt DC voltage between the two electrodes. When the user's palm skin experiences changes in sweat gland secretion due to emotions, cognitive load, or changes in autonomic nervous activity, the conductivity of the skin surface changes, causing a slight change in the loop current. The measurement circuit monitors this current change in real time and converts it into a skin conductivity level signal, sampling at a frequency of 32 times per second.

[0027] A miniature wide-angle camera, facing the user, is discreetly installed on the upper edge of the seat back. This camera features an ultra-small package, only eight millimeters in diameter, and its casing blends seamlessly with the seat material. The camera's optical axis is slightly tilted downwards to ensure it captures the user's frontal facial area. Its image sensor is two megapixels, capturing a grayscale video stream at one frame per second. To fully protect user privacy, a dedicated privacy filtering chip is integrated within the camera module. This chip processes the image data immediately after it leaves the sensor. Its workflow is as follows: First, a lightweight face detection algorithm is run to locate the face bounding box in the image; then, the algorithm extracts only image patches of the eye and mouth regions within the bounding boxes; next, a pre-trained feature extraction network converts these two image patches into fixed-length geometric feature vectors, such as numerical sequences describing the degree of eye opening, the position of the inner and outer corners of the eyes, and the position and curvature of the corners of the mouth; finally, the original video stream data is immediately and permanently discarded from the chip's memory, and only the two extracted feature vectors are sent to the edge computing gateway via the internal integrated circuit bus. The entire processing is completed inside the camera module, ensuring that no raw facial image leaves the device.

[0028] The edge computing gateway, serving as the system's local intelligent hub, is integrated into the internal cavity of the armrest on one side of the seat. The gateway employs an industrial-grade embedded system design, with a multi-core processor featuring an independent digital signal processing core and neural network acceleration unit. It incorporates a signal preprocessing module, a feature extraction engine, and a local lightweight state assessment model. Its primary function is to perform preliminary fusion and feature extraction of raw data from the multimodal sensing array.

[0029] The signal preprocessing module is specifically designed to handle the massive amounts of data from the flexible pressure distribution sensing pad. This module receives 9,750 pressure data points per second. The first step in preprocessing is noise reduction, employing an algorithm combining a moving average filter and a median filter to filter out high-frequency noise caused by minor user movements or environmental vibrations. The second step is normalization, dividing the pressure value of each sensing unit by its historically statistically obtained maximum possible pressure value, converting absolute pressure into a relative pressure value between 0 and 1. After preprocessing, the module generates two key types of pressure distribution derived data. The first is a user posture pressure center trajectory map. The algorithm calculates the pressure-weighted center coordinates of all activated sensing units in real time, outputting ten center point coordinate sequences per second. The second is a static pressure distribution heatmap. The algorithm integrates the normalized pressure matrix every five seconds to generate a two-dimensional matrix reflecting the pressure distribution across the seat and backrest areas.

[0030] The feature extraction engine processes signals from the photoplethysmography (PPG) sensor, skin conductance contact electrodes, and camera in parallel. For the PPG signal, the engine first performs bandpass filtering, retaining frequency components from 0.5 Hz to 5 Hz to remove baseline drift and high-frequency noise. Subsequently, the engine detects each PPG peak in the time domain and calculates the interval between consecutive peaks to obtain the instantaneous heart rate sequence. In frequency domain analysis, the engine performs a Fast Fourier Transform on the five-minute instantaneous heart rate sequence to calculate the power spectral density of heart rate variability. The feature extraction engine extracts the ratio of low-frequency power to high-frequency power as the core feature, reflecting the balance between sympathetic and parasympathetic activity in the autonomic nervous system.

[0031] For skin conductance signals, the feature extraction engine first performs low-pass filtering to smooth out physiological tremor noise. Then, the algorithm identifies the baseline value of the skin conductance level, which is the average signal value over a relatively calm period. The engine continuously monitors signal fluctuations relative to the baseline; when a fluctuation amplitude exceeds a preset threshold and exhibits a specific rise time pattern, it is marked as a skin conductance response event, and the peak amplitude, rise time, and recovery time of this event are extracted as features.

[0032] For the facial geometric feature vectors transmitted from the camera, the feature extraction engine performs temporal analysis. For the eye feature vectors, the engine calculates the rate of change in eye opening and closing values ​​between consecutive frames, forming an eye movement sequence. For the mouth feature vectors, the engine calculates the change in the upward angle of the corners of the mouth, forming a mouth corner movement sequence. These sequences reflect the activity level of facial micro-expressions.

[0033] The local lightweight state assessment model running within the edge computing gateway is a pruned and quantized neural network model used for preliminary, low-latency anomaly detection. This model takes multimodal features from the most recent 30-second window as input, including the standard deviation of the stress center coordinates, the low-frequency to high-frequency power ratio of heart rate variability, the number of skin conductance response events, average eye movement, and average corner of the mouth movement. The model outputs a binary warning flag indicating whether the user may be experiencing significant physical discomfort or emotional distress. Regardless of whether the warning flag is triggered, the edge computing gateway encapsulates the processed multimodal feature data into a standard data packet. The data packet includes a timestamp, a unique device identifier, and all the aforementioned feature data. The gateway, through its integrated 4G or 5G wireless communication module, uploads the data packet to the cloud-based collaborative analysis and decision-making platform in real time, every ten seconds, via Transmission Control Protocol (TCP) and Internet Protocol (IP). In cases of poor wireless signal, the gateway has local data caching capabilities, capable of caching data for up to 24 hours, resuming transmission once the network is restored.

[0034] The cloud-based collaborative analysis and decision-making platform is deployed on a cloud server cluster with high availability and elastic computing capabilities. The platform consists of three core subsystems: a user state fusion analysis module, a healing strategy generation engine, and a system resource scheduler. The user state fusion analysis module is the brain of the platform, responsible for in-depth mining and comprehensive analysis of massive amounts of incoming data.

[0035] This module first receives a feature data stream from the edge computing gateway. Simultaneously, it accesses real-time weather station data deployed within the community via an application programming interface (API), including ambient temperature, relative humidity, light intensity, and wind speed. This environmental context data is crucial for accurately interpreting users' physiological and psychological signals. The core of the user state fusion analysis module is a multi-task learning model based on a deep neural network. Please refer to the appendix. Figure 2 The model employs an encoder-decoder architecture. The encoder consists of three layers of long short-term memory networks, responsible for encoding the input temporal feature sequence and capturing its temporal dependencies. The encoded feature vectors are fed into three parallel decoder branches, each consisting of a fully connected neural network, responsible for outputting an evaluation index in one dimension.

[0036] The first decoder branch outputs a physiological fatigue index. This index is a continuous value between 0 and 1, with higher values ​​indicating greater fatigue. The model's training objective is to make this index comprehensively reflect a decrease in heart rate variability, an increase in the degree of pressure center trajectory sway, and the duration for which the user maintains the same sitting posture. Specifically, a prolonged low ratio of low-frequency to high-frequency heart rate variability, a consistently high standard deviation of the pressure center coordinates, and a sitting posture with no significant change in pressure distribution for more than twenty minutes will all lead to an increase in the physiological fatigue index.

[0037] The second decoder branch outputs an emotional valence index. This index is normalized to a range of -1 to +1. Negative values ​​represent negative emotional tendencies, such as frustration or anxiety; positive values ​​represent positive emotional tendencies, such as calm or pleasure; and values ​​near 0 represent neutral emotions. The model's inferences are based on the combined characteristics of mouth corner movement, eye movement, and skin conductance response patterns. For example, a sustained increase in the angle of the mouth corner, moderate eye movement, and fewer skin conductance response events would be inferred as a high positive valence index. Conversely, low mouth corner movement, low eye movement, but accompanied by occasional strong skin conductance responses might be inferred as a negative valence index.

[0038] The third decoder branch outputs a social interaction willingness index. This index, too, is a value between 0 and 1. Its estimation relies on behavioral posture analysis. The model uses pressure distribution data to infer the user's torso orientation and, combined with the seat's preset installation orientation data, calculates the angle between the user's head orientation and the direction of the main pedestrian paths in the community. The smaller the angle, the more the user is facing the public space. Simultaneously, the model records the user's time spent alone in the current seat. Combining orientation and time spent alone, the model assesses the user's potential openness to social interaction. For example, a user facing the path and spending more than ten minutes alone may correspond to a higher social interaction willingness index.

[0039] This multi-task learning model is updated using a periodic online incremental learning mechanism. Every weekend, the system extracts all anonymized user feature data from the past week, along with effect feedback data indirectly labeled from subsequent therapeutic intervention records, from the cloud database to form a new training sample set. The system uses the backpropagation algorithm to train on these new samples with a small learning rate, fine-tuning the model parameters. This mechanism enables the model to adapt to the common behavioral patterns of elderly populations in different communities and to dynamically adjust in response to long-term changes in environmental factors such as seasons and climate.

[0040] The healing strategy generation engine is the platform's decision center. It receives three evaluation indices from the user state fusion analysis module in real time, while simultaneously acquiring the current time, day of the week, and real-time weather data, and querying the user's anonymous historical preference profile. This profile records the user's historical response to various healing strategies, such as under which emotional valence index the aromatherapy release resulted in the most significant increase. Please refer to the appendix. Figure 3 At the core of the engine is a strategy library containing 12 basic healing strategies across three main categories: environmental regulation, sensory stimulation, and social facilitation.

[0041] Environmental regulation strategies include two types. Strategy one is localized microclimate regulation within the seat, which creates a comfortable microclimate around the user by controlling the temperature control system within the seat. Strategy two is background soundscape generation, which plays soothing and natural sounds or instrumental music that blend with the current environment.

[0042] Sensory stimulation strategies include three types. Strategy three is directional aroma release, which releases specially formulated plant essential oil molecules through an atomizer. Strategy four is seat rhythm guidance, which creates extremely gentle, rhythmic vertical vibrations in the seat. Strategy five is soothing lighting, which projects a soft, slowly changing halo of light from the top of the backrest.

[0043] Social facilitation strategies include two types. Strategy Six: Pushing age-friendly interactive content, displaying news summaries, health tips, or simple interactive games on the armrest's display screen. Strategy Seven: Proximity seat-linked prompts, encouraging anonymous, lightweight interactions among nearby users under specific conditions.

[0044] The healing strategy generation engine's workflow is a multi-condition matching and optimization process. The engine first determines a "profile" of the current user's state based on a combination of three evaluation indices. For example, a high physical fatigue index, a low emotional valence index, and a low willingness to social interaction index constitute a "fatigued and depressed" profile. For each "profile," the strategy library predefines several candidate strategy combinations and their initial weights. The engine then adjusts these weights based on the current time, weather, and the user's historical preference profile. For example, on a hot afternoon, the weight of the microclimate regulation strategy will significantly increase for the "fatigued and depressed" profile; and historical data showing a positive response to sandalwood essential oil will increase the weight of the sandalwood formula in the targeted aroma release strategy.

[0045] Furthermore, the engine introduces a long-term benefit optimization mechanism based on reinforcement learning. The system treats each healing intervention as a decision action, using the degree of improvement in the user's combined physiological fatigue index and emotional valence index within the next 30 minutes after receiving the intervention as the reward signal. The system maintains a policy value network, which takes user state characteristics as input and outputs the long-term expected value of each healing strategy. Through a policy gradient algorithm, the system continuously updates this value network. The core update rule of the policy gradient algorithm can be stated as follows: the direction of policy parameter updates is proportional to the long-term expected benefit of the action. This means that policy combinations that have historically brought more lasting positive effects will gradually increase their probability of being selected in similar future states. This mechanism enables the system to go beyond simple rule matching and autonomously learn the sequence of healing strategy combinations that can produce the most lasting positive effects.

[0046] Ultimately, the healing strategy generation engine outputs a personalized sequence of complex healing instructions. This sequence includes not only which strategies to execute, but also the parameters of the strategies, such as target temperature, aroma type, light color frequency, vibration amplitude, and the suggested execution duration and intensity level.

[0047] The system resource scheduler is the command center for the platform's interaction with the physical world. Please refer to the appendix. Figure 4 Once the healing strategy generation engine outputs a composite healing command, the command is first sent to the system resource scheduler. The scheduler's primary responsibility is to check the current status and system power consumption of all execution components involved in the command. It uses the heartbeat packets and status data periodically reported by the execution components to understand whether each component is online, idle, has a fault, or has reached the end of its service life.

[0048] Subsequently, the scheduler decomposes and schedules the instructions according to a set of preset, non-overridable priority rules and coordination logic. The priority rules stipulate that any strategy involving the user's physiological safety has the highest execution priority. For example, if a user is detected to have signs of fainting, the system will immediately terminate all non-emergency treatment strategies and initiate the highest-priority alarm and community management notification process. Secondly, emergency reassurance strategies for extreme negative emotions have high priority.

[0049] Coordination logic defines the timing and parameter coupling relationships when different strategies are executed in combination. For example, when deciding to implement soothing lighting, the coordination logic requires the system resource scheduler to synchronously query the frequency parameters of the current lighting color and match a soothing background sound from the soundscape library that resonates with or complements the current light color. It then instructs the background soundscape generation unit to start simultaneously, generating an enhanced sensory synergy effect and avoiding the abruptness that might result from a single sensory stimulus. As another example, when implementing seat kinetic guidance, the coordination logic temporarily reduces the speed of the fan in the local microclimate regulation system to prevent airflow from interfering with the user's perception of vibration.

[0050] After completing security and coordination verification, the system resource scheduler decomposes the high-level healing instruction sequence into a series of specific, executable low-level control commands. These commands are sent to the edge computing gateway of the target public seat via message queues, the Internet, and wireless networks, and then forwarded by the gateway to the corresponding environmental control and healing execution unit via the internal bus.

[0051] The environmental regulation and healing execution units are the terminals at which the system generates physical healing effects. Specifically, they include multiple precision execution components integrated within the seat. Within the structural frame beneath the seat surface, semiconductor cooling and heating elements and a miniature centrifugal fan are installed, together forming a localized temperature control system. The semiconductor cooling and heating elements are in close contact with a heat-conducting plate at the bottom of the seat surface, and the miniature fan accelerates air circulation within the cavity beneath the seat surface. This system receives temperature setting commands from the cloud and can generate a temperature difference of ±5 degrees Celsius relative to the ambient temperature on the seat surface, executing microclimate regulation within the environmental regulation strategy.

[0052] Embedded behind the top lining of the seat back is a full-range miniature speaker array. This array consists of four 20mm diameter neodymium magnet speaker units arranged in a specific phase to create a spatial sound field behind the user's head for background music playback. Next to the speaker array are several groups of multi-color LEDs. Each group contains four chips: red, green, blue, and warm white. Through pulse-width modulation (PWM) technology, they mix colors and adjust brightness, producing a continuous spectral change from cool blue to warm yellow for implementing a soothing lighting strategy.

[0053] Inside the front of each armrest on either side of the seat is a miniature ultrasonic atomizer. The atomizing plate of the atomizer operates at a frequency of 1.7 MHz, capable of breaking down liquid water molecules into fine particles approximately five micrometers in diameter. The atomizer connects to a pluggable, replaceable capsule containing essential oils. The capsule is pre-filled with various blends of essential oils and water, such as lavender, sweet orange, and sandalwood. Upon receiving a directional aroma release command, the designated atomizer activates, atomizing the essential oil molecules and slowly releasing them from the fine grille at the front of the armrest, typically for one to three minutes.

[0054] At the center of the seat's bottom support structure is a low-power, low-noise linear motor. Coupled to the main seat structure via an elastic rubber pad, this motor generates simple harmonic vibrations in the vertical direction with an amplitude not exceeding three millimeters and a frequency between 0.5 Hz and 2 Hz. This extremely gentle vibration, simulating a slight rocking motion or breathing rhythm, is used to implement a seat rhythm guidance strategy designed to promote relaxation through proprioceptive stimulation.

[0055] A six-inch low-power e-ink display is embedded in the side of the armrest of the seat. This display consumes power only when updating content, and consumes zero power when displaying static images. It is used for pushing text and image information, and the displayed font is specially enlarged and optimized for high contrast, making it suitable for elderly readers. The display connects to an edge computing gateway via an internal bus to receive and display interactive content delivered from the cloud.

[0056] In addition, each seat is equipped with a near-field sensing module based on Bluetooth Low Energy beacons at its base. This module periodically broadcasts a signal containing the seat's unique anonymous identifier while simultaneously scanning and receiving similar signals broadcast by other nearby seats. By measuring the received signal strength indicator, the edge computing gateway can estimate the approximate distance to nearby seats and determine whether any other seats within a three-meter radius are occupied. This information is a key input for implementing the nearby seat linkage prompt function in social facilitation strategies.

[0057] The specific execution process of the nearby seat linkage prompt function reflects the anonymous social interaction of the system.

Claims

1. A community-based elderly-friendly public seating system with health monitoring and emotional therapy functions, characterized in that, include: A multimodal sensing array integrated into the public seat body is used to collect users' physiological and behavioral signals; An environmental control and healing execution unit deployed inside and around the seat is used to perform physical therapy interventions; A cloud-based collaborative analysis and decision-making platform is used for status assessment and strategy decision-making. A local edge computing gateway, integrated into the public seat body, is used to perform preliminary fusion processing on the raw data collected by the multimodal sensing array, and upload the processed feature data to the cloud-based collaborative analysis and decision-making platform via a wireless communication network. The multimodal sensing array includes a flexible pressure distribution sensing pad, a non-contact photoplethysmography (PPG) sensor, a skin conductivity contact electrode, and a miniature wide-angle camera. The flexible pressure distribution sensing pad is embedded in the seat surface and backrest of the seat and consists of miniature piezoresistive sensing units arranged in a matrix, used to collect the pressure value of the contact surface in real time at a preset sampling frequency. The non-contact PPG sensor and the skin conductivity contact electrode are integrated into the inner surface of the seat armrest. The non-contact PPG sensor uses a specific wavelength light-emitting diode and a photodetector arranged opposite each other to extract pulse wave signals by detecting the periodic changes in capillary blood volume in the user's fingertips or palms. The skin conductivity contact electrode is made of inert metal and operates in constant voltage mode to measure the conductivity changes of the user's palm skin surface; the miniature wide-angle camera is concealed on the upper edge of the seat back and faces the user, with a built-in privacy filter chip, used to capture the user's facial video stream at a preset frame rate, and after completing face region detection and geometric feature extraction of the eye and mouth regions locally, discards the original video data and only uploads the extracted feature vectors; the local edge computing gateway has a built-in signal preprocessing module, feature extraction engine and local lightweight state evaluation model; the signal preprocessing module is used to perform noise reduction and normalization processing on the pressure data from the flexible pressure distribution sensing pad, and generate a user sitting posture pressure center trajectory map and a static pressure distribution heat map; The feature extraction engine is used to perform time-domain and frequency-domain analysis on the pulse wave signal from the non-contact photoplethysmography (PPG) sensor to calculate the ratio of low-frequency power to high-frequency power in heart rate and heart rate variability. This is used to perform event-related potential analysis on the skin conductivity signals from the skin conductivity contact electrodes to extract the baseline value and event-related fluctuation amplitude of the skin conductivity level; The system includes a cloud-based collaborative analysis and decision-making platform for performing temporal analysis on facial geometric feature vectors from the miniature wide-angle camera, calculating the rate of change in eye opening and closing and the sequence of changes in the angle of mouth upward movement between consecutive frames. The cloud-based collaborative analysis and decision-making platform includes a user state fusion analysis module, a healing strategy generation engine, and a system resource scheduler. The user state fusion analysis module receives temporal feature data streams from the local edge computing gateway and accesses real-time environmental data from the community weather station, running a multi-task learning model based on a deep neural network. The multi-task learning model takes the temporal feature data as input and outputs evaluation indices in parallel across three dimensions: physiological fatigue index, emotional valence index, and social interaction willingness index. The physiological fatigue index is quantified by comprehensively considering decreased heart rate variability, increased stress center trajectory swaying, and duration of holding a specific sitting posture. The emotional valence index ranges from -1 to +1, and is jointly inferred based on mouth corner movement, eye movement, and skin conductance response patterns. The social interaction willingness index is estimated by analyzing the deviation angle of the user's head orientation relative to the community path and the duration of solitude in the seat. The healing strategy generation engine is used to match and generate a personalized composite healing instruction sequence from a preset healing strategy library based on three evaluation indices output in real time by the user state fusion analysis module, combined with the current time, weather data, and user historical preference profile; the healing strategy library includes three basic healing strategies: environmental adjustment, sensory stimulation, and social promotion. The system resource scheduler is used to check the current status and energy consumption of each execution component in the environmental control and healing execution unit after the healing strategy generation engine outputs the composite healing command, and decompose the composite healing command into specific control commands and issue them for execution according to preset priority rules and coordination logic; The environmental control and healing execution unit includes a local temperature control system composed of a semiconductor cooling and heating element and a micro fan integrated under the seat surface, a full-band micro speaker array and multi-color light-emitting diode beads integrated on the top of the seat back, micro ultrasonic atomizers located at the front of the armrests on both sides of the seat, a low-power linear motor installed at the bottom of the seat, a low-power electronic ink display screen located on the side of the armrests, and a near-field sensing module based on Bluetooth low-power beacons.

2. The community elderly-friendly public seating system with health monitoring and emotional healing functions according to claim 1, characterized in that, The multi-task learning model in the user state fusion analysis module is updated using a periodic online incremental learning mechanism. The online incremental learning mechanism periodically extracts anonymized user data and healing effect feedback data from the cloud database as new training samples to fine-tune the parameters of the multi-task learning model.

3. A community-based elderly-friendly public seating system with health monitoring and emotional therapy functions as described in claim 1, characterized in that, The healing strategy generation engine introduces a long-term benefit optimization mechanism based on reinforcement learning when matching strategies. The long-term benefit optimization mechanism treats each healing intervention as a decision action and uses the degree of improvement in the user's physiological fatigue index and emotional valence index over a subsequent period as a reward signal. It continuously optimizes the mapping relationship from user state to healing strategy through a policy gradient algorithm.

4. A community-based elderly-friendly public seating system with health monitoring and emotional therapy functions as described in claim 1, characterized in that, The execution process of the adjacent seat linkage prompt function in the social promotion strategy is as follows: When the system evaluates that a user's social interaction willingness index is continuously higher than the threshold of 0.5 for more than 5 minutes, and the near field perception module detects that another seat within 3 meters to the side of the user is occupied, the system resource scheduler simultaneously pushes a lightweight anonymous interaction prompt to the e-ink display screens of the two seats.

5. A community-based elderly-friendly public seating system with health monitoring and emotional therapy functions as described in claim 1, characterized in that, The system also includes a visual data dashboard for community managers; the visual data dashboard is presented via a web page and displays the real-time and historical usage density of all public seats in the community, the trend of regional average emotional valence index changes, and the statistics of the frequency of use of various healing strategies in the form of heat maps.

6. A community-based elderly-friendly public seating system with health monitoring and emotional therapy functions as described in claim 1, characterized in that, The local lightweight state assessment model within the local edge computing gateway is used to take multimodal features within the most recent time window as input and output a binary warning sign indicating whether the user may be in a state of significant physical discomfort or emotional abnormality.

7. A community-based elderly-friendly public seating system with health monitoring and emotional therapy functions as described in claim 1, characterized in that, The signal preprocessing module uses an algorithm combining a moving average filter and a median filter to reduce noise in the pressure data; the normalization process divides the pressure value of each micro piezoresistive sensing unit by its historically obtained maximum possible pressure value, converting it into a relative pressure value between 0 and 1.

8. A community-based elderly-friendly public seating system with health monitoring and emotional therapy functions as described in claim 1, characterized in that, The feature extraction engine calculates the ratio of low-frequency power to high-frequency power for heart rate variability as follows: it performs a fast Fourier transform on a heart rate interval sequence of a preset duration to obtain the power spectral density, and then extracts the ratio of low-frequency power to high-frequency power.

9. A community-based elderly-friendly public seating system with health monitoring and emotional therapy functions as described in claim 1, characterized in that, The miniature ultrasonic atomizer in the environmental regulation and healing execution unit is connected to a replaceable plant essential oil capsule for atomizing and releasing specially formulated plant essential oil molecules when executing a targeted aroma release strategy.

10. A community-based elderly-friendly public seating system with health monitoring and emotional therapy functions as described in claim 1, characterized in that, The system resource scheduler, based on the collaborative logic requirement, simultaneously initiates a background soundscape generation strategy that matches the frequency of the soothing light and color illumination strategy when executing the soothing light and color illumination strategy.

Images