A health monitoring system for a geriatric robot

Abstract

The application discloses a health monitoring system of an old-age robot, relates to the technical field of health monitoring, constructs an indoor multi-modal data pool, generates an encrypted feature vector, establishes a steady physiological behavior baseline according to historical encrypted feature vectors, outputs a health risk grade and an intervention instruction set, converts the intervention instruction set into a robot action, a voice or a visual prompt instruction and issues the same for execution, and realizes health monitoring in all time periods and all scenes through integration of multi-modal sensors, without the need for the old people to wear equipment, adoption of feature compression and homomorphic encryption technology ensures data security, can update a threshold value in real time, accurately monitors the health status of the old people, carries out health risk prediction and personalized intervention in combination with behavior analysis such as falling and still abnormality, dynamically adjusts a prompt mode according to the ability of the old people, optimizes data processing efficiency through edge-cloud collaborative architecture, guarantees data privacy and stability, and improves the life quality and safety.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology, and in particular to a health monitoring system for an elderly care robot. Background Technology

[0002] Currently, traditional health monitoring systems mostly rely on single vital sign detection devices, and these devices often require elderly people to actively wear them or intervene, leading to certain limitations in practical applications. Furthermore, existing systems still lack the technology to monitor environmental changes, elderly people with mobility issues, or those requiring non-contact monitoring, failing to meet the needs for 24 / 7, all-around, and unobtrusive monitoring.

[0003] Currently, Chinese invention patent application number CN202410979754.X discloses a health monitoring method based on an elderly care robot and the elderly care robot thereof. The method includes: acquiring historical posts published by users on social media platforms based on the elderly care robot; extracting depressive symptom features, emotional course features, and text history features based on the historical posts using a preset large language model; integrating and concatenating the depressive symptom features, emotional course features, and text history features to construct a final feature vector; inputting the final feature vector into a pre-trained prediction model, and outputting a predicted value after multiple iterations of the prediction model; and generating an interpretive output based on the predicted value.

[0004] The aforementioned technologies are insufficient for continuous monitoring and personalized intervention of the vital signs and behavioral status of the elderly in all scenarios without wearing protective devices. Summary of the Invention

[0005] The technical problem solved by this invention is that existing technologies are unable to achieve continuous monitoring and personalized intervention of the vital signs and behavioral status of the elderly in all scenarios without wearing a device.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] A health monitoring system for an elderly care robot includes a vital signs data acquisition module, a data preprocessing and privacy protection module, an individual baseline modeling module, a risk prediction and intervention decision-making module, and an interactive execution module;

[0008] The vital signs data acquisition module is used to construct an indoor multimodal data pool;

[0009] The data preprocessing and privacy protection module is used to extract features, compress and securely encapsulate the multimodal data pool, and generate encrypted feature vectors.

[0010] The individual baseline modeling module is used to establish a steady-state physiological behavior baseline based on historical encrypted feature vectors;

[0011] The risk prediction and intervention decision-making module is used to compare real-time encrypted feature vectors with the baseline of steady-state physiological behavior and output health risk level and intervention instruction set;

[0012] The interactive execution module is used to convert the intervention instruction set into robot action, voice or visual prompts and issue them for execution.

[0013] Preferably, the vital signs data acquisition module includes a microseismic radar unit, a depth camera unit, a UWB positioning unit, and an environmental parameter unit;

[0014] The microseismic radar unit is used to transmit continuous frequency-modulated waves and analyze echo phase changes, and output respiratory interval data and cardiac interval data.

[0015] The depth camera unit is used to acquire time-of-flight depth maps and output attitude skeleton coordinate data based on a skeleton fitting algorithm.

[0016] The UWB positioning unit is used to capture joint centroid trajectory data in time difference positioning mode;

[0017] The environmental parameter unit is used to collect real-time data on carbon dioxide concentration, temperature, humidity, and illuminance, and output environmental status data.

[0018] Preferably, the data preprocessing and privacy protection module includes a feature compression unit, an encryption encapsulation unit, and a dynamic transmission unit;

[0019] The feature compression unit is used to perform attention weighting on respiratory interval data, heartbeat interval data, posture skeleton coordinate data, joint centroid trajectory data and environmental state data, delete redundant features and generate compressed feature vectors;

[0020] The encryption encapsulation unit is used to perform homomorphic encryption on the compressed feature vector to generate an encrypted feature vector.

[0021] The dynamic transmission unit is used to switch between 5G, Wi-Fi 6E and local cache to a transmission latency of less than a preset threshold when the transmission delay of the encrypted feature vector is lower than a preset threshold, based on the wireless channel quality.

[0022] Preferably, the individual baseline modeling module includes a temporal clustering unit, a pose mapping unit, and a threshold self-updating unit;

[0023] The temporal clustering unit is used to perform temporal clustering on historical encrypted feature vectors according to daily and weekly cyclic windows and output steady-state segment center curves;

[0024] The posture mapping unit is used to learn the coupling relationship between breathing, heartbeat and posture under sitting, lying and standing conditions, and generate a posture mapping model.

[0025] The threshold self-updating unit is used to update the individual baseline and record the version number when the center curve of the steady-state segment changes below the mean square error threshold for three consecutive periods.

[0026] Preferably, the risk prediction and intervention decision-making module includes a multi-task prediction unit, an anomaly assessment unit, and an intervention generation unit;

[0027] The multi-task prediction unit is used to receive the encrypted feature vector output by the data preprocessing and privacy protection module. The encrypted feature vector is arranged in a time series and includes respiratory interval data, heartbeat interval data, posture skeleton coordinate data, joint centroid trajectory data and environmental state data.

[0028] The encoder-decoder network distinguishes each data channel and extracts time-dependent features. The decoder generates heart rate trend sequence, respiratory trend sequence, fall warning probability value sequence and long-term immobility prediction value sequence according to the task identifier.

[0029] The heart rate trend sequence, respiratory trend sequence, fall warning probability value sequence, and long-term immobility prediction value sequence are mapped one-to-one with the input timestamp and stored in the prediction cache area;

[0030] The anomaly assessment unit is used to read the heart rate trend sequence and respiratory trend sequence in the prediction buffer, and call the center curve of the steady-state physiological behavior baseline saved by the individual baseline modeling module.

[0031] The real-time drift is calculated using vector difference operations, and confidence values ​​for cardiac abnormalities and respiratory abnormalities are generated using a Mahalanobis distance-based metric.

[0032] Set the probability threshold for the probability value sequence of fall warning signs and generate fall anomaly confidence values;

[0033] The long-term stationary prediction value sequence is compared with the activity threshold of the attitude skeleton coordinate data to generate a stationary anomaly confidence value.

[0034] The anomaly confidence values ​​are mapped to the original timestamps, combined into anomaly confidence vectors, and stored in the evaluation results table;

[0035] The intervention generation unit is used to read the evaluation result table and parse the anomaly confidence vector according to the priority order of anomaly types.

[0036] The intervention generation unit has a built-in abnormal type intervention template mapping table and an elderly historical feedback record table. It sequentially calls the abnormal type intervention template mapping table and the elderly historical feedback record table to generate an initial intervention instruction, and performs weight adjustment according to the corresponding abnormal type and interaction method entries in the historical feedback record table, and outputs the adjustment intervention instruction.

[0037] The adjustment intervention instructions include action instructions, voice instructions, and visual prompt instructions. After being formatted into a unified instruction set queue, they are written into the intervention instruction set queue and accompanied by an abnormal confidence index for scheduling by the interactive execution module.

[0038] Preferably, the interactive execution module includes an interactive mode matching unit, a load balancing unit, and a feedback recording unit;

[0039] The interaction matching unit is used to combine the elderly’s hearing, vision and cognitive assessment files to select a set of interaction methods from voice prompts, facial LED prompts and robotic arm touch prompts.

[0040] The load balancing unit is used to calculate the attention resources occupied by the interaction and adjust the prompt frequency so that the total prompt duration per unit time does not exceed the preset attention load limit.

[0041] The feedback recording unit is used to record the elderly person's response delay, action completion rate, and facial expression similarity, and write the interaction validity into the feedback database.

[0042] Preferably, the feature compression unit performs weight allocation on multimodal features through a cross-attention mechanism, retains the features with the highest correlation to respiratory interval data, heart rate interval data, and posture skeleton coordinate data, and deletes redundant features until the feature dimension is reduced by a preset first percentage after compression.

[0043] Preferably, when the threshold self-updating unit detects that the average offset between the center curve of the steady-state segment and the current baseline is lower than the preset mean square error threshold for three consecutive periods, it automatically updates the individual baseline and generates a new version number.

[0044] Preferably, the intervention generation unit maps the abnormality type to the intervention method by looking up a table. When the probability of a fall warning is higher than a preset threshold, it generates a robotic arm support action command. When the confidence level of a heartbeat abnormality is higher than a preset threshold, it generates a remote call command.

[0045] Preferably, the load balancing unit dynamically adjusts the prompt interval by calculating the elderly person's gaze dwell time and voice response delay in real time, and automatically extends the next prompt interval when the attention resource occupancy rate exceeds the preset upper limit.

[0046] The beneficial effects of this invention are as follows: This invention achieves all-time, all-scenario health monitoring by integrating multimodal sensors, eliminating the need for elderly people to wear devices. It employs feature compression and homomorphic encryption technologies to ensure data security, updates thresholds in real time, accurately monitors the health status of the elderly, and combines behavioral analysis such as falls and abnormal stillness to predict health risks and provide personalized interventions. It dynamically adjusts prompts according to the elderly's abilities and optimizes data processing efficiency through an edge-cloud collaborative architecture, ensuring data privacy and stability, and improving quality of life and safety. Attached Figure Description

[0047] Figure 1 This is a basic flowchart of a health monitoring system for an elderly care robot provided in one embodiment of the present invention;

[0048] Figure 2 This is a physical illustration of an application of the present invention;

[0049] Figure 3 This diagram illustrates the effect of a risk prediction and intervention decision-making module provided in one embodiment of the present invention. Detailed Implementation

[0050] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0051] Example, refer to Figure 1 This paper presents a health monitoring system for elderly care robots, which includes a vital signs data acquisition module, a data preprocessing and privacy protection module, an individual baseline modeling module, a risk prediction and intervention decision-making module, and an interactive execution module.

[0052] The vital signs data acquisition module is used to build an indoor multimodal data pool.

[0053] The data preprocessing and privacy protection module is used to extract features, compress and securely encapsulate the multimodal data pool, and generate encrypted feature vectors.

[0054] The individual baseline modeling module is used to establish a steady-state physiological behavior baseline based on historical encrypted feature vectors.

[0055] The risk prediction and intervention decision-making module compares real-time encrypted feature vectors with the baseline of steady-state physiological behavior, and outputs the health risk level and intervention instruction set.

[0056] The interactive execution module is used to convert intervention instruction sets into robot actions, voice or visual prompts and issue them for execution.

[0057] This invention achieves all-time, all-scenario health monitoring by integrating multimodal sensors, eliminating the need for elderly people to wear devices. It employs feature compression and homomorphic encryption technologies to ensure data security. Based on an individualized baseline model, it can update thresholds in real time to accurately monitor the health status of the elderly. Through multi-task prediction, combined with behavioral analysis such as falls and abnormal stillness, it can predict health risks and provide personalized interventions. The system dynamically adjusts prompts according to the elderly person's abilities and optimizes data processing efficiency through an edge-cloud collaborative architecture, ensuring data privacy and stability, and improving quality of life and safety.

[0058] The vital signs data acquisition module includes a microseismic radar unit, a depth camera unit, a UWB positioning unit, and an environmental parameter unit.

[0059] The microseismic radar unit is used to transmit continuous frequency modulated waves and analyze the echo phase changes, outputting respiratory interval data and cardiac interval data.

[0060] The microseismic radar unit accurately detects respiratory intervals and cardiac intervals by emitting continuous frequency-modulated waves and analyzing echo phase changes. This unit can monitor the breathing and heart rate of elderly individuals without them noticing, making it suitable for situations where wearing a device is not possible, such as at night or during exercise. The output data can be used for early warning of diseases such as heart disease and sleep apnea.

[0061] The depth camera unit is used to acquire time-of-flight depth maps and outputs attitude skeleton coordinate data based on a skeleton fitting algorithm.

[0062] The depth camera unit utilizes time-of-flight technology to generate depth maps and employs a skeleton fitting algorithm to extract posture skeleton coordinate data, accurately monitoring postural changes in the elderly. This unit plays a crucial role in fall detection, activity assessment, and monitoring of changes in motor ability, and can operate stably in complex environments.

[0063] The UWB positioning unit is used to capture joint centroid trajectory data in time difference positioning mode.

[0064] The UWB positioning unit uses time-of-flight positioning technology to capture the joint center of mass trajectory data of elderly individuals, helping to accurately track their movement trajectories and activity patterns. This unit can monitor in real time whether the elderly person is active within a preset area and promptly detect abnormal behavior or falls.

[0065] The environmental parameter unit is used to collect real-time data on carbon dioxide concentration, temperature, humidity, and illuminance, and output environmental status data.

[0066] The environmental parameter unit collects real-time environmental data such as carbon dioxide concentration, temperature, humidity, and illuminance to assess the health and safety of the environment in which the elderly live. For example, abnormal carbon dioxide concentration or changes in temperature and humidity may indicate air quality problems or temperature control malfunctions, thus providing a safer living environment for the elderly.

[0067] The main function of the vital signs data acquisition module is to provide comprehensive, real-time, and non-invasive health data collection. It comprehensively monitors the vital signs, behavioral postures, and environmental conditions of the elderly through multiple sensing technologies. This module can operate continuously in complex environments, ensuring the continuity and accuracy of health monitoring. By integrating different types of data sources, this module can provide real-time vital signs data, supporting subsequent data processing, analysis, and health assessment.

[0068] The data preprocessing and privacy protection module includes a feature compression unit, an encryption encapsulation unit, and a dynamic transmission unit.

[0069] The feature compression unit is used to perform attention weighting on respiratory interval data, heartbeat interval data, posture skeleton coordinate data, joint centroid trajectory data and environmental state data, remove redundant features and generate compressed feature vectors.

[0070] The feature compression unit assigns weights to multimodal features through a cross-attention mechanism, retains the features with the highest correlation to respiratory interval data, heart rate interval data, and posture skeleton coordinate data, and deletes redundant features until the feature dimension is reduced by a preset first percentage after compression.

[0071] The feature compression unit uses an attention-weighted mechanism to perform multi-dimensional feature filtering on respiratory interval data, heart rate interval data, posture skeletal coordinate data, joint centroid trajectory data, and environmental state data, removing redundant information irrelevant to the target and generating compressed feature vectors. A cross-attention mechanism is used to assign weights to multimodal features, ensuring that data most relevant to changes in vital signs is retained, reducing the feature dimensionality to a preset range, decreasing computational burden, and improving the efficiency and accuracy of subsequent processing.

[0072] The encryption encapsulation unit is used to perform homomorphic encryption on the compressed feature vector to generate an encrypted feature vector.

[0073] After compression, the encryption encapsulation unit performs homomorphic encryption on the feature vector to generate an encrypted feature vector, which can be directly used for calculation and analysis in the cloud or on collaborative nodes without decryption. This process ensures the privacy and security of vital signs and behavioral data, avoiding the leakage of sensitive information during transmission and storage.

[0074] The dynamic transmission unit is used to switch between 5G, Wi-Fi 6E and local cache to a transmission latency of less than a preset threshold based on the wireless channel quality.

[0075] The dynamic transmission unit monitors the wireless channel status in real time and dynamically switches between 5G, Wi-Fi 6E, and local buffering based on latency and bandwidth conditions to ensure the stability and timeliness of encrypted feature vector transmission. By scheduling different transmission paths, data transmission latency is controlled within a preset threshold, providing reliable data input for subsequent real-time risk prediction.

[0076] The main function of the data preprocessing and privacy protection module is to efficiently process and securely encapsulate the collected multimodal vital signs and environmental data, providing high-quality input for subsequent analysis and ensuring data privacy throughout the entire process. This module achieves efficient data transmission and privacy protection through feature extraction, redundant compression, and homomorphic encryption, enabling the system to complete health status analysis without exposing raw sensitive information.

[0077] The individual baseline modeling module includes a temporal clustering unit, a pose mapping unit, and a threshold self-updating unit.

[0078] The temporal clustering unit is used to perform temporal clustering on historical encrypted feature vectors in daily and weekly cyclic windows and output the center curve of the steady-state segment.

[0079] The temporal clustering unit performs temporal clustering on historical encrypted feature vectors using daily and weekly cyclic windows, outputting the center curve of the steady-state segment. This curve describes the long-term patterns of vital signs and posture in the elderly under stable conditions. Through this unit, the system can extract periodic patterns and construct individualized reference curves, providing a benchmark for subsequent anomaly detection.

[0080] The posture mapping unit is used to learn the coupling relationship between breathing, heartbeat and posture under sitting, lying and standing conditions, and to generate a posture mapping model.

[0081] The posture mapping unit analyzes the coupling relationship between respiration, heart rate, and posture in sitting, lying, and standing positions to generate a posture mapping model, enabling dynamic correction of vital sign data by posture changes. This model ensures the continuity and comparability of monitoring data under different postures, allowing anomaly detection to be based on a unified individual baseline.

[0082] The threshold self-updating unit is used to update the individual baseline and record the version number when the change of the center curve in the steady-state section is lower than the mean square error threshold for three consecutive periods.

[0083] When the threshold self-updating unit detects that the average offset between the center curve of the steady-state segment and the current baseline is lower than the preset mean square error threshold for three consecutive periods, it automatically updates the individual baseline and generates a new version number.

[0084] The threshold self-updating unit monitors the deviation of the steady-state segment center curve from the current individual baseline. When the change is lower than the set mean square error threshold for three consecutive periods, it automatically updates the individual baseline and records the version number. Through adaptive threshold adjustment, this unit keeps the baseline consistent with the elderly person's current health status, ensuring long-term accuracy and sensitivity of the monitoring.

[0085] The main function of the individual baseline modeling module is to establish individualized physiological and behavioral baselines for older adults based on historical multimodal health data, enabling differentiated monitoring of different individuals. Through dynamic modeling and adaptive updates, this module can provide long-term stable reference curves, providing a reliable basis for anomaly detection and health trend analysis, and avoiding false alarms and false negatives caused by using uniform thresholds.

[0086] The risk prediction and intervention decision-making module includes a multi-task prediction unit, an anomaly assessment unit, and an intervention generation unit.

[0087] The multi-task prediction unit is used to receive encrypted feature vectors output by the data preprocessing and privacy protection module. The encrypted feature vectors are arranged in time series and include respiratory interval data, heartbeat interval data, posture skeleton coordinate data, joint centroid trajectory data, and environmental state data.

[0088] The encoder-decoder network distinguishes each data channel and extracts time-dependent features. The decoder generates heart rate trend sequence, respiratory trend sequence, fall warning probability value sequence and long-term immobility prediction value sequence according to the task identifier.

[0089] The heart rate trend sequence, respiratory trend sequence, fall warning probability value sequence, and long-term immobility prediction value sequence are mapped one-to-one with the input timestamps and stored in the prediction cache.

[0090] The multi-task prediction unit utilizes time-series encrypted feature vectors to perform multi-task parallel modeling of respiration, heartbeat, posture, and trajectory data, generating predictions for heartbeat trends, respiration trends, fall prognoses, and prolonged immobility. By aligning the output sequences with time and writing them to the prediction buffer, it provides high-time-accuracy predictions of health changes, laying the foundation for subsequent anomaly detection.

[0091] The anomaly assessment unit is used to read the heart rate trend sequence and respiratory trend sequence in the prediction buffer and call the center curve of the steady-state physiological behavior baseline saved by the individual baseline modeling module.

[0092] Real-time drift is calculated using vector difference operations, and confidence values ​​for cardiac and respiratory abnormalities are generated using a Mahalanobis distance-based metric.

[0093] Set the probability threshold for the probability value sequence of fall warning signs and generate fall anomaly confidence values.

[0094] The static anomaly confidence value is generated by comparing the long-term stationary predicted value sequence with the activity threshold of the attitude skeleton coordinate data.

[0095] The anomaly confidence values ​​are mapped to the original timestamps, combined into an anomaly confidence vector, and stored in the evaluation results table.

[0096] The anomaly assessment unit performs differential analysis between the predicted results and the steady-state center curve output by the individual baseline modeling module, calculates the real-time drift of heartbeat and respiration, and generates anomaly confidence scores using Mahalanobis distance. Combining the fall warning probability with the resting threshold, a complete anomaly confidence vector is formed, achieving multi-dimensional quantification of health risk and providing clear data support for intervention strategies.

[0097] The intervention generation unit is used to read the evaluation result table and parse the anomaly confidence vector according to the priority order of the anomaly type.

[0098] The intervention generation unit has a built-in abnormal type intervention template mapping table and an elderly historical feedback record table. It sequentially calls the abnormal type intervention template mapping table and the elderly historical feedback record table to generate initial intervention instructions, and adjusts the weights according to the corresponding abnormal type and interaction method entries in the historical feedback record table, and outputs the adjusted intervention instructions.

[0099] The intervention instructions include action instructions, voice instructions, and visual cues. After being formatted into a unified instruction set queue, they are written into the intervention instruction set queue and accompanied by an anomaly confidence index for scheduling by the interactive execution module.

[0100] The intervention generation unit maps the abnormality type to the intervention method by looking up a table. When the probability of a fall warning is higher than a preset threshold, it generates a robotic arm support action command. When the confidence level of a heartbeat abnormality is higher than a preset threshold, it generates a remote call command.

[0101] The intervention generation unit parses the abnormal confidence vector and calls the preset intervention template library and historical feedback record table to generate action, voice, and visual prompts. Through priority and feedback weighting adjustments, it outputs the intervention plan that best suits the current state and uniformly formats it into an intervention instruction set for the execution module to call, realizing an immediate closed-loop response to health risks.

[0102] The risk prediction and intervention decision-making module transforms multimodal vital signs and behavioral data into actionable health risk assessment and intervention instructions. Through trend prediction, anomaly assessment, and instruction generation, this module achieves closed-loop management from real-time data to proactive intervention, ensuring that changes in the health status of the elderly can be identified promptly and targeted measures can be taken, forming a continuous risk control chain.

[0103] The interactive execution module includes an interactive mode matching unit, a load balancing unit, and a feedback recording unit.

[0104] The interaction matching unit is used to combine the elderly person's hearing, vision and cognitive assessment files to select a set of interaction methods from voice prompts, facial LED prompts and robotic arm touch prompts.

[0105] The interaction matching unit selects a combination of voice prompts, facial LED prompts, and robotic arm touch prompts based on the elderly person's hearing, vision, and cognitive assessment records, ensuring that the prompting method matches the elderly person's perceptual abilities. Through precise matching, this unit improves the delivery rate of intervention instructions and reduces information loss.

[0106] The load balancing unit is used to calculate the attention resources occupied by the interaction and adjust the prompt frequency so that the total prompt duration per unit time does not exceed the preset attention load limit.

[0107] The load balancing unit dynamically adjusts the prompt interval by calculating the elderly person's gaze dwell time and voice response delay in real time. When the attention resource occupancy rate exceeds the preset limit, it automatically extends the next prompt interval.

[0108] The load balancing unit calculates the attentional resources consumed by the interaction and adjusts the frequency and interval of prompts based on the real-time interaction load. By monitoring the elderly person's gaze dwell time and voice response delay, when the attentional resource occupancy rate exceeds a preset limit, the interval between the next prompt is automatically extended to ensure that the interaction is within an acceptable cognitive range and to avoid overstimulation or neglect.

[0109] The feedback recording unit is used to record the elderly's response delay, action completion rate, and facial expression similarity, and write the effectiveness of the interaction into the feedback database.

[0110] The feedback recording unit records the elderly person's response delay, action completion rate, and facial expression similarity, and writes this interaction effectiveness data into the feedback database. Through continuously accumulated feedback information, the system can dynamically optimize prompting strategies, improving the efficiency of intervention execution and long-term adaptability.

[0111] The interactive execution module transforms the intervention instructions generated by the risk prediction and intervention decision-making module into multimodal prompts tailored to the individual characteristics of the elderly, thus achieving a closed-loop human-computer interaction for health management. This module can select appropriate interaction methods based on the elderly person's hearing, vision, and cognitive level, and dynamically adjust the interaction strategy by monitoring interaction load and feedback data in real time. This ensures that intervention instructions are effectively perceived and executed, while simultaneously creating an interactive database that can be optimized.

[0112] This system integrates multimodal sensors, utilizing devices such as micro-seismic radar, depth cameras, UWB positioning units, and environmental parameter acquisition units to comprehensively collect vital signs, behavioral postures, joint movements, and environmental data of the elderly. This non-intrusive monitoring method eliminates the need for the elderly to actively wear devices, solving the inconvenience of traditional wearable devices and enabling all-time, all-scenario health monitoring. Through data preprocessing and privacy protection modules, feature compression and homomorphic encryption technologies are employed to ensure data security and privacy protection during transmission. Simultaneously, the feature compression unit effectively removes redundant data, retaining only key health features and improving the system's data processing efficiency. Based on historical encrypted feature vectors, the system generates individualized physiological behavioral baselines through temporal clustering and posture mapping methods. Each elderly person has an independent individual baseline, and the system can dynamically update thresholds based on baseline differences, ensuring the accuracy and personalization of health monitoring and avoiding the false alarms and missed alarms of traditional uniform thresholds. The system uses a multi-task prediction unit to predict trends in important vital signs such as heart rate and respiration, while combining this with behavioral analysis of fall warning signs and abnormal rest to monitor health risks in real time. The anomaly assessment unit generates a risk assessment based on the prediction results and, in conjunction with the intervention generation unit, provides personalized intervention instructions, including voice prompts and robotic arm support, to intervene and respond promptly, preventing major health events. Through the interactive execution module, the system can dynamically adjust the interaction method according to the elderly person's hearing, vision, and cognitive abilities. The load balancing unit can reasonably arrange the interaction frequency to ensure that the elderly person receives timely health guidance and support without incurring cognitive burden. The feedback recording unit continuously optimizes the interaction effect and enhances the system's adaptability by tracking the elderly person's responses. This system adopts an edge-cloud collaborative architecture, utilizing edge computing for data processing and encryption to ensure data privacy and efficient transmission; simultaneously, the cloud performs deep analysis and prediction, providing the system with continuous learning and optimization capabilities. The collaborative application of multiple technologies not only improves the system's real-time performance but also ensures stability and reliability during long-term, high-intensity use. In summary, the health monitoring system of the elderly care robot of this invention, by integrating multimodal sensors, personalized physiological modeling, and intelligent health prediction and intervention, can not only comprehensively and seamlessly monitor the health status of the elderly but also adjust intervention strategies based on real-time data and the elderly person's feedback, thereby improving the quality of life and safety of the elderly.

[0113] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0114] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A health monitoring system for an elderly care robot, characterized in that, It includes a vital signs data acquisition module, a data preprocessing and privacy protection module, an individual baseline modeling module, a risk prediction and intervention decision-making module, and an interactive execution module; The vital signs data acquisition module is used to construct an indoor multimodal data pool; The data preprocessing and privacy protection module is used to extract features, compress and securely encapsulate the multimodal data pool, and generate encrypted feature vectors. The individual baseline modeling module is used to establish a steady-state physiological behavior baseline based on historical encrypted feature vectors; The risk prediction and intervention decision-making module is used to compare real-time encrypted feature vectors with the baseline of steady-state physiological behavior and output health risk level and intervention instruction set; The interactive execution module is used to convert the intervention instruction set into robot action, voice or visual prompts and issue them for execution. The risk prediction and intervention decision-making module includes a multi-task prediction unit, an anomaly assessment unit, and an intervention generation unit. The multi-task prediction unit is used to receive the encrypted feature vector output by the data preprocessing and privacy protection module. The encrypted feature vector is arranged in a time series and includes respiratory interval data, heartbeat interval data, posture skeleton coordinate data, joint centroid trajectory data and environmental state data. The encoder-decoder network distinguishes each data channel and extracts time-dependent features. The decoder generates heart rate trend sequence, respiratory trend sequence, fall warning probability value sequence and long-term immobility prediction value sequence according to the task identifier. The heart rate trend sequence, respiratory trend sequence, fall warning probability value sequence, and long-term immobility prediction value sequence are mapped one-to-one with the input timestamp and stored in the prediction cache area; The anomaly assessment unit is used to read the heart rate trend sequence and respiratory trend sequence in the prediction buffer, and call the center curve of the steady-state physiological behavior baseline saved by the individual baseline modeling module. The real-time drift is calculated using vector difference operations, and confidence values ​​for cardiac abnormalities and respiratory abnormalities are generated using a Mahalanobis distance-based metric. Set the probability threshold for the probability value sequence of fall warning signs and generate fall anomaly confidence values; The long-term stationary prediction value sequence is compared with the activity threshold of the attitude skeleton coordinate data to generate a stationary anomaly confidence value. The anomaly confidence values ​​are mapped to the original timestamps, combined into anomaly confidence vectors, and stored in the evaluation results table; The intervention generation unit is used to read the evaluation result table and parse the anomaly confidence vector according to the priority order of anomaly types. The intervention generation unit has a built-in abnormal type intervention template mapping table and an elderly historical feedback record table. It sequentially calls the abnormal type intervention template mapping table and the elderly historical feedback record table to generate an initial intervention instruction, and performs weight adjustment according to the corresponding abnormal type and interaction method entries in the historical feedback record table, and outputs the adjustment intervention instruction. The adjustment intervention instructions include action instructions, voice instructions, and visual prompt instructions. After being formatted into a unified instruction set queue, they are written into the intervention instruction set queue and accompanied by an abnormal confidence index for scheduling by the interactive execution module.

2. The health monitoring system for an elderly care robot as described in claim 1, characterized in that, The vital signs data acquisition module includes a microseismic radar unit, a depth camera unit, a UWB positioning unit, and an environmental parameter unit. The microseismic radar unit is used to transmit continuous frequency-modulated waves and analyze echo phase changes, and output respiratory interval data and cardiac interval data. The depth camera unit is used to acquire time-of-flight depth maps and output attitude skeleton coordinate data based on a skeleton fitting algorithm. The UWB positioning unit is used to capture joint centroid trajectory data in time difference positioning mode; The environmental parameter unit is used to collect real-time data on carbon dioxide concentration, temperature, humidity, and illuminance, and output environmental status data.

3. The health monitoring system for an elderly care robot as described in claim 2, characterized in that, The data preprocessing and privacy protection module includes a feature compression unit, an encryption and encapsulation unit, and a dynamic transmission unit. The feature compression unit is used to perform attention weighting on respiratory interval data, heartbeat interval data, posture skeleton coordinate data, joint centroid trajectory data and environmental state data, delete redundant features and generate compressed feature vectors; The encryption encapsulation unit is used to perform homomorphic encryption on the compressed feature vector to generate an encrypted feature vector. The dynamic transmission unit is used to switch between 5G, Wi-Fi 6E and local cache to a transmission latency of less than a preset threshold when the transmission delay of the encrypted feature vector is lower than a preset threshold, based on the wireless channel quality.

4. The health monitoring system for an elderly care robot as described in claim 3, characterized in that, The individual baseline modeling module includes a temporal clustering unit, a pose mapping unit, and a threshold self-updating unit; The temporal clustering unit is used to perform temporal clustering on historical encrypted feature vectors according to daily and weekly cyclic windows and output steady-state segment center curves; The posture mapping unit is used to learn the coupling relationship between breathing, heartbeat and posture under sitting, lying and standing conditions, and generate a posture mapping model. The threshold self-updating unit is used to update the individual baseline and record the version number when the center curve of the steady-state segment changes below the mean square error threshold for three consecutive periods.

5. The health monitoring system for an elderly care robot as described in claim 1, characterized in that, The interactive execution module includes an interactive mode matching unit, a load balancing unit, and a feedback recording unit; The interaction matching unit is used to combine the elderly’s hearing, vision and cognitive assessment files to select a set of interaction methods from voice prompts, facial LED prompts and robotic arm touch prompts. The load balancing unit is used to calculate the attention resources occupied by the interaction and adjust the prompt frequency so that the total prompt duration per unit time does not exceed the preset attention load limit. The feedback recording unit is used to record the elderly person's response delay, action completion rate, and facial expression similarity, and write the interaction validity into the feedback database.

6. The health monitoring system for an elderly care robot as described in claim 3, characterized in that, The feature compression unit assigns weights to multimodal features through a cross-attention mechanism, retains the features with the highest correlation to respiratory interval data, heart rate interval data, and posture skeleton coordinate data, and deletes redundant features until the feature dimension is reduced by a preset first percentage after compression.

7. The health monitoring system for an elderly care robot as described in claim 4, characterized in that, When the threshold self-updating unit detects that the average offset between the center curve of the steady-state segment and the current baseline is lower than the preset mean square error threshold for three consecutive cycles, it automatically updates the individual baseline and generates a new version number.

8. The health monitoring system for an elderly care robot as described in claim 7, characterized in that, The intervention generation unit maps the abnormality type to the intervention method by looking up a table. When the probability of a fall warning is higher than a preset threshold, it generates a robotic arm support action command. When the confidence level of a heartbeat abnormality is higher than a preset threshold, it generates a remote call command.

9. The health monitoring system for an elderly care robot as described in claim 5, characterized in that, The load balancing unit dynamically adjusts the prompt interval by calculating the elderly person's gaze dwell time and voice response delay in real time. When the attention resource occupancy rate exceeds the preset limit, it automatically extends the next prompt interval.

Images