Intelligent old-age service management system and control method

Abstract

The application discloses a kind of wisdom endowment service management systems and control method, including double sensing data acquisition unit, edge intelligent computing unit, early warning protection linkage unit and service management background unit.The application is not worn, breaks through the use barrier, adapts to all age old people group;Through double-sensing isomerism fusion, accurate algorithm, solve single-sensing identification bottleneck, high recognition accuracy;Through fall precursor, early warning, early prediction grading, form protection closed loop, improve the value of old-age security;Utilize edge integration deployment, low-power design, give consideration to privacy security and low-cost popularization;System architecture is perfect, adapts to home scene, and operation and maintenance cost is low.

Description

Technical Field

[0001] This application relates to the field of intelligent management system technology, and in particular to a smart elderly care service management system and control method. Background Technology

[0002] With the accelerating aging of the population, the safety of elderly people living alone at home is becoming increasingly prominent. Among these risks, falls when getting up to use the toilet at night are one of the core threats to their lives and health. Statistics show that falls at night account for more than 65% of all falls among elderly people living alone. Furthermore, because no one is there to provide timely assistance after a fall, serious complications such as fractures and intracranial hemorrhage are easily caused. Therefore, accurate identification of nighttime falls and early warning systems for elderly people living alone have become a core requirement in the field of smart elderly care service management.

[0003] In existing smart elderly care service management systems, the technical solutions for elderly people getting up and falling early warning can be mainly divided into three categories: The first category is wearable device solutions, which collect body posture or movement data by having the elderly wear smart bracelets, watches, and other devices to monitor getting up and falling; the second category is single-sensor non-wearable solutions, which mostly use millimeter-wave radar or cameras as a single sensing unit and are deployed in the bedroom to collect motion data; the third category is simple alarm linkage solutions, which use trigger sensors (such as bed edge pressure sensors) to detect when the elderly person gets out of bed and then trigger an alarm.

[0004] However, the above technical solution has the following drawbacks:

[0005] Elderly people living alone, especially the very old and frail, generally have a strong aversion to wearable devices, and wearing the devices at night can seriously affect their sleep quality. At the same time, the elderly are prone to forgetting to wear the devices when they get up at night, or the devices may not be worn securely due to hand tremors or loose skin, causing data collection interruptions or deviations, making it impossible to achieve continuous and stable monitoring and resulting in extremely low reliability of early warnings.

[0006] In bedroom scenarios at night, the use of a single millimeter-wave radar can lead to problems. The electromagnetic shielding of the radar signal by bedding can cause the point cloud data of an elderly person's movement to be obscured by the subtle movements of the bedding, resulting in misjudgments of the intention to get up (e.g., misinterpreting turning over or stretching as getting up) or missed judgments (e.g., failing to recognize a slight movement by a frail elderly person). Using a camera presents serious privacy risks, and the recognition accuracy drops significantly in low-light conditions at night. Furthermore, it requires substantial computing power, making low-cost deployment in home settings difficult. In addition, existing single-sensor solutions often employ fixed threshold judgment modes, which cannot adapt to the individualized getting-up habits of different elderly individuals (e.g., frail, hemiplegic, or unilaterally-leaning elderly people), further reducing recognition accuracy.

[0007] Most existing solutions can only detect falls after they occur, failing to anticipate pre-fall warning signs. This results in the elderly falling before the alarm is triggered, missing the best opportunity for rescue. Some solutions with warning signs do not combine the elderly person's health level with the risk of the environment (such as slippery ground) to provide tiered warnings, which can easily lead to ineffective responses from children and caregivers, resulting in a waste of care resources. At the same time, most existing solutions lack proactive protection linkage mechanisms, only able to remind relevant personnel through sound or SMS alarms, failing to proactively reduce the probability of falls from the environmental level. A closed loop of warning and protection has not yet been formed.

[0008] Therefore, a smart elderly care service management system and control method are proposed. Summary of the Invention

[0009] This application aims to at least partially solve one of the technical problems in the aforementioned technologies.

[0010] To achieve the above objectives, the first aspect of this application proposes a smart elderly care service management system, including a dual-sensor data acquisition unit, an edge intelligent computing unit, an early warning and protection linkage unit, and a service management backend unit;

[0011] The dual-sensor data acquisition unit includes a main sensing module and an auxiliary sensing module. The main sensing module is a 24 GHz narrow beam millimeter-wave radar module, which is used to collect three-dimensional spatial attitude point cloud data of elderly people living alone at home during the process of getting up. The auxiliary sensing module is an ultra-thin patch piezoelectric micro-vibration sensor, which is used to collect micro-vibration force, vibration frequency and time sequence characteristic data of the bed during the process of elderly people getting up.

[0012] The edge intelligent computing unit is an STM32H743 microcontroller. The edge intelligent computing unit is connected to the dual-sensor data acquisition unit by wires and is used to receive the three-dimensional spatial attitude point cloud data and micro-vibration data output by the dual-sensor data acquisition unit. The edge intelligent computing unit has built-in dual-source noise collaborative filtering algorithm, standing-specific heterogeneous fusion feature set temporal linkage judgment algorithm, edge lightweight individual adaptive temporal learning algorithm, fall precursor multi-dimensional accurate prediction algorithm and graded early warning logic.

[0013] The early warning and protection linkage unit includes a soft night light, a lightweight vibration reminder, and a magnetic handrail unlocking module. The early warning and protection linkage unit is wirelessly connected to the edge intelligent computing unit in low power to receive the early warning signal and protection linkage signal output by the edge intelligent computing unit and execute the corresponding early warning action and protection linkage action.

[0014] The service management backend unit is wirelessly connected to the edge intelligent computing unit to receive and store the elderly getting up records, risk warning records and equipment operation status data uploaded by the edge intelligent computing unit. It supports children, community grid workers and caregivers to access and view and perform manual fine-tuning operations through mobile app.

[0015] By constructing a complete system architecture consisting of a dual-sensor data acquisition unit, an edge intelligent computing unit, an early warning and protection linkage unit, and a service management backend unit, the system achieves collaborative work among multiple units. This addresses the shortcomings of existing technologies, such as low accuracy of single-sensor identification and lack of system-level management and linkage. The heterogeneous fusion of dual sensors ensures comprehensive data acquisition from the hardware level, edge computing ensures efficient local processing, and the service management backend supplements the management and interaction needs of elderly care services. At the same time, the dual-sensor mutual verification mechanism reduces interference from invalid signals at the source, thereby improving the reliability of system operation.

[0016] In addition, the smart elderly care service management system and control method proposed in this application may also have the following additional technical features:

[0017] As a further description of the above technical solution:

[0018] The 24GHz narrow-beam millimeter-wave radar module has a ranging range of 0.1 meters to 5 meters, a detection angle of ±30 degrees, and a power consumption of no more than 0.5 watts. It supports switching between a 1Hz low-frequency scanning mode and a 10Hz high-frequency scanning mode. The ultra-thin patch-type piezoelectric micro-vibration sensor has a sampling frequency of 5Hz to 100Hz and a power consumption of no more than 0.1 watts. It adopts low-cost, low-power civilian-grade sensing parameters, solving the shortcomings of existing technologies such as high computing power and high cost of camera solutions, or high power consumption and difficulty in 24-hour operation of other sensing solutions. The high and low frequency scanning and sleep-wake mode switching design takes into account the monitoring accuracy and low power consumption requirements, ensuring that the system can be deployed stably at home for a long time and reducing the cost of use.

[0019] As a further description of the above technical solution:

[0020] The 24 GHz narrow beam millimeter-wave radar module is wall-mounted above the head of the elderly person's bed at a height of 1.8 to 2.0 meters and a downward angle of 30 to 45 degrees. The detection area covers the elderly person's upper body and hip core area for getting up.

[0021] The ultra-thin patch-type piezoelectric micro-vibration sensor is attached to the bed frame, legs, or under the bed board without drilling. The collection area covers the core vibration area of ​​the bed when the elderly get up and prop themselves up on the bed, and when their buttocks leave the bed. It has a tolerance of no less than ±5 cm for installation position deviation. It precisely defines the radar installation position, angle, and micro-vibration sensor attachment area in the small bedroom and the pain points of bedding obstruction, solving the monitoring blind spots and data acquisition deviation problems caused by unreasonable sensor deployment in existing technologies. The patch-type installation without drilling and the tolerance design reduce the difficulty of system deployment, adapt to various types of home beds, and improve the universality of home scenarios.

[0022] As a further description of the above technical solution:

[0023] The dual-source noise collaborative filtering algorithm includes radar-side filtering. Based on the electromagnetic reflectivity characteristics of 24 GHz radar, the reflectivity threshold difference between the human body and the bedding is calibrated, dynamic reflectivity filtering rules are set, and bedding point cloud noise with a reflectivity lower than that of the human body is filtered out. At the same time, based on the temporal continuity of the action, discrete point clouds of irregular micro-movements of the bedding are removed, while continuous action point clouds of human limbs are retained.

[0024] Micro-vibration filtering is based on the difference in vibration frequency between human limb exertion and bedding micro-movement. A human exertion vibration threshold of 5 Hz to 15 Hz and a bedding micro-movement filtering threshold of 20 Hz to 50 Hz are set. Combined with vibration energy threshold judgment, invalid vibration signals of bedding micro-movement are filtered out.

[0025] Dual-source collaborative filtering is implemented, and a dual-sensor noise verification mechanism is established. Invalid radar point cloud signals and invalid micro-vibration vibration signals are mutually verified. Dual filtering of bedding interference noise ensures that all data entering the feature extraction stage are valid human motion data.

[0026] Addressing the core deficiency of existing single-radar solutions that generate significant noise due to obstruction by bedding, this paper employs a triple logic approach: electromagnetic characteristic filtering at the radar end, frequency threshold filtering at the micro-vibration end, and dual-source mutual verification. This achieves a noise filtering rate of 99.5%, ensuring that all data entering subsequent processing is valid human motion data. This algorithmically improves the quality of basic data for motion recognition, laying the foundation for accurate identification.

[0027] As a further description of the above technical solution:

[0028] In the time-series linkage determination algorithm of the heterogeneous fusion feature set for stand-up, the heterogeneous fusion feature set for stand-up includes 3 radar spatial attitude features and 3 micro-vibration force feedback features;

[0029] The three radar spatial attitude characteristics are: shoulder elevation amplitude not less than 3 cm, hip displacement from the bed not less than 2 cm, and trunk tilt angle not less than 15 degrees.

[0030] The three micro-vibration force feedback characteristics are: the bed frame micro-vibration energy is not less than the preset threshold, the duration of bed support vibration is not less than 200 milliseconds, and the vibration frequency is matched with the human force exertion frequency band of 5 Hz to 15 Hz.

[0031] The temporal linkage judgment rule requires that the logic of 6 core features must be sequential and simultaneous, strictly matching the natural behavioral trajectory of the elderly getting up at night. The feature must last for a duration of not less than 300 milliseconds to be judged as a valid intention to get up. If the temporal logic is not met or the number of features is insufficient, it is directly judged as a non-getting-up action, and the subsequent judgment process is terminated.

[0032] Abandoning the judgment mode of fixed large movement features or single features in existing technologies, the system uses 6 core features and time-sequential logic to accurately match the elderly’s small and slow getting-up movements. This solves the problem that existing solutions easily misjudge turning over or stretching as getting up, making the accuracy of getting up intention recognition ≥99% and the misjudgment rate <1%, greatly improving the recognition accuracy.

[0033] As a further description of the above technical solution:

[0034] The edge-based lightweight individual adaptive temporal learning algorithm includes a self-learning phase. After the system starts, it automatically enters a 3-day adaptive learning cycle. The edge intelligent computing unit collects and stores the target elderly person's standing movement feature parameters locally, automatically fine-tunes the threshold range of 6 core features, adapts to the individual's differences in standing range, force intensity, and movement time, and automatically locks the optimal threshold after learning is completed.

[0035] During the manual fine-tuning stage, children, community grid workers, or caregivers can fine-tune the feature thresholds in one dimension through the mobile app of the service management backend unit. The fine-tuning accuracy is ±0.5 cm for shoulder elevation amplitude, ±1 Hz for vibration frequency, and ±50 ms for vibration duration.

[0036] It solves the shortcomings of existing technologies where fixed threshold algorithms cannot adapt to the personalized getting-up habits of the elderly, the frail, and those with hemiplegia. The 3-day self-learning cycle requires no manual intervention, and the manual fine-tuning function is adapted to special elderly groups. Moreover, all learning data is stored and processed locally, which not only improves the individual adaptability of the system, but also ensures privacy and security.

[0037] As a further description of the above technical solution:

[0038] In the multi-dimensional accurate prediction algorithm for fall precursors, the feature system of high-risk fall precursors includes premonitory postural imbalance, trunk tilt angle greater than 30 degrees, shoulder elevation with left-right imbalance difference greater than 2 cm, and sudden sinking of the spatial point cloud trajectory after the buttocks leave the bed. Any one of the three features must be satisfied and last for a duration of not less than 100 milliseconds.

[0039] The ultra-thin patch piezoelectric micro-vibration sensor detected a sudden drop in the vibration force of the support bed, with the vibration energy falling below 50% or more of the preset threshold.

[0040] Slow movement is a precursor to a complete standing up motion that takes more than 5 seconds;

[0041] If any one of the warning signs is met, it is determined to be a high-risk state for falling, and the early warning process is immediately initiated. When multiple warning signs are superimposed, it is directly upgraded to an emergency warning.

[0042] To overcome the lag limitation of existing technologies that can only detect falls after they have occurred, a warning feature system is constructed based on three dimensions: postural imbalance, insufficient force, and slow movement. This system enables early warnings with a response time of 800ms-1s, providing crucial time for elderly people to save themselves and receive remote assistance. The multi-dimensional feature overlay enhances the rules for early warning, further improving the accuracy of the warnings and the targeted nature of emergency responses.

[0043] As a further description of the above technical solution:

[0044] The tiered early warning logic includes low risk, where only a valid intention to get up is detected, there are no signs of imbalance, no local or remote alarms, and the time and duration of getting up are recorded only through the edge intelligent computing unit, and the recorded data is uploaded to the service management backend unit for storage;

[0045] Medium risk: Slight postural imbalance was detected, with no signs of a fall or scene risks. The soft night light was automatically kept on by the edge intelligent computing unit. There were no sound or vibration alarms or remote pushes. The risk status was only recorded and uploaded to the service management backend unit.

[0046] High risk: If a fall warning sign or overlapping scene risks are detected, the edge intelligent computing unit controls the soft night light to stay on and the light vibration reminder to be activated. At the same time, it pushes SMS, mini-program alarms and direct telephone alarms to the mobile devices of children, community grid workers and caregivers connected to the service management backend unit. The pushed information includes the fall risk level, bedroom location and warning time, and the warning record is uploaded to the service management backend unit for archiving in real time.

[0047] This solution addresses the issues of ineffective alerts and wasted nursing resources caused by the one-size-fits-all approach of existing technologies. It adopts a three-tiered differentiated early warning strategy (low, medium, and high), combining the elderly's health status with real-time scenario risks. Under the premise of ensuring elderly care safety, it rationally allocates nursing resources. Remote alarm and backend data archiving functions enable traceability of risk events and improve the management efficiency of smart elderly care.

[0048] As a further description of the above technical solution:

[0049] When the protective linkage action detects a valid intention to get up, the edge intelligent computing unit controls the magnetic handrail unlocking module to unlock the magnetic handrail. After getting up, it automatically locks and uploads the linkage record to the service management backend unit. It connects to the bedroom floor anti-slip mat micro-sensor module. When it detects the risk of a wet floor and the elderly person's intention to get up, the edge intelligent computing unit automatically raises the warning level by one level and uploads the scene risk information to the service management backend unit.

[0050] Overcoming the limitations of passive alarms in existing technologies, this system employs proactive protective measures such as magnetic handrail unlocking and enhanced scenario-based risk warnings to directly reduce the probability of falls among the elderly from an environmental perspective. The system also links to a backend recording function, completing the warning-protection loop and upgrading the system from risk alerts to proactive risk avoidance, significantly improving the safety and security of the elderly.

[0051] The second aspect of this application proposes a control method for a smart elderly care service management system, comprising the following steps:

[0052] S1: After the dual-sensor data acquisition unit is initialized and deployed, the system enters a low-power monitoring state. The 24 GHz narrow-beam millimeter-wave radar module scans at a frequency of 1 GHz, and the ultra-thin patch piezoelectric micro-vibration sensor is in a dormant state.

[0053] S2: The ultra-thin patch piezoelectric micro-vibration sensor is instantly awakened after detecting a vibration signal. At the same time, it triggers the 24 GHz narrow beam millimeter-wave radar module to switch to 10 GHz high-frequency scanning mode. The dual sensor data acquisition unit simultaneously acquires three-dimensional spatial attitude point cloud data and micro-vibration data and transmits them to the edge intelligent computing unit.

[0054] S3: The edge intelligent computing unit filters out bedding interference noise through a dual-source noise collaborative filtering algorithm to obtain effective human motion data;

[0055] S4: The edge intelligent computing unit uses a temporal linkage judgment algorithm based on the heterogeneous fusion feature set dedicated to getting up to extract features and make temporal judgments on the effective data of human body movements to determine whether it is a valid intention to get up.

[0056] If not, return to S1;

[0057] If so, execute S5;

[0058] S5: The edge intelligent computing unit uses a lightweight individual adaptive temporal learning algorithm at the edge to fine-tune the feature threshold to adapt to the individual differences of the target elderly and improve the accuracy of recognizing the intention to get up.

[0059] S6: The edge intelligent computing unit uses a multi-dimensional and accurate prediction algorithm for fall warning signs to monitor the warning features of data during the effective standing process and determine whether there is a high risk of falling. If not, it records the relevant standing data and uploads it to the service management backend unit, then returns to S1.

[0060] If so, execute S7;

[0061] S7: The edge intelligent computing unit combines the elderly’s preset health level with the real-time scene risk in the bedroom, generates an early warning signal through hierarchical early warning logic, controls the early warning and protection linkage unit to execute the corresponding early warning action, executes the scene-based protection linkage action, and uploads the early warning and linkage data to the service management backend unit.

[0062] S8: After the warning is lifted or the person gets up, the dual-sensor data acquisition unit resumes low-power monitoring and returns to S1.

[0063] The system clearly defines the complete workflow from low-power monitoring to data acquisition, processing, judgment, early warning, linkage, and return to low power consumption. It solves the problems of unclear system working logic and poor connection between links in the existing technology, which cause running lag and response delay. The low-power wake-up-sleep loop logic further reduces the power consumption of the system for 24 hours of operation (<1.2W), and takes into account both monitoring continuity and cost control.

[0064] Advantages of this invention:

[0065] The smart elderly care service management system and control method of this application feature a wearable-free design, breaking through usage barriers and making it suitable for elderly people of all ages.

[0066] By using dual-sensor heterogeneous fusion and precise algorithms, the bottleneck of single-sensor recognition is solved, resulting in high recognition accuracy.

[0067] By identifying and predicting early warning signs of falls, a closed-loop protection system is formed to enhance the safety value of elderly care.

[0068] Leveraging edge-integrated deployment and low-power design, it balances privacy and security with low-cost adoption;

[0069] The system has a well-developed architecture, is adapted to home environments, and has low operation and maintenance costs.

[0070] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0071] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0072] Figure 1 This is a system schematic diagram of a smart elderly care service management system and control method according to an embodiment of this application;

[0073] Figure 2 This is a flowchart of a smart elderly care service management system and control method according to an embodiment of this application. Detailed Implementation

[0074] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0075] The following is in conjunction with the appendix Figure 1 This application describes the smart elderly care service management system of Embodiment 1.

[0076] First, the hardware is installed: The millimeter-wave radar is wall-mounted above the left side of the elderly person's bed, at a height of 1.9m and a 38° angle. It is secured with a bracket, accurately covering the upper body and hip area, avoiding electromagnetic blind spots directly above the bedding. A micro-vibration sensor is installed using 3M adhesive, non-drilling patches, 10cm above the left leg of the bed frame. The sensor covers the core area of ​​bed vibration when the elderly person gets up, supporting themselves on the bed and lifting their hips off. The installation position has a tolerance of ±5cm. An edge computing unit is integrated inside the radar module housing and wired directly to the dual-sensor module via DuPont wires (transmission distance <1m, latency ≤8ms). It connects to the early warning and protection linkage unit via Bluetooth Low Energy (BLE). 5.0) Wireless connection (communication distance ≤ 5m); Deployment of linked devices: a soft night light is installed on the wall above the bed at a height of 1.5m, and a lightweight vibration reminder is fixed to the inside of the headboard; a magnetic armrest is installed on the right side of the headboard and connected to the unlocking module via wire; Overall deployment adaptation requires a minimum bedroom area of ​​6㎡, is compatible with a 1.5m single bed, requires no modification to the bed frame or additional wiring, is powered by 220V AC mains, and the entire deployment takes < 30 minutes.

[0077] Secondly, all algorithms in this embodiment are deployed on the STM32H743 edge computing unit and implemented using C language programming. The specific formulas and calculation steps of the core algorithms are as follows:

[0078] A dual-source noise collaborative filtering algorithm filters out invalid point clouds and micro-motion noise caused by occlusion from bedding, while retaining valid human motion data. This is implemented in three steps:

[0079] For radar-side electromagnetic reflectivity filtering, firstly, through calibration experiments, the 24GHz radar reflectivity threshold range for human skin / bones was determined [R, R] = [0.6, 0.9], and the reflectivity range for bedding was determined [R, R] = [0.1, 0.3]. A dynamic reflectivity filtering threshold R = 0.4 (taking the median reflectivity of the human body and bedding) was then set, and each point cloud data R acquired by the radar was filtered.

[0080] ;

[0081] Where R is the reflectance of the i-th point cloud, R' is the filtered reflectance, and point clouds with a reflectance of 0 are marked as invalid noise and are removed.

[0082] Radar-side timing continuity filtering

[0083] Calculate the displacement change ΔS of point clouds over three consecutive frames (i=1,2,...,n, where n is the number of point clouds), and set the temporal continuity threshold ΔS = 0.5cm (discrete displacement of bedding, continuous displacement of human body movements):

[0084] ;

[0085] If there are 2 or more frames with ΔS > ΔS in 3 consecutive frames, the point cloud is determined to be bedding micro-motion noise and is removed; otherwise, it is retained as a valid point cloud of human motion.

[0086] For micro-vibration frequency and energy filtering, a Fast Fourier Transform (FFT) is performed on the vibration signal s(t) acquired by the micro-vibration sensor to obtain the frequency spectrum F(f). The threshold range of human force vibration frequency is set as [f, f] = [5Hz, 15Hz], and the threshold range of bedding micro-movement frequency is set as [f, f] = [20Hz, 50Hz].

[0087] ;

[0088] Simultaneously calculate the energy E of the vibration signal, and set the energy threshold E = 0.1mJ (the vibration energy of a human body exerting force is higher than that of a slight movement of the bedding): ;

[0089] Where T is the signal acquisition time (taken as 200ms). If F(f) ∈ [5Hz, 15Hz] and E ≥ E, it is determined to be a human force vibration signal; otherwise, it is a bedding micro-motion noise and is discarded.

[0090] The dual-source collaborative filtering method determines human motion data as valid only when the radar retains a valid point cloud and the micro-vibration end retains a valid vibration signal, and then proceeds to the subsequent feature extraction stage. If either sensor determines the data as noise, the entire dataset is determined as invalid data, and subsequent processing is terminated.

[0091] Then, a time-series linkage judgment algorithm based on a heterogeneous fusion feature set specifically for standing up is used to extract core features:

[0092] The shoulder elevation amplitude H was measured, and the Z-axis height change of key shoulder points was extracted using radar point cloud data. The height difference between time t and the initial supine position was calculated. ;

[0093] Where Z is the initial shoulder height, and the threshold H is set to 3cm;

[0094] The micro-displacement D of the buttocks off the bed was used to extract the Z-axis height change of key points on the buttocks (the Z-axis height increased significantly upon leaving the bed), with a threshold value of D=2cm. ;

[0095] The torso tilt angle θ is calculated using the spatial coordinates of key points at the shoulders and waist, with a threshold of θ = 15°. ;

[0096] Where (X, Y, Z) are the coordinates of the key points of the waist at time t;

[0097] Bed frame micro-vibration energy E: set threshold E=0.15mJ.

[0098] The duration of the bed vibration is recorded as T: the duration for which the micro-vibration signal meets the effective conditions is recorded, with a threshold value of T=200ms.

[0099] Vibration frequency f: The result of FFT transformation must satisfy f∈[5Hz, 15Hz];

[0100] The timing-based judgment rule requires that all six core features meet the threshold requirements: H≥3cm, D≥2cm, θ≥15°, E≥0.15mJ, T≥200ms, and f∈[5Hz, 15Hz]. The feature timing-based progressive logic is as follows: effective micro-vibration signal (E, T, f satisfied) → shoulder elevation (H satisfied) → trunk tilt (θ satisfied) → buttocks off the bed (D satisfied), and the duration of the entire feature satisfaction process is T≥300ms, in order to be judged as a valid intention to get up. If any of the above conditions are not met, it is judged as a non-getting-up action such as turning over or stretching, and the subsequent process is terminated.

[0101] Next, a lightweight individual adaptive temporal learning algorithm is performed at the edge:

[0102] The system automatically adapts to the individualized getting-up habits of different elderly people. The specific steps are as follows:

[0103] During the self-learning phase (3-day cycle), effective standing-up movement characteristics of the target elderly were recorded daily, forming a sample set S = {S, S, ..., S}, where S = [H, D, θ, E, T, f] (k is the number of effective standing-ups per day). The mean μ and standard deviation σ of each feature were calculated for the 3-day sample set, and the mean - 0.5σ was used as the personalized feature threshold (adapting to individual differences in movement amplitude). ;

[0104] Where Th' is the personalized threshold of the i-th feature, μ is the 3-day mean of the i-th feature, and σ is the standard deviation. For example, if an elderly person's shoulder elevation mean μ = 4.5cm and standard deviation σ = 0.8cm, then the personalized threshold Th' = 4.5 - 0.5 × 0.8 = 4.1cm.

[0105] During the manual fine-tuning stage, children can input fine-tuning instructions via a mini-program to adjust the threshold of a single feature. The fine-tuning formula is as follows: ;

[0106] Where Th'' is the final threshold, Δ is the fine-tuning step size (H / D step size 0.5cm, θ step size 1°, E step size 0.02mJ, T step size 50ms, f step size 1Hz), and δ is the fine-tuning coefficient (user input +1 or -1, corresponding to the threshold being adjusted up or down).

[0107] Privacy is guaranteed; all sample data is stored in the Flash memory of the STM32H743. After learning is complete, the original samples are automatically deleted, and only the adjusted threshold parameters are retained. There is no cloud transmission.

[0108] Then, a multi-dimensional, precise prediction algorithm for fall precursors is implemented:

[0109] Three-dimensional precursor feature calculation, posture imbalance precursor:

[0110] The trunk tilt angle θ > 30°;

[0111] The difference in shoulder elevation imbalance ΔH = |HH| > 2cm (H / H is the left and right shoulder elevation amplitude);

[0112] The amount of sinking of the buttocks after leaving the bed is ΔZ = ZZ>1cm (Z is the maximum height of the buttocks after leaving the bed, Z is the height at time t).

[0113] If any of the above three conditions are met and the duration is ≥100ms, it is determined to be a precursor to attitude imbalance;

[0114] Signs of insufficient power:

[0115] Calculate the rate of change of micro-vibration energy ΔE / Δt. If ΔE / Δt < -0.5 mJ / s (sudden energy drop) and the current energy E < 0.5 × E (E is the personalized adjusted energy threshold), it is determined to be a precursor to insufficient force.

[0116] Early signs of slowed movement:

[0117] Record the duration T from the effective micro-vibration to the complete standing up process from the buttocks off the bed. If T > 5s, it is considered a precursor to bradykinesia.

[0118] The prediction rule is that if any one of the warning signs is met, it is judged as a high risk of falling and the early warning process is initiated. If two or more warning signs are met, it is upgraded to an emergency warning, and the alarm and linkage actions are strengthened.

[0119] Finally, a low-power cooperative scheduling algorithm is performed:

[0120] The master-slave sensor collaborative sleep-wake strategy is adopted to achieve low power consumption. When stationary, the millimeter-wave radar scans at a frequency of 1Hz (power consumption 0.1W), the micro-vibration sensor goes into sleep (power consumption <0.01W), and the total power consumption of the device is 0.25W.

[0121] When the wake-up is triggered, the micro-vibration sensor detects a vibration signal (amplitude > 0.1g), instantly wakes up and triggers the millimeter-wave radar to switch to 10Hz high-frequency scanning (power consumption 0.4W), with a total power consumption of 1.08W.

[0122] When the system returns to sleep mode, if no valid human motion data is detected for 30 consecutive seconds, the millimeter-wave radar will resume 1Hz scanning, and the micro-vibration sensor will go into sleep mode.

[0123] Example 2 involves operating and controlling the system deployed in Example 1, such as... Figure 2 As shown, the specific steps are as follows:

[0124] 1) After the system initialization is completed, it enters the low power monitoring state, the millimeter-wave radar scans at a frequency of 1Hz, and the micro-vibration sensor goes into sleep mode; the service management backend unit completes the preset of the health level of the elderly (such as setting the elderly and frail to high risk).

[0125] 2) When an elderly person turns over at night, the bed vibrates. The micro-vibration sensor detects a vibration signal with an amplitude of 0.15g, instantly waking the bed and triggering the millimeter-wave radar to switch to 10Hz high-frequency scanning. The two sensors simultaneously collect data and transmit it to the edge computing unit (transmission delay 6ms).

[0126] 3) The edge computing unit executes a dual-source noise collaborative filtering algorithm to filter out invalid point clouds and vibration noise generated by the slight movement of the bedding, and obtains effective human motion data (noise filtering rate 99.6%).

[0127] 4) Based on the time-series linkage judgment algorithm of the heterogeneous fusion feature set dedicated to getting up, 6 core features are extracted. The current action is determined to be turning over (buttocks off the bed feature is not met), the subsequent process is terminated, and the system returns to the low power monitoring state.

[0128] 5) After 30 minutes, the elderly man prepared to get up to use the toilet. The action of pushing himself up on the bed triggered the micro-vibration sensor to wake up. The two sensors collected data simultaneously. After noise filtering, the extracted 6 features all met the threshold requirements, and the timing was consistent with micro-vibration → shoulder lifting → trunk tilting → buttocks leaving the bed, lasting for 380ms. This was determined to be a valid intention to get up.

[0129] 6) The edge computing unit executes an individual adaptive learning algorithm, which fine-tunes the feature thresholds based on the learning data from the previous 3 days (e.g., the shoulder elevation threshold is adjusted from 3cm to 3.2cm) to improve recognition accuracy;

[0130] 7) Execute the multi-dimensional accurate prediction algorithm for fall warning signs. If a trunk tilt angle of 35° (>30°) is detected and lasts for 120ms, it is judged as a high risk of fall. At the same time, combined with the elderly person's high-risk health level, the high-risk warning process is initiated.

[0131] 8) Edge computing unit output commands: control the soft night light to stay on, start the light vibration reminder, push alarm information to the children and community grid members' mini-program, trigger direct dial alarm, and simultaneously control the magnetic handrail unlocking module to unlock for the elderly to grip and support;

[0132] 9) After the elderly person gets up, the system detects that the trajectory of the buttocks is stable after leaving the bed, determines that the warning is lifted, and controls the magnetic handrail to lock automatically; the dual sensor acquisition unit resumes low power monitoring state, and the edge computing unit uploads the record of this getting up and the warning record to the service management backend unit for archiving.

[0133] The above embodiment selected 10 elderly people living alone at home (including 3 frail elderly people and 2 hemiplegic elderly people) for a one-month test to verify the reliability and effectiveness of the system, as shown in the table below:

[0134] Test metrics Design Goals Test Results Accuracy of recognizing the intention to get up ≥99% 99.20% Misjudgment rate of intention to get up ＜1% 0.80% Fall warning time 800ms-1s Average 920ms Accuracy of fall warning ≥99% 99.30% Individual adaptive fit rate 100% 100% (All elderly individuals can be accurately identified without manual adjustments) Total power consumption ＜1.2W Average 1.05W Total power consumption during sleep mode ＜0.3W Average 0.22W Stable operation without failure for 24 hours support 30 consecutive days of trouble-free operation

[0135] The above embodiment achieves accurate recognition of elderly people living alone at night when they get up to use the toilet and provides early warning of fall signs. It also has advantages such as being wearable-free, having low false alarms, low power consumption, and strong individual adaptability, fully meeting the actual needs of home-based elderly care scenarios and can be directly promoted and applied in batches.

[0136] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0137] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0138] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A smart elderly care service management system, characterized in that, It includes a dual-sensor data acquisition unit, an edge intelligent computing unit, an early warning and protection linkage unit, and a service management backend unit; The dual-sensor data acquisition unit includes a main sensing module and an auxiliary sensing module. The main sensing module is a 24 GHz narrow beam millimeter-wave radar module, which is used to collect three-dimensional spatial attitude point cloud data of elderly people living alone at home during the process of getting up. The auxiliary sensing module is an ultra-thin patch piezoelectric micro-vibration sensor, which is used to collect micro-vibration force, vibration frequency and time sequence characteristic data of the bed during the process of elderly people getting up. The edge intelligent computing unit is an STM32H743 microcontroller. The edge intelligent computing unit is connected to the dual-sensor data acquisition unit by wires and is used to receive the three-dimensional spatial attitude point cloud data and micro-vibration data output by the dual-sensor data acquisition unit. The edge intelligent computing unit has built-in dual-source noise collaborative filtering algorithm, standing-specific heterogeneous fusion feature set temporal linkage judgment algorithm, edge lightweight individual adaptive temporal learning algorithm, fall precursor multi-dimensional accurate prediction algorithm and graded early warning logic. The early warning and protection linkage unit includes a soft night light, a lightweight vibration reminder, and a magnetic handrail unlocking module. The early warning and protection linkage unit is wirelessly connected to the edge intelligent computing unit in low power to receive the early warning signal and protection linkage signal output by the edge intelligent computing unit and execute the corresponding early warning action and protection linkage action. The service management backend unit is wirelessly connected to the edge intelligent computing unit to receive and store the elderly getting up records, risk warning records and equipment operation status data uploaded by the edge intelligent computing unit. It supports children, community grid workers and caregivers to access and view and perform manual fine-tuning operations through mobile app.

2. The intelligent elderly care service management system according to claim 1, characterized in that, The 24GHz narrow-beam millimeter-wave radar module has a ranging range of 0.1 meters to 5 meters, a detection angle of ±30 degrees, and a power consumption of no more than 0.5 watts. It supports switching between a 1Hz low-frequency scanning mode and a 10Hz high-frequency scanning mode. The ultra-thin patch-type piezoelectric micro-vibration sensor has a sampling frequency of 5Hz to 100Hz and a power consumption of no more than 0.1 watts.

3. The intelligent elderly care service management system according to claim 1, characterized in that, The 24 GHz narrow beam millimeter-wave radar module is wall-mounted above the head of the elderly person's bed at a height of 1.8 to 2.0 meters and a downward angle of 30 to 45 degrees. The detection area covers the elderly person's upper body and hip core area for getting up. The ultra-thin patch-type piezoelectric micro-vibration sensor is attached to the bed frame, legs, or under the bed board without drilling. The collection area covers the core area of ​​bed vibration when the elderly get up and support themselves on the bed, and when their buttocks leave the bed. The error tolerance for installation position deviation is no less than ±5 cm.

4. The intelligent elderly care service management system according to claim 1, characterized in that, The dual-source noise collaborative filtering algorithm includes radar-side filtering. Based on the electromagnetic reflectivity characteristics of 24 GHz radar, the reflectivity threshold difference between the human body and the bedding is calibrated, dynamic reflectivity filtering rules are set, and bedding point cloud noise with a reflectivity lower than that of the human body is filtered out. At the same time, based on the temporal continuity of the action, discrete point clouds of irregular micro-movements of the bedding are removed, while continuous action point clouds of human limbs are retained. Micro-vibration filtering is based on the difference in vibration frequency between human limb exertion and bedding micro-movement. A human exertion vibration threshold of 5 Hz to 15 Hz and a bedding micro-movement filtering threshold of 20 Hz to 50 Hz are set. Combined with vibration energy threshold judgment, invalid vibration signals of bedding micro-movement are filtered out. Dual-source collaborative filtering establishes a dual-sensor noise verification mechanism, cross-verifies invalid radar point cloud signals with invalid micro-vibration signals, and double-filters bedding interference noise to ensure that all data entering the feature extraction stage are valid human motion data.

5. The intelligent elderly care service management system according to claim 1, characterized in that, In the time-series linkage determination algorithm of the heterogeneous fusion feature set for stand-up, the heterogeneous fusion feature set for stand-up includes 3 radar spatial attitude features and 3 micro-vibration force feedback features; The three radar spatial attitude characteristics are: shoulder elevation amplitude not less than 3 cm, hip displacement from the bed not less than 2 cm, and trunk tilt angle not less than 15 degrees. The three micro-vibration force feedback characteristics are: the bed frame micro-vibration energy is not less than the preset threshold, the duration of bed support vibration is not less than 200 milliseconds, and the vibration frequency is matched with the human force exertion frequency band of 5 Hz to 15 Hz. The temporal linkage judgment rule requires that 6 core features must meet the logic of temporal progression and simultaneous validity, strictly match the natural behavioral trajectory of the elderly getting up at night, and the feature must be continuous for a duration of not less than 300 milliseconds to be judged as a valid intention to get up. If the temporal logic is not met or the number of features is insufficient, it is directly judged as a non-getting-up action, and the subsequent judgment process is terminated.

6. The intelligent elderly care service management system according to claim 1, characterized in that, The edge-based lightweight individual adaptive temporal learning algorithm includes a self-learning phase. After the system starts, it automatically enters a 3-day adaptive learning cycle. The edge intelligent computing unit collects and stores the target elderly person's standing movement feature parameters locally, automatically fine-tunes the threshold range of 6 core features, adapts to the individual's differences in standing range, force intensity, and movement time, and automatically locks the optimal threshold after learning is completed. During the manual fine-tuning stage, children, community grid workers, or caregivers can fine-tune the feature thresholds in one dimension through the mobile app of the service management backend unit. The fine-tuning accuracy is ±0.5 cm for shoulder elevation amplitude, ±1 Hz for vibration frequency, and ±50 ms for vibration duration.

7. The intelligent elderly care service management system according to claim 1, characterized in that, In the multi-dimensional accurate prediction algorithm for fall precursors, the feature system of high-risk fall precursors includes premonitory postural imbalance, trunk tilt angle greater than 30 degrees, shoulder elevation with left-right imbalance difference greater than 2 cm, and sudden sinking of the spatial point cloud trajectory after the buttocks leave the bed. Any one of the three features must be satisfied and last for a duration of not less than 100 milliseconds. The ultra-thin patch piezoelectric micro-vibration sensor detected a sudden drop in the vibration force of the support bed, with the vibration energy falling below 50% or more of the preset threshold. Slow movement is a precursor to a complete standing up motion that takes more than 5 seconds; If any one of the warning signs is met, it is determined to be a high-risk state for a fall, and the early warning process is immediately initiated. When multiple warning signs are superimposed, it is directly upgraded to an emergency warning.

8. The intelligent elderly care service management system according to claim 1, characterized in that, The tiered early warning logic includes low risk, where only a valid intention to get up is detected, there are no signs of imbalance, no local or remote alarms, and the time and duration of getting up are recorded only through the edge intelligent computing unit, and the recorded data is uploaded to the service management backend unit for storage; Medium risk: Slight postural imbalance was detected, with no signs of a fall or scene risks. The soft night light was automatically kept on by the edge intelligent computing unit. There were no sound or vibration alarms or remote pushes. The risk status was only recorded and uploaded to the service management backend unit. In cases of high risk, if signs of a fall are detected or the risk of a fall is compounded, the edge intelligent computing unit will control the soft night light to stay on and activate the light vibration reminder. At the same time, it will push SMS, mini-program alarms and direct telephone alarms to the mobile devices of children, community grid workers and caregivers connected to the service management backend unit. The pushed information includes the fall risk level, bedroom location and warning time, and the warning record will be uploaded to the service management backend unit for archiving in real time.

9. The intelligent elderly care service management system according to claim 1, characterized in that, When the protective linkage action detects a valid intention to get up, the edge intelligent computing unit controls the magnetic handrail unlocking module to unlock the magnetic handrail. After getting up, it automatically locks and uploads the linkage record to the service management backend unit. It connects to the bedroom floor anti-slip mat micro-sensor module. When it detects a risk of slippery floor and the elderly person's intention to get up, the edge intelligent computing unit automatically raises the warning level by one level and uploads the scene risk information to the service management backend unit.

10. The control method for the intelligent elderly care service management system according to any one of claims 1-9, characterized in that, Includes the following steps: S1: After the dual-sensor data acquisition unit is initialized and deployed, the system enters a low-power monitoring state. The 24 GHz narrow-beam millimeter-wave radar module scans at a frequency of 1 GHz, and the ultra-thin patch piezoelectric micro-vibration sensor is in a dormant state. S2: The ultra-thin patch piezoelectric micro-vibration sensor is instantly awakened after detecting a vibration signal. At the same time, it triggers the 24 GHz narrow beam millimeter-wave radar module to switch to 10 GHz high-frequency scanning mode. The dual sensor data acquisition unit simultaneously acquires three-dimensional spatial attitude point cloud data and micro-vibration data and transmits them to the edge intelligent computing unit. S3: The edge intelligent computing unit filters out bedding interference noise through a dual-source noise collaborative filtering algorithm to obtain effective human motion data; S4: The edge intelligent computing unit uses a temporal linkage judgment algorithm based on the heterogeneous fusion feature set dedicated to getting up to extract features and make temporal judgments on the effective data of human body movements to determine whether it is a valid intention to get up. If not, return to S1; If so, execute S5; S5: The edge intelligent computing unit uses a lightweight individual adaptive temporal learning algorithm at the edge to fine-tune the feature threshold to adapt to the individual differences of the target elderly and improve the accuracy of recognizing the intention to get up. S6: The edge intelligent computing unit uses a multi-dimensional and accurate prediction algorithm for fall warning signs to monitor the warning features of data during the effective standing process and determine whether there is a high risk of falling. If not, it records the relevant standing data and uploads it to the service management backend unit, then returns to S1. If so, execute S7; S7: The edge intelligent computing unit combines the elderly’s preset health level with the real-time scene risk in the bedroom, generates an early warning signal through hierarchical early warning logic, controls the early warning and protection linkage unit to execute the corresponding early warning action, executes the scene-based protection linkage action, and uploads the early warning and linkage data to the service management backend unit. S8: After the warning is lifted or the person gets up, the dual-sensor data acquisition unit resumes low-power monitoring and returns to S1.

Images