Intelligent nursing device for old-age service

Abstract

The application discloses a kind of wisdom old-age care watching devices dedicated to old-age service, including non-contact behavior perception module, bed body vital sign perception module, voice interaction evaluation module, environmental state acquisition module, indoor positioning module, event anchoring fusion module, digital twin state update module, service scheduling module and privacy security module.Event anchoring fusion module determines the time when the abnormal occurs as event anchor point when producing initial event, intercepts the point cloud behavior characteristics, bed body pressure characteristics, voice response characteristics, environmental characteristics and positioning characteristics in the time window before and after, carries out time alignment, conflict verification and confidence fusion, generates care risk grade.Digital twin state update module updates state label accordingly, and service scheduling module outputs disposal strategy.The application can review and judge fall, night bed leaving, sleep abnormality and cognitive response decline trend without relying on camera and forced wearing equipment.

Description

Technical Field

[0001] This invention relates to the fields of smart elderly care, Internet of Things (IoT) care, and health monitoring, specifically to a smart elderly care device for elderly care services. Background Technology

[0002] With the increasing demand for home-based, community-based, and institutional elderly care, daily safety monitoring, nighttime bed-leaving monitoring, fall risk identification, and health status tracking are gradually becoming important components of elderly care services. Existing elderly care equipment typically includes wearable alarm devices, camera monitoring devices, mattress monitoring devices, voice call devices, and elderly care service platforms.

[0003] Wearable devices can collect heart rate, activity status, or location information, but their effectiveness depends on continuous wear by the elderly. For very old, disabled, or cognitively impaired seniors, there may be instances of forgetting to wear them, charging interruptions, removal during bathing, or discomfort, leading to interruptions in monitoring data. While camera monitoring devices can provide direct observation of the elderly's condition, privacy concerns arise in private settings such as bedrooms and bathrooms, hindering long-term adoption.

[0004] Millimeter-wave radar, bed pressure sensors, and voice interaction devices can reduce dependence on wearing them and privacy risks to some extent. However, when using millimeter-wave radar alone for fall detection, it is easy to misjudge actions such as sitting down quickly, lying down voluntarily, or bending over to pick up objects as falls; when using bed sensors alone, they can only cover the bed status and it is difficult to determine the specific situation after the elderly person gets out of bed; when using voice interaction data alone, it is easily affected by noise, the elderly person's condition that day, dialect, or response habits.

[0005] Existing smart elderly care service platforms primarily focus on information aggregation, alarm notifications, and service dispatch. They typically only upload data collected from different devices to the platform for display or to trigger alarms, lacking a processing mechanism for multi-source feature verification around the moment the same abnormal event occurs. Therefore, current technology still requires an elderly care device that does not rely on cameras, does not require mandatory device wearing, and can collaboratively assess multi-source data to reduce false alarms, minimize privacy risks, and improve the targeted nature of elderly care service responses. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a smart elderly care device for elderly care services, which can review and judge the care risks such as falls, getting out of bed at night, abnormal sleep and cognitive decline without relying on cameras or forcing the elderly to wear the device, and output corresponding care and treatment strategies based on the judgment results.

[0007] The above-mentioned technical objective of the present invention is achieved through the following technical solution: a smart elderly care device for elderly care services, comprising a non-contact behavior perception module, a bed vital signs perception module, a voice interaction assessment module, an environmental status acquisition module, an indoor positioning module, an event anchoring fusion module, a digital twin status update module, a service scheduling module, and a privacy and security module. The non-contact behavior perception module collects millimeter-wave radar point cloud data in the space where the elderly person is located, and extracts the human body height center, posture category, movement speed and stopping position from the millimeter-wave radar point cloud data. The bed vital signs sensing module collects bed pressure time-series signals and extracts in-bed status, out-of-bed status, turning frequency, respiratory rate and heart rate from the bed pressure time-series signals. The voice interaction evaluation module extracts response delay, pause ratio, speech rate fluctuation and sound intensity change when the elderly respond to voice. The environmental status acquisition module collects data on light, temperature, humidity, noise, and air quality in the space where the elderly person is located. The indoor positioning module determines the functional area where the elderly person is located and outputs the location information of the bedside, bathroom, living room or doorway; When an abnormal initial event occurs in any of the following modules: the non-contact behavior perception module, the bed vital signs perception module, the voice interaction assessment module, the environmental status acquisition module, or the indoor positioning module, the event anchoring fusion module determines the time of occurrence of the abnormal initial event as the event anchor point. It then extracts point cloud features, bed pressure features, voice features, environmental features, and positioning features within a preset time window before and after the event anchor point. The module performs time alignment, conflict verification, and confidence fusion on the point cloud features, bed pressure features, voice features, environmental features, and positioning features to generate a care risk level. The digital twin status update module updates the elderly person's status tag according to the care risk level; The service scheduling module outputs care treatment strategies based on the status label and the care risk level. The privacy and security module performs local preprocessing, feature-based storage, encrypted transmission, and access control for the data collection, processing, transmission, storage, and retrieval processes.

[0008] Furthermore, the event anchoring fusion module includes an abnormal initial event triggering unit, an event anchor point determination unit, a time window interception unit, a feature time alignment unit, a conflict data verification unit, a confidence correction unit, and a risk level generation unit; The abnormal initial event triggering unit receives abnormal initial events from the non-contact behavior perception module, the bed vital signs perception module, the voice interaction evaluation module, the environmental status acquisition module, or the indoor positioning module. The event anchor point determination unit determines the occurrence time of the abnormal initial event as the event anchor point; The time window extraction unit extracts multi-source features within the forward and backward time windows based on the event anchor point. The feature time alignment unit aligns point cloud features, bed pressure features, voice features, environmental features, and positioning features with different sampling frequencies to the same time axis according to timestamp mapping, sliding window statistics, or resampling methods. The conflict data verification unit determines whether there is a mutually supportive or mutually conflicting relationship between features from different sources. The confidence correction unit corrects the confidence of the corresponding feature based on the mutually supporting or conflicting relationships. The risk level generation unit generates a care risk level based on the revised confidence level.

[0009] Furthermore, the abnormal initial event includes at least one of the following events: The body's center of gravity descends rapidly within a preset time. The human body posture changes from standing or walking to a low posture; The human body remains in a low position continuously near the bed, in the bathroom, or in the corridor area; The bed pressure timing signal indicates that the person left the bed at night and did not return to a stable bed state within a preset time. The respiratory rate or heart rate exceeds the preset range; The voice interaction evaluation module failed to obtain a valid response within a preset number of voice confirmation attempts; The elderly person's range of daytime activities decreased over several consecutive assessment periods; Specifically, for the initial abnormal event of a fall, the forward time window is 5 to 30 seconds, and the backward time window is 10 to 60 seconds.

[0010] Furthermore, when the event anchoring fusion module reviews a fall event, it first monitors the duration of the low-position stay after the non-contact behavior perception module detects a rapid descent of the human body's height center and determines whether an initial low-position stay has been formed. If an initial low-level pause does not occur, output "Cancel alarm". If an initial low-level stay is formed, the bed pressure characteristics output by the bed vital signs sensing module are read, and it is determined whether a stable bed pressure distribution has been formed. If a stable pressure distribution is formed when getting into bed, then based on the indoor positioning area verification results output by the indoor positioning module, the system will output whether to actively get into bed or switch between sitting and lying down. If a stable pressure distribution is not established, the voice response result output by the voice interaction evaluation module is read, and it is determined whether there is a valid response. If no valid response is found, the system will combine the indoor location area verification results to output a high risk of fall when the elderly person is located at the bedside, in the bathroom, or in the corridor area, and output a delayed verification when the elderly person is not located at the bedside, in the bathroom, or in the corridor area. If a valid response is received, the system continues to monitor changes in posture after the response and outputs information such as sitting-lying transition, bending over, or delayed verification based on whether the user continues to remain in a low position after the response and the results of the indoor positioning area verification.

[0011] Furthermore, the stable bed pressure distribution is determined based on the total pressure change, pressure distribution duration, and pressure distribution stability in the bed pressure time-series signal; When, within the backward time window after the body's height center rapidly descends, a pressure distribution with a duration reaching a preset duration appears in the bed pressure timing signal, and the pressure distribution matches the pressure distribution corresponding to the bed state, the event anchoring fusion module reduces the fall confidence. When, within the backward time window after the body's height center rapidly descends, the bed pressure timing signal does not show a stable bed pressure distribution corresponding to the in-bed state, and the voice interaction evaluation module does not receive an effective response, the event anchoring fusion module increases the fall confidence.

[0012] Furthermore, when the event anchoring fusion module verifies the nighttime bed-leaving event, it uses the moment when the bed vital signs sensing module detects the nighttime bed-leaving as the event anchor point, and extracts the indoor positioning features, environmental features, and bed pressure features within the backward time window after the event anchor point. When the indoor positioning features show that the elderly person enters the bathroom and returns to the bed within a preset time, and the bed pressure features return to a stable state in the bed, the event anchoring fusion module records the nighttime bed-leaving event as a normal nighttime bed-leaving event. When indoor positioning features indicate that an elderly person has stayed in the bathroom, bedside, or corridor area for more than a preset time after leaving the bed, and the bed pressure features have not returned to a stable state in bed, the event anchoring fusion module increases the risk level of abnormal nighttime bed leaving. When environmental characteristics indicate that the light intensity is below a preset threshold, the service scheduling module outputs a light-on reminder or a nighttime safety reminder.

[0013] Furthermore, when the digital twin state update module judges the declining trend of cognitive response, it acquires voice interaction features, sleep state features, nighttime bed-leaving features, and daytime activity features over multiple consecutive evaluation periods. The voice interaction features include response latency, pause ratio, and speech rate fluctuation; The sleep state characteristics include sleep continuity or the number of sleep interruptions; The nighttime bed-leaving characteristics include the number of nighttime bed-leavings or the duration of bed-leavings; The characteristics of daytime activity include the range or duration of daytime activity; When, within multiple consecutive assessment cycles, there is an increase in response delay, an increase in the proportion of pauses, or an increase in speech rate fluctuations, along with a decrease in sleep continuity, an increase in the number of times the elderly person gets out of bed at night, and a reduction in the range of daytime activities, the digital twin status update module updates the elderly person's status label to a cognitive response decline trend label and sends the cognitive response decline trend label to the service scheduling module.

[0014] Furthermore, the digital twin status update module includes a basic profile unit, a daily routine model unit, a sleep state model unit, a behavioral activity model unit, an interaction response model unit, and a risk label update unit; The basic record unit records the elderly person's age, mobility level, whether they live alone, and care level; The daily routine model unit records sleep time, wake-up time, nighttime bed-leaving habits, and daytime activity area; The sleep state model unit updates sleep continuity, number of times the patient gets out of bed at night, frequency of turning over, and number of abnormal breathing events based on bed pressure characteristics; The behavioral activity model unit updates the daytime activity range, activity duration, number of low-level stays, and number of abnormal stationary times based on point cloud features and positioning features; The interactive response model unit updates the response delay, pause ratio, speech completion rate, and no-response frequency based on speech features; The risk label updating unit updates at least one status label based on the care risk level, including high risk of nighttime falls, high risk of bedside falls, abnormal risk of staying in the bathroom, abnormal risk of sleep breathing, frequent risk of getting out of bed at night, and a trend of declining cognitive response or a shrinking range of activity.

[0015] Furthermore, the privacy and security module includes a local preprocessing unit, a feature storage unit, an encrypted communication unit, an access control unit, and a log tracking unit; The local preprocessing unit extracts point cloud features, bed pressure features, and voice features locally from millimeter-wave radar point cloud data, bed pressure time-series signals, and voice response data. The feature storage unit stores point cloud features, attitude category, bed pressure features, response delay, pause ratio, speech rate fluctuation, sound intensity change, risk level and status label, but does not store video images, complete voice content or complete original pressure waveform for a long time. The encrypted communication unit encrypts and transmits data uploaded to the service scheduling module, nursing station terminal, or family member terminal. The access control unit displays data at different granularities according to the access permissions of family members, caregivers, and institutional administrators; The log tracking unit records the viewing of abnormal events, the output of handling strategies, and the manual confirmation operations.

[0016] Furthermore, the service scheduling module outputs a tiered care management strategy based on the level of care risk. When the care risk level is low, the service scheduling module records the event and outputs a strategy to continue observation. When the care risk level is medium risk, the service scheduling module outputs voice reminders, secondary confirmations, or sends attention reminders to family members through the user interaction module; When the care risk level is high, the service scheduling module sends a manual verification task to the caregiver terminal or the nursing station terminal; When the care risk level is emergency risk, the service dispatch module outputs emergency response strategies to the family member terminal, caregiver terminal or emergency call system; The non-contact behavior perception module, bed vital signs perception module, voice interaction assessment module, environmental status acquisition module, indoor positioning module, event anchoring fusion module, digital twin status update module, service scheduling module, and privacy and security module are integrated into the same device, or distributed in elderly care rooms, beds, bedside terminals, edge computing terminals, nursing station terminals, or cloud servers, and connected through wired or wireless communication.

[0017] Compared with the prior art, the present invention has at least the following beneficial effects: 1. This invention collects multi-source features through a non-contact behavior sensing module, a bed vital signs sensing module, a voice interaction assessment module, an environmental status acquisition module, and an indoor positioning module. The event anchoring fusion module extracts data within a time window before and after the occurrence of the abnormal initial event, enabling data from different sources to be reviewed and judged around the same event, avoiding false alarms or missed judgments caused by independent alarms from each module.

[0018] 2. This invention enables the device to distinguish between different situations such as high-risk falls, voluntary bed-getting, sitting-to-lying transitions, bending-over events, and normal nighttime bed-getting by performing time alignment, conflict verification, and confidence fusion on point cloud behavioral features, bed pressure features, voice response features, environmental features, and positioning features. This reduces false alarms caused by a single millimeter-wave radar or a single bed sensor.

[0019] 3. This invention verifies the changes in the body's height center after a rapid descent by examining the bed's pressure characteristics, and combines this with voice response results and indoor positioning area assessment to determine the risk of falls. This makes fall alarms no longer solely dependent on a single posture change, thus improving the rationality of fall event assessment.

[0020] 4. This invention uses a digital twin status update module to combine voice interaction features, sleep status features, nighttime bed-leaving features, and daytime activity features over multiple consecutive assessment periods to update status labels for cognitive response decline trends, avoiding the need to judge the elderly's status based on only one voice response or a single voice feature.

[0021] 5. This invention extracts and saves feature data locally through a privacy and security module, without storing video images, complete voice content, or complete original pressure waveforms for a long time. The service scheduling module outputs observation records, voice reminders, manual verification, or emergency response strategies based on the care risk level and status label. This reduces privacy risks while enabling the risk identification results to be integrated with the elderly care service handling process. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the modular structure of a smart elderly care device specifically designed for elderly care services according to the present invention. Figure 2 This is a flowchart of the event anchoring fusion processing in this invention; Figure 3 This is a logic diagram for verifying fall events in this invention; Figure 4 This is a logic diagram for judging the decline trend of cognitive response in this invention. Detailed Implementation

[0023] The following is combined with Figures 1 to 4 The specific embodiments of the present invention will be described below.

[0024] In one specific implementation, the smart elderly care device includes a non-contact behavior sensing module, a bed-based vital signs sensing module, a voice interaction assessment module, an environmental status acquisition module, an indoor positioning module, an event anchoring and fusion module, a digital twin status update module, a service scheduling module, and a privacy and security module. The non-contact behavior sensing module can employ a millimeter-wave radar sensor. The millimeter-wave radar sensor can be installed above the bedroom wall, diagonally above the foot of the bed, above the bathroom door, or on the ceiling of the elderly care facility room. The installation location should ideally cover the bedside, room corridors, bathroom entrances, and the elderly person's daily activity areas. The millimeter-wave radar sensor does not output video images, but instead outputs information such as reflective point clouds, human body height center, movement speed, posture category, and dwelling position. Here, the human body height center can be understood as the center of height change representing the main body's position in the point cloud. Posture categories can include standing, sitting, lying down, bending over, and low-position dwelling. Movement speed can include horizontal movement speed and vertical height change speed. Dwelling position can cooperate with the indoor positioning module to determine whether the elderly person is in an area such as the bedside, bathroom, living room, doorway, or corridor.

[0025] The bed vital signs sensing module can employ pressure film sensors, piezoelectric film sensors, bed foot weighing sensors, or micro-vibration sensors located under the mattress or above the bed frame. This module does not require altering the main structure of the bed; it only needs to acquire the bed pressure timing signal. After the elderly person gets into bed, the bed pressure timing signal shows a stable pressure distribution. After the elderly person gets out of bed, the total pressure decreases, and the pressure distribution disappears or weakens significantly. When the elderly person turns over, the pressure distribution shows short-term changes. Breathing movements cause low-frequency periodic fluctuations in the bed pressure signal, and heartbeats cause weak periodic changes in the micro-vibration signal. The bed vital signs sensing module can then extract information on in-bed status, out-of-bed status, turning frequency, respiratory rate, heart rate, and sleep continuity. Sleep continuity can be derived by combining data from continuous nighttime in-bed time, number of times the elderly person gets out of bed, number of times they turn over, and number of interruptions in body movement.

[0026] The voice interaction assessment module can be installed in bedside terminals, smart speakers, nursing call terminals, or mobile terminals. This module primarily collects features during the elderly person's response process, without requiring the complete audio recording to be saved. In use, the user interaction module can issue voice prompts for morning greetings, medication reminders, nighttime bed-leaning reminders, or suspected fall checks. For example, the system can prompt "Are you safe now?", "Do you need help?", or "Have you taken your medication?". After the elderly person responds, the voice interaction assessment module extracts features such as response delay, pause ratio, speech rate fluctuation, volume variation, number of no-response instances, and interaction completion rate. Response delay is the time between the end of the voice prompt and the elderly person's response. Pause ratio is the ratio of pause duration to total response time. Speech rate fluctuation can be obtained based on the number of syllables per unit time or changes in the rhythm of speech segments. Volume variation reflects whether the elderly person's voice weakens significantly. This module only outputs feature values, reducing privacy concerns associated with long-term storage of complete audio recordings.

[0027] The environmental status acquisition module may include one or more of the following: temperature and humidity sensors, light sensors, noise sensors, air quality sensors, smoke sensors, gas sensors, and door magnetic sensors. Light characteristics can be used to determine if a low-light environment exists when someone gets out of bed at night. Noise characteristics can be used to determine if a lack of voice response may be due to environmental noise interference. Temperature, humidity, and air quality can be used to help determine if nighttime discomfort, sleep interruptions, or abnormal breathing in the elderly are related to the environment. Door magnetic characteristics can help determine if the elderly person is heading towards the door or if there is a risk of them leaving home. The data collected by the environmental status acquisition module is not used as the sole result of health assessment, but rather to adjust the confidence level in event assessments.

[0028] The indoor positioning module is used to determine the functional area where the elderly person is located. This module can employ one or more methods, including millimeter-wave radar area recognition, Bluetooth Low Energy beacons, Wi-Fi fingerprinting, UWB positioning, door sensor status, and bed presence status. In typical home-based elderly care scenarios, high-precision positioning devices may not be necessary. Instead, the bedroom, bedside, bathroom, living room, doorway, and corridor can be divided into several functional areas, and the elderly person's location can be determined by combining millimeter-wave radar detection of the area, bed pressure status, and door sensor status. In nursing home scenarios, a caregiver's terminal positioning can also be added for subsequent service scheduling.

[0029] The event anchoring fusion module is the processing module used in this embodiment for multi-source verification and judgment. Figure 2 The event anchoring fusion processing flow is illustrated. When any sensing module generates an abnormal initial event, the event anchoring fusion module does not immediately output a final alarm based on the result of a single module. Instead, it first determines the time of occurrence of the abnormal initial event and uses this time as the event anchor point. Then, the event anchoring fusion module uses the event anchor point as a reference to extract multi-source features within the forward and backward time windows. The forward time window is used to observe whether the elderly person accelerated their movement, changed their posture, or left the bed before the abnormality occurred. The backward time window is used to observe whether the elderly person remained in a low position continuously after the abnormality occurred, whether a stable pressure distribution on the bed was formed, whether there was an effective voice response, and whether they entered a high-risk area.

[0030] When dealing with fall-related anomalies, the forward time window can be set to 5 to 30 seconds, and the backward time window can be set to 10 to 60 seconds. This range effectively covers the main movement processes before and after the fall. For nighttime bed-getting events, the backward time window can be set to several minutes to tens of minutes to determine whether the elderly person has returned to bed normally. For assessing the trend of declining cognitive function, the assessment cycle can be set daily, and multiple consecutive assessment cycles can be 7, 14, or 30 consecutive days, which can be adjusted according to the elderly person's level of care.

[0031] After acquiring multi-source features, the event anchoring fusion module first performs time alignment. Different modules use different data sampling methods. Millimeter-wave radar point cloud data is typically output frame by frame, bed pressure signals are continuous time-series signals, voice responses are event-based data, environmental data can be collected at fixed intervals, and positioning data can be triggered by regional changes. To enable comparison of these data around the same event, the feature time alignment unit can map various data types to the same time axis using timestamp mapping, sliding window statistics, or resampling methods. For example, the period from 10 seconds before the event anchor point to 30 seconds after the event anchor point can be divided into several small windows, and within each small window, changes in the human body height center, bed pressure, voice response results, and positioning area changes can be statistically analyzed.

[0032] After time alignment is completed, the event anchoring and fusion module performs conflict verification. Conflict verification is not simply about determining whether an alarm has been triggered, but rather whether features from different sources are mutually supportive or contradictory. For example, if millimeter-wave radar shows a rapid descent of the body's center of height, but the bed pressure characteristics show a stable pressure distribution upon getting into bed within a short period, then the event is more likely the elderly person actively getting into bed or transitioning from sitting to lying down, rather than a fall. Similarly, if the voice interaction assessment module does not receive a valid response, but the environmental status acquisition module shows high noise levels, the confidence level of the lack of voice response can be reduced. Furthermore, if bed pressure indicates getting out of bed, but millimeter-wave radar shows the elderly person remaining at a low position near the bed, further investigation is needed to determine whether they were sitting on the bedside, fell at the bedside, or bent over to tidy up items.

[0033] The confidence correction unit adjusts the confidence level of each feature based on the conflict verification results. A risk score from 0 to 100 can be set, or a level system of low, medium, high, and emergency risks can be used. If multiple features support each other, the confidence level of the corresponding event is increased. If a contradictory feature that can explain the anomaly appears, the confidence level of the corresponding event is decreased. For example, if the body's center of height rapidly descends, the body remains in a low position, a stable pressure distribution on the bed is not established, there is no effective voice response, and the location features indicate the location is in the bathroom or bedside area, the fall risk level can be increased. If, after the body's center of height descends, the bed pressure returns to a stable state, and the elderly person provides an effective response, the fall risk level can be decreased.

[0034] Figure 3The logic for verifying a fall event is illustrated. In a specific scenario, an elderly person gets up from the bedside at night, and millimeter-wave radar detects a rapid descent of the body's center of height. The non-contact behavior perception module sends this event as a suspected fall-related anomaly to the event anchoring and fusion module. The event anchoring and fusion module first monitors the duration of the low-position stay. If the body's center of height only drops briefly and then quickly recovers, it indicates that the elderly person may have simply bent down to pick something up or sat down briefly, and the system outputs "cancel alarm" or "low-risk record." If an initial low-position stay is formed, the bed pressure characteristics are read. If a stable bed-up pressure distribution is formed within the backward time window, it indicates that the elderly person may have actively gone to bed, and the system, combined with location characteristics, outputs "actively went to bed" or "sitting / lying down transition." If a stable bed-up pressure distribution is not formed, the voice response result is read. The system can issue one or more voice confirmations through the user interaction module, such as "Do you need help?" If no effective response is received, and the location characteristics show that the elderly person is located at the bedside, in the bathroom, or in the corridor area, the event anchoring and fusion module outputs "high-risk fall." If no effective response is received but the location is not in a high-risk area, a delayed verification or manual confirmation is output. If a valid response is received, continue to observe the posture changes after the response. If the system continues to remain in a low position after the response, it can output a delayed verification; if the system no longer remains in a low position after the response, it can output a sitting-lying transition or bending event in conjunction with the indoor positioning area.

[0035] Stable pressure distribution upon getting into bed can be determined by changes in total pressure, duration of pressure distribution, and stability of pressure distribution. Specifically, after an elderly person gets into bed, the bed pressure signal will show a significant increase in total pressure, which will remain within a relatively stable range for a certain period of time. The location of the pressure distribution corresponds to the area where the elderly person is lying down, and the duration reaches a preset time. This preset time can be 5 to 30 seconds. If this pressure distribution occurs after the center of body height has descended, it indicates that the elderly person is more likely to have assumed a lying position in bed. If the bed pressure does not recover after the center of body height has descended, and the elderly person does not respond effectively, it indicates that the elderly person may not be in bed, and the risk of fall needs to be increased.

[0036] In nighttime bed-leaving events, the bed vital signs sensing module generates an initial abnormal nighttime bed-leaving event when it detects the disappearance of bed pressure distribution or a significant drop in total pressure. The event anchoring and fusion module uses the time of bed-leaving as the event anchor point, extracting location features, environmental features, and bed pressure features within a specified time window. If the location features show that the elderly person enters the bathroom and returns to the bed within a preset time, and the bed pressure stabilizes, the system records it as a normal nighttime bed-leaving event and does not send a high-risk alarm to family members or caregivers. If the elderly person stays in the bathroom, bedside, or corridor area for an extended period after leaving the bed, and the bed pressure does not stabilize, the risk level of the abnormal nighttime bed-leaving event is increased. If the illumination features show low nighttime illuminance, the service scheduling module can first issue a light-on reminder or safety reminder through the user interaction module. If both low-position lingering and lack of voice response occur simultaneously, the system can proceed to the fall event review logic.

[0037] For sleep abnormalities, the bed-based vital signs sensing module can continuously extract respiratory rate, heart rate, frequency of turning over, and sleep continuity throughout the night. When the respiratory rate continuously exceeds the preset range, or when sleep is repeatedly interrupted by getting out of bed, turning over, or body movement during the night, the event anchoring and fusion module can record such events as sleep abnormalities. If it is only a short-term abnormality accompanied by turning over or short-term body movement, the risk level can be reduced. If it occurs multiple times in the same night and repeats for several consecutive nights, the digital twin status update module can use it as sleep status trend data for subsequent status label updates.

[0038] Figure 4 The logic for judging the decline in cognitive response is illustrated. This logic does not directly determine the elderly person's cognitive state based on a single voice response, but rather jointly summarizes voice interaction characteristics, sleep state characteristics, nighttime bed-leaving characteristics, and daytime activity characteristics over multiple consecutive assessment periods. Voice interaction characteristics include response delay, pause ratio, and speech rate fluctuations. Sleep state characteristics include sleep continuity or the number of sleep interruptions. Nighttime bed-leaving characteristics include the number of times the elderly person gets out of bed at night or the duration of bed-leaving. Daytime activity characteristics include the range or duration of daytime activities. The system can issue simple greetings or reminders at fixed times each day and record the elderly person's responses. For example, the system can issue a greeting like "How are you feeling today?" in the morning, or a confirmation prompt at medication time. The voice interaction assessment module only extracts response characteristics and does not permanently store complete voice content.

[0039] Over multiple consecutive assessment periods, if the voice interaction characteristics exhibit at least one of the following: increased response delay, increased pause rate, or increased speech rate fluctuation, and simultaneously meet the criteria of decreased sleep continuity, increased nighttime outings, and reduced daytime activity range, the digital twin status update module will update the elderly person's status label to a cognitive response decline trend label. This label is used for follow-up reminders and care suggestions in elderly care services and is not considered a medical diagnosis. If the above combined conditions are not met, the original status label will be maintained and observation will continue. This can reduce misjudgments caused by the elderly person's daily fatigue, noise interference, emotional changes, or occasional lack of response.

[0040] The digital twin status update module can include a basic profile unit, a daily routine model unit, a sleep state model unit, a behavioral activity model unit, an interaction response model unit, and a risk label update unit. The basic profile unit records the elderly person's age, mobility level, whether they live alone, and care level. The daily routine model unit records the elderly person's usual bedtime, wake-up time, nighttime bed-leaving habits, and daytime activity area. The sleep state model unit updates sleep continuity, nighttime bed-leaving frequency, turning frequency, and number of abnormal breathing events based on bed pressure characteristics. The behavioral activity model unit updates daytime activity range, activity duration, number of low-position stays, and number of abnormal stillnesses based on point cloud and location features. The interaction response model unit updates response delay, pause ratio, speech completion rate, and frequency of no response based on voice features. The risk label update unit updates status labels such as high risk of nighttime falls, high risk of bedside falls, abnormal risk of bathroom stays, risk of sleep breathing abnormalities, risk of frequent nighttime bed-leavings, and trends of declining cognitive response or shrinking activity range based on care risk levels and multi-period trend data.

[0041] The service dispatch module outputs care management strategies based on the care risk level and status label. At low risk, the system can simply record the event and continue observation. At medium risk, the user interaction module can output voice reminders, secondary confirmations, or send attention reminders to family members. At high risk, a manual verification task can be sent to the caregiver's terminal or the care station terminal. In case of emergency risk, an emergency management strategy can be output to the family member's terminal, caregiver's terminal, or the emergency call system. In elderly care institutions, the service dispatch module can also combine the caregiver's location and current task status to send verification tasks to the nearest caregiver's terminal. After processing, the caregiver can confirm the result on their terminal, and this result is then sent as manual feedback to the digital twin status update module.

[0042] The privacy and security module is integrated throughout the data acquisition, processing, transmission, storage, and retrieval process. Millimeter-wave radar does not acquire video images; the system saves point cloud features, posture categories, and risk assessment results. The voice interaction assessment module can perform feature extraction locally, saving features such as response latency, pause ratio, speech rate fluctuations, and volume changes, but does not permanently save complete voice content. The bedside vital signs sensing module can save bedside pressure characteristics, respiratory rate, heart rate cycle, and bed exit status, but does not permanently save complete raw pressure waveforms. Data uploaded to the service scheduling module, nursing station terminal, or family member terminal can be transmitted encrypted. Different personnel can view data at different granularities. For example, family members can view risk alerts and handling suggestions, caregivers can view rooms and event types requiring verification, and institutional management can view statistics and treatment records. Abnormal event viewing, treatment strategy output, and manual confirmation operations can be logged for easy subsequent traceability.

[0043] In home-based elderly care scenarios, a millimeter-wave radar sensor can be installed in the bedroom, a pressure film sensor can be placed under the mattress, a voice interaction terminal can be set up at the bedside, and light, temperature, humidity, and noise sensors can be deployed in the bedroom and bathroom areas. An edge computing terminal can be placed near the bedside table or the electrical control box to run the event anchoring fusion module, the digital twin state update module, and the privacy and security module. Normally, the system only records feature data and status tags. If the elderly person is suspected of falling, fails to return to bed at night, or is unresponsive to voice commands, the system will then verify the multi-source features in the event window according to the event anchoring fusion process and send a corresponding reminder to the family member.

[0044] In institutional elderly care settings, each room can be equipped with a non-contact behavior sensing module, a bed vital signs sensing module, and a user interaction module. Risk levels, status tags, and manual verification tasks can be uploaded to the nursing station terminal. Each room can first extract point cloud features, bed pressure features, and voice response features locally, and then upload the low-privacy feature data or risk results. Upon receiving a high-risk event, the nursing station terminal can assign tasks based on room number, elderly person's status tag, and caregiver's location. This deployment method facilitates the use of monitoring devices in multi-room elderly care facilities without adding complex mechanical equipment.

[0045] In the above embodiments, modules can be added, removed, or merged according to actual needs. For example, the voice interaction evaluation module and the user interaction module can be integrated into the same bedside terminal; the event anchoring fusion module, the digital twin status update module, and the privacy and security module can be set in the same edge computing terminal; the service scheduling module can be set in the nursing station server, cloud platform, or family terminal application. As long as it is possible to extract multi-source features within the time window before and after the occurrence of the abnormal initial event, using the time of the abnormality as the event anchor point, and perform time alignment, conflict verification, and confidence fusion, it is considered an implementation of the technical solution of this invention.

Claims

1. A smart elderly care monitoring device specifically designed for elderly care services, characterized in that, It includes a non-contact behavior perception module, a bed vital signs perception module, a voice interaction assessment module, an environmental status acquisition module, an indoor positioning module, an event anchoring and fusion module, a digital twin status update module, a service scheduling module, and a privacy and security module. The non-contact behavior perception module is used to collect millimeter-wave radar point cloud data in the space where the elderly are located and extract point cloud behavior features. The bed vital signs sensing module is used to collect bed pressure time-series signals and extract bed pressure characteristics; The voice interaction evaluation module is used to extract voice response features when the elderly make voice responses; The environmental status acquisition module is used to collect environmental characteristics within the space where the elderly person is located; The indoor positioning module is used to determine the functional area where the elderly person is located and outputs positioning features; When any of the above modules generates an abnormal initial event, the event anchoring and fusion module determines the time of occurrence of the abnormal initial event as the event anchor point, and extracts point cloud behavioral features, bed pressure features, voice response features, environmental features and positioning features within a preset time window before and after the event anchor point. The extracted features are then time-aligned, conflict-checked and confidence-fused to generate a care risk level. The digital twin status update module updates the elderly person's status tag according to the care risk level; The service scheduling module outputs care treatment strategies based on the care risk level and status label. The privacy and security module is used to impose security constraints on the data collection, processing, transmission, storage, and retrieval processes.

2. The intelligent elderly care device for elderly care services according to claim 1, characterized in that, The event anchoring fusion module includes an abnormal initial event triggering unit, an event anchor point determination unit, a time window interception unit, a feature time alignment unit, a conflict verification unit, a confidence correction unit, and a risk level generation unit. The abnormal initial event triggering unit is used to receive abnormal initial events; The event anchor point determination unit is used to determine the event anchor point; The time window extraction unit is used to extract multi-source features within the forward and backward time windows of the event anchor point; The feature time alignment unit is used to align multi-source features with different sampling frequencies to the same time axis; The conflict verification unit is used to determine the mutual support relationship or mutual conflict relationship between multi-source features; The confidence correction unit is used to correct the confidence of the corresponding feature according to the mutual support relationship or mutual conflict relationship; The risk level generation unit is used to generate a care risk level based on the corrected confidence level.

3. The intelligent elderly care device for elderly care services according to claim 1, characterized in that, The abnormal initial event includes at least one of the following: rapid descent of the body's center of gravity, prolonged low-lying position, failure to return to bed after leaving it at night, lack of effective voice response, abnormal sleep patterns, or abnormal activity trends.

4. The intelligent elderly care device for elderly care services according to claim 1, characterized in that, When the event anchoring fusion module reviews a fall event, it monitors the duration of low-position stay after the non-contact behavior perception module detects a rapid descent of the human body's height center and determines whether an initial low-position stay has been formed. If an initial low-level pause does not occur, output "Cancel alarm". If an initial low-level dwell is formed, the bed pressure characteristics are read, and it is determined whether a stable bed pressure distribution has been formed. If a stable pressure distribution is formed when getting into bed, the system will output the active option to get into bed or switch between sitting and lying down, based on the positioning characteristics. If a stable pressure distribution for getting into bed is not established, the voice response characteristics are read, and high-risk fall, delayed review, sitting-to-lying transition, or bending-over events are output based on whether a valid response is obtained and the location characteristics.

5. A smart elderly care device for elderly care services according to claim 4, characterized in that, The stable bed pressure distribution is determined based on the total pressure change, pressure distribution duration, and pressure distribution stability in the bed pressure time-series signal. When a stable bed-going pressure distribution matching the bed-going state appears within the backward time window after the body's height center rapidly descends, the event anchoring fusion module reduces the fall confidence. If a stable pressure distribution matching the bed-going state does not appear within the backward time window after the body's height center rapidly descends, and no effective voice response is obtained, the event anchoring fusion module increases the fall confidence level.

6. The intelligent elderly care device for elderly care services according to claim 1, characterized in that, When the event anchoring fusion module reviews nighttime bed-leaving events, it takes the moment when the bed vital signs sensing module detects the nighttime bed-leaving event as the event anchor point and extracts the positioning features, environmental features, and bed pressure features within the backward time window after the event anchor point. When the location data shows that the elderly person enters the bathroom and returns to the bed within a preset time, and the bed pressure characteristics return to a stable state, it is recorded as a normal nighttime bed-leaving event. When location data shows that an elderly person stays in the bathroom, bedside, or corridor area for more than a preset time after getting out of bed, and the bed pressure characteristics have not returned to a stable state while still in bed, the risk level of abnormal nighttime bed leaving is increased.

7. The intelligent elderly care device for elderly care services according to claim 1, characterized in that, When the digital twin state update module judges the decline trend of cognitive response, it acquires voice interaction features, sleep state features, nighttime bed-leaving features, and daytime activity features over multiple consecutive evaluation periods. The voice interaction features include response latency, pause ratio, and speech rate fluctuation; The sleep state characteristics include sleep continuity; The nighttime bed-leaning characteristics include the number of nighttime bed-leanings; The daytime activity characteristics include the range of daytime activities; When, within multiple consecutive evaluation periods, the voice interaction features exhibit at least one of the following: increased response delay, increased pause rate, or increased speech rate fluctuation, and simultaneously meet the criteria of decreased sleep continuity, increased nighttime bed-getting frequency, and reduced daytime activity range, the digital twin status update module updates the elderly person's status label to a cognitive response decline trend label and sends the cognitive response decline trend label to the service scheduling module.

8. The intelligent elderly care device for elderly care services according to claim 1, characterized in that, The digital twin status update module updates status tags based on care risk level and multi-period trend data. The status tags include at least one of the following: high risk of nighttime falls, high risk of bedside falls, abnormal risk of staying in the bathroom, abnormal risk of sleep breathing, frequent risk of getting out of bed at night, and a trend of declining cognitive response or shrinking activity range.

9. A smart elderly care device for elderly care services according to claim 1, characterized in that, The privacy and security module extracts feature data locally from millimeter-wave radar point cloud data, bed pressure time-series signals, and voice response data, and saves point cloud features, attitude categories, bed pressure features, voice response features, risk levels, and status labels. The privacy and security module does not store video images, complete voice content, or complete original pressure waveforms for extended periods, and it encrypts and controls the transmission and access permissions of data uploaded to the service scheduling module, nursing station terminal, or family member terminal.

10. A smart elderly care device for elderly care services according to claim 1, characterized in that, The service scheduling module outputs a graded care and treatment strategy based on the care risk level. When the care risk level is low, output the observation strategy. When the care risk level is medium risk, output voice reminders, secondary confirmations, or reminders for family members to pay attention; When the care risk level is high, a manual verification task is assigned. When the care risk level is classified as emergency risk, an emergency response strategy will be implemented. The non-contact behavior perception module, bed vital signs perception module, voice interaction assessment module, environmental status acquisition module, indoor positioning module, event anchoring fusion module, digital twin status update module, service scheduling module, and privacy and security module are integrated or distributed and connected via wired or wireless communication.

Images