Millimeter wave radar and vision fusion-based method and system for identifying abnormal behaviors of elderly living at home

Abstract

The invention relates to the technical field of intelligent old-age care monitoring, and discloses a method for identifying abnormal behaviors of home elderly based on millimeter-wave radar and vision fusion, which comprises the following steps: S1, multi-modal data acquisition and privacy protection configuration: deploying a configurable sensing suite in a home environment of the elderly, comprising a millimeter wave radar, a visual sensor (infrared and visible light), environment sensors (doors and windows, water, illumination and ground vibration) and wearable equipment (such as a smart bracelet / chest card). According to the method and the system for identifying the abnormal behaviors of the home elderly based on millimeter-wave radar and vision fusion, all-around behavior monitoring is realized on the premise of guaranteeing the privacy of the elderly through multi-modal sensor fusion and a hierarchical privacy protection mechanism, and the millimeter-wave radar and the vision sensor are complementarily fused, so that the privacy worry of a pure vision technology is overcome, and the safety of the home elderly is improved. And the limitation of a single radar on complex behavior identification is solved, and meanwhile, a configurable sensing suite and a hardware-level privacy switch are adopted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent elderly care monitoring, in particular to a home elderly abnormal behavior recognition method and system based on millimeter wave radar and visual fusion. BACKGROUND

[0002] With the intensification of population aging trend, the safety monitoring problem of the elderly living alone is increasingly prominent, the traditional video monitoring method has the risk of privacy leakage, and is greatly affected by the light condition; while the single sensor technology such as millimeter wave radar can protect privacy, but the precision is limited in complex behavior recognition.

[0003] In the prior art, the scheme based on wearable devices has the problem of forgetting to wear by the elderly, and most systems cannot work normally in a weak network environment, have a high false positive rate, and lack long-term health trend analysis capability, therefore, there is an urgent need for an elderly abnormal behavior recognition scheme which can protect privacy, accurately recognize, and has network resilience. SUMMARY

[0004] (I) Technical problem solved In view of the shortcomings of the prior art, the present application provides a home elderly abnormal behavior recognition method and system based on millimeter wave radar and visual fusion, which has the advantages of good privacy protection, high recognition accuracy, strong network resilience, and support for long-term health monitoring, and solves the problems of traditional monitoring schemes, such as high risk of privacy leakage, poor environmental adaptability, high false positive rate, and lack of trend analysis capability.

[0005] (II) Technical scheme In order to achieve the purpose of good privacy protection, high recognition accuracy, strong network resilience, and support for long-term health monitoring, the present application provides the following technical scheme: a home elderly abnormal behavior recognition method based on millimeter wave radar and visual fusion, comprising the following operation steps: Step S1: multi-modal data acquisition and privacy protection configuration Deploy a configurable sensor kit in the home environment of the elderly, including millimeter wave radar, visual sensor (infrared + visible light), environmental sensor (door and window, water use, light, ground vibration), and wearable device (such as smart bracelet / chest card), and dynamically enable or shield the sensors according to the user privacy settings to ensure reliable data acquisition under the premise of protecting privacy; Step S2: edge real-time behavior perception and preliminary identification Run a high-reliability fall detection algorithm and a local scene understanding engine on the local edge node to process sensor data in real time, identify behaviors such as "fall", "get up", "sit down", etc., and perform multi-sensor cross-validation on high-risk events in combination with the "three-time verification" principle to control false positives; Step S3: Data caching and disaster recovery handling for weak / outage networks In the event of network instability or interruption, edge nodes encrypt and store the collected data and alarm events locally, and automatically synchronize them to the cloud after the network is restored, ensuring that there is no gap in monitoring. Step S4: Cloud-based behavioral modeling and long-term trend analysis Establish a “digital twin” behavioral profile for the elderly in the cloud, learn their daily routines, behavioral styles and other personalized baselines, and achieve early warning of slow changes in abnormalities through long-term data analysis of activity levels, sleep patterns, gait stability, etc. Step S5: False Alarm Feedback and Continuous Model Optimization The system receives user feedback on false alarms through the App interface and uses this feedback data to continuously optimize the edge and cloud recognition models to improve recognition accuracy. Step S6: Multi-role collaborative service and closed-loop response Based on the identification results, a multi-level response mechanism is triggered, from seamless self-check to emergency rescue. Information synchronization and service loop are achieved through a multi-role interaction platform (elderly end, family / caregiver end, community / institution end), and ecological service resources (such as meal delivery, medical care, rehabilitation, etc.) are connected.

[0006] Preferably, the multimodal data acquisition in step S1 specifically includes: Millimeter-wave radar is used to collect the elderly’s movement trajectory, vital signs and micro-motion information to obtain behavioral data that is not affected by light conditions. Infrared and visible light video data are collected by a visual sensor and processed into a skeletal key point sequence in real time at the edge to avoid leakage of the original image data. The system collects multi-dimensional environmental parameters such as door and window status, water usage, light intensity, and ground vibration through a group of environmental sensors. Wearable devices are used to collect physiological parameters such as heart rate and blood oxygen saturation of the elderly, as well as their outdoor activity trajectories.

[0007] Preferably, the edge intelligence processing in step S2 specifically includes: An abnormal behavior recognition model based on multimodal data fusion is constructed. This model uses an attention mechanism to dynamically weight the importance of data from different sensors. The highly reliable fall detection algorithm analyzes multi-dimensional features such as the rate of change of the human body's center of mass, changes in posture angle, and duration of stillness, and combines these with a set threshold judgment rule to achieve fall recognition.

[0008] A system for identifying abnormal behavior of elderly people living at home based on millimeter-wave radar and vision fusion, including a smart elderly care monitoring system, which includes a humanized perception layer, a resilient and intelligent edge layer, a cloud-based smart brain and model and services, and an interaction and ecosystem layer. The human-centered perception layer is used to achieve reliable and configurable data collection while ensuring privacy and user experience. The resilience and intelligent edge layer are used to achieve local high-reliability decision-making, ensuring that the core functions of the system remain intact in weak network / network outage environments; The cloud-based intelligent brain and model are used to achieve personalized analysis, long-term trend mining, and system self-evolution. The services, interactions, and ecosystem layers are used to build a collaborative, proactive, and positive health promotion closed loop.

[0009] Preferably, the humanized perception layer includes a configurable sensing suite and dynamic perception and privacy mechanisms; The configurable sensing kit is a tiered system, including a basic version, an enhanced version, and wearable device supplements (such as smart bracelets / badges). The dynamic sensing and privacy mechanism includes user-definable "privacy time / space", hardware-level privacy switch and data anonymization processing.

[0010] Preferably, the resilience and intelligent edge layer includes a highly reliable fall detection algorithm, a local scene understanding engine, a "three-times verification" alarm principle, and edge data caching and synchronization; The highly reliable fall detection algorithm is used to run a carefully optimized, low-computation fall detection model (based on skeletal points or radar trajectories) independently at the edge. Its primary goal is to control false alarms, preferring to miss alarms rather than have frequent false alarms. The local scene understanding engine is used to understand normal patterns such as "getting up", "sitting on the sofa", and "walking around in the kitchen", reducing misjudgments of these behaviors; The “three-step verification” alarm principle is used to prevent suspected high-risk events (such as falls) from being immediately reported to the cloud. The edge node will cross-verify the data from multiple sensors (for example, radar detects a rapid fall + an environmental microphone (optional, requires authorization) captures an impact sound / groan + no movement for a long time afterward) before triggering a high-level alarm. The edge data caching and synchronization are used to encrypt and store all data and alarm events locally when the network is interrupted, and automatically synchronize them after the network is restored, leaving no monitoring gap.

[0011] Preferably, the cloud-based intelligent brain and model include establishing a "digital twin" behavior profile, early warning of slow-changing anomalies, a false alarm feedback learning cycle, and multi-household learning and knowledge transfer; The establishment of a "digital twin" behavioral profile is used to learn the elderly's daily routines and behavioral styles, such as walking speed, range of activity, and dominant hand, to form a unique baseline. The early warning system for slow-change abnormalities analyzes long-term data such as activity levels, sleep patterns, and gait stability to indicate a "weakness trend" weeks or months in advance, thus buying valuable time for intervention. The false alarm feedback learning loop provides a simple App feedback interface, where family members or the elderly can mark "false alarms". The system collects this feedback to continuously optimize the edge and cloud models, making them more accurate with use. The multi-resident learning and knowledge transfer mechanism, under strict anonymity and aggregation, enables the system to learn common patterns from massive user data, thereby improving the initial accuracy of the system for new users.

[0012] Preferably, the service, interaction, and ecosystem layer includes a multi-role interaction platform and a service closed loop; The multi-role interaction platform includes an elderly user interface (minimalist interaction), a family member / caregiver interface (App / Mini Program), and a community / institution interface (management backend). The service loop includes a multi-level response mechanism and access to ecosystem services.

[0013] (III) Beneficial Effects Compared with existing technologies, this invention provides a method and system for identifying abnormal behaviors of elderly people living at home based on millimeter-wave radar and visual fusion, which has the following beneficial effects: 1. This method and system for identifying abnormal behaviors of elderly people living at home based on millimeter-wave radar and visual fusion achieves comprehensive behavior monitoring while protecting the privacy of the elderly through multimodal sensor fusion and hierarchical privacy protection mechanism. By adopting complementary fusion of millimeter-wave radar and visual sensors, it overcomes the privacy concerns of pure visual technology and solves the limitations of single radar in complex behavior recognition. At the same time, through configurable sensor kits and hardware-level privacy switches, the elderly can independently control the monitoring permissions, realizing humanized monitoring.

[0014] 2. This method and system for identifying abnormal behavior of elderly people living at home based on millimeter-wave radar and vision fusion significantly improves the recognition accuracy and reduces false alarms through edge intelligence and a three-level verification mechanism. A lightweight recognition model is deployed at the local edge to achieve real-time behavior analysis and fall detection. The "three-level verification" principle is adopted to cross-verify high-risk events through multi-sensor data, effectively avoiding false alarms caused by normal activities (such as sitting down or lying down) and improving system reliability.

[0015] 3. This method and system for identifying abnormal behaviors of elderly people living at home, based on the fusion of millimeter-wave radar and vision, achieves early warning of health status through cloud-based digital twins and long-term trend analysis, establishes personalized behavioral profiles for the elderly, and identifies frailty trends weeks or months in advance by analyzing long-term data such as activity levels, sleep patterns, and gait stability, thus gaining valuable time for early intervention and realizing the transformation from passive emergency response to proactive prevention.

[0016] 4. This home-based elderly abnormal behavior recognition system, based on millimeter-wave radar and visual fusion, constructs a complete closed loop for elderly care monitoring through multi-level response and service ecosystem integration. It initiates differentiated responses from unnoticed self-checks to emergency rescues based on the risk level of the event. At the same time, it integrates resources from communities, medical institutions, and domestic services to form a multi-party collaborative health promotion ecosystem, truly realizing an innovative model of technology-enabled elderly care services. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method for identifying abnormal behavior of elderly people at home according to the present invention; Figure 2 This is a schematic diagram of the intelligent elderly care monitoring system of the present invention; Figure 3 This is a schematic diagram of the multi-role interaction platform of the present invention; Figure 4 This is a schematic diagram of the service closed loop of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figures 1-4 A method for identifying abnormal behavior of elderly people living at home based on millimeter-wave radar and vision fusion includes the following steps: Step S1: Multimodal data acquisition and privacy protection configuration Deploy configurable sensor suites in the home environment for the elderly, including millimeter-wave radar, visual sensors (infrared + visible light), environmental sensors (doors and windows, water usage, light intensity, ground vibration), and wearable devices (such as smart bracelets / badges), and dynamically enable or disable sensors according to user privacy settings to ensure reliable data collection while protecting privacy. Step S2: Real-time behavior perception and preliminary identification at the edge A highly reliable fall detection algorithm and a local scene understanding engine are run on local edge nodes to process sensor data in real time, identify behaviors such as "falling", "getting up", and "sitting down", and perform multi-sensor cross-verification of high-risk events in combination with the "three-times verification" principle to control false alarms; Step S3: Data caching and disaster recovery handling for weak / outage networks In the event of network instability or interruption, edge nodes encrypt and store the collected data and alarm events locally, and automatically synchronize them to the cloud after the network is restored, ensuring that there is no gap in monitoring. Step S4: Cloud-based behavioral modeling and long-term trend analysis Establish a “digital twin” behavioral profile for the elderly in the cloud, learn their daily routines, behavioral styles and other personalized baselines, and achieve early warning of slow changes in abnormalities through long-term data analysis of activity levels, sleep patterns, gait stability, etc. Step S5: False Alarm Feedback and Continuous Model Optimization The system receives user feedback on false alarms through the App interface and uses this feedback data to continuously optimize the edge and cloud recognition models to improve recognition accuracy. Step S6: Multi-role collaborative service and closed-loop response Based on the identification results, a multi-level response mechanism is triggered, from seamless self-check to emergency rescue. Information synchronization and service loop are achieved through a multi-role interaction platform (elderly end, family / caregiver end, community / institution end), and ecological service resources (such as meal delivery, medical care, rehabilitation, etc.) are connected.

[0020] A system for identifying abnormal behavior of elderly people living at home based on millimeter-wave radar and vision fusion, including a smart elderly care monitoring system, which includes a humanized perception layer, a resilient and intelligent edge layer, a cloud-based smart brain and model and service, and an interaction and ecosystem layer. The human-centered perception layer is used to achieve reliable and configurable data collection while ensuring privacy and user experience. The resilient and intelligent edge layer is used to enable local high-reliability decision-making, ensuring that the core functions of the system remain intact in weak network / network outage environments; Cloud-based intelligent brains and models are used to achieve personalized analysis, long-term trend mining, and system self-evolution. The service, interaction, and ecosystem layers are used to build a collaborative, proactive, and positive health promotion closed loop.

[0021] In the implementation of the case, the humanized perception layer includes configurable sensor kits and dynamic perception and privacy mechanisms; The configurable sensing kit is a tiered system, including a basic version, an enhanced version, and wearable device supplements (such as smart bracelets / badges). The basic version only deploys an environmental sensor group (doors and windows, water usage, lighting, ground vibration) and a low-power non-visual life radar (which only detects presence and vital signs, without micro-motion recognition), making it suitable for users who are extremely sensitive to privacy. The enhanced version uses visual sensors (infrared + visible light) in public areas (living room, kitchen) and strictly processes them as skeletal points at the edge. Millimeter-wave radar is used in bedrooms / bathrooms, but with clear indicator lights and physical switches, the elderly can disable the sensing function of specific areas with one click. Wearable devices (such as smart bracelets / badges) are used to supplement outdoor activity monitoring and one-click SOS function, forming indoor and outdoor coverage.

[0022] This hierarchical perception solution establishes an intelligent monitoring system that respects personal privacy while providing effective security, providing a reliable data foundation for subsequent abnormal behavior identification and analysis, and ensuring the practical feasibility and user acceptance of the technical solution.

[0023] Dynamic awareness and privacy mechanisms include user-definable "privacy time / space", hardware-level privacy switches, and data anonymization processing; Among them, the user-definable "privacy time / space" allows seniors to set "do not disturb time periods" or turn off specific sensors.

[0024] Hardware-level privacy switch: A physical switch is placed on the sensor to provide the most intuitive sense of control.

[0025] Data anonymization: In the cloud, pseudonyms are used to replace direct personal identification information.

[0026] This privacy mechanism significantly enhances users' trust and acceptance of the monitoring system, resolves the prominent contradiction in the field of intelligent monitoring between "security and privacy," and lays a solid foundation for the long-term stable operation and widespread application of the system.

[0027] In the case implementation, the resilience and intelligent edge layer includes a highly reliable fall detection algorithm, a local scene understanding engine, a "three-time verification" alarm principle, and edge data caching and synchronization; The highly reliable fall detection algorithm is used to run a carefully optimized, low-computation fall detection model (based on skeletal points or radar trajectories) independently at the edge. Its primary goal is to control false alarms, preferring to miss alarms rather than have frequent false alarms. The local scene understanding engine is used to understand normal patterns such as "getting up", "sitting on the sofa", and "walking in the kitchen", reducing misjudgments of these behaviors; The "three-step verification" alarm principle is used to prevent suspected high-risk events (such as falls) from being immediately reported to the cloud. The edge nodes will cross-verify the data from multiple sensors (for example, radar detects a rapid fall + an environmental microphone (optional, requires authorization) captures the impact sound / groan + no movement for a long time afterward) before triggering a high-level alarm. Edge data caching and synchronization are used when the network is interrupted. All data and alarm events are encrypted and stored locally, and automatically synchronized after the network is restored, leaving no monitoring gap.

[0028] In the implementation of the case, the cloud-based intelligent brain and model include the establishment of "digital twin" behavioral profiles, early warning of slow-changing anomalies, false alarm feedback learning cycle, and multi-household learning and knowledge transfer; A "digital twin" behavioral profile is created to learn the elderly's daily routines and behavioral styles, such as walking speed, range of activity, and dominant hand, forming a unique baseline. Early warning of slow-change abnormalities can indicate a "weakness trend" weeks or months in advance by analyzing long-term data such as activity level, sleep patterns, and gait stability, thus buying valuable time for intervention. The false alarm feedback learning loop provides a simple app feedback interface where family members or seniors can mark "false alarms." The system collects this feedback to continuously optimize the edge and cloud models, making them more accurate with use. Multi-resident learning and knowledge transfer are used to enable the system to learn common patterns from massive user data under strict anonymity and aggregation, thereby improving the initial accuracy of the system for new users.

[0029] In the implementation of the case, the service, interaction, and ecosystem layers include a multi-role interaction platform and a service loop; The multi-role interaction platform includes a user interface for seniors (minimalist interaction), a family member / caregiver interface (App / Mini Program), and a community / institution interface (management backend). The elderly-friendly (minimalist interaction) version includes the following: Voice assistant: Used for proactive inquiries ("Did I sleep well today?"), setting reminders, and controlling smart home devices; Smart speakers offer proactive care: they regularly broadcast weather, news, and messages from friends and family, and gently remind users to move around when they are sitting for long periods. "Digital Companion": Displays health achievements (such as "Completed walking goals for 7 consecutive days") and plays memory albums to provide positive motivation through a simple interface.

[0030] Among them, the family / caregiver interface (App / Mini Program): "Safety Notification": A daily briefing stating "everything is normal" greatly alleviates the anxiety of family members; Abstract animation playback: When checking for abnormal events, view the animation reconstructed from skeletal points, with absolute privacy protection; Health Trends Report: Visual charts show long-term changes in activity, sleep, weight, etc.

[0031] Among them, the community / institutional side (management backend): Resource Dispatch Center: Based on system alerts (such as multiple elderly people showing reduced activity), intelligently dispatches caregivers to their care. Community health insights: Anonymized statistics on the overall health status of elderly people in the community, used for public service planning.

[0032] The service loop includes a multi-level response mechanism and access to ecosystem services; The multi-level response mechanism includes the following: L0 Seamless Self-Check: The system automatically processes and marks most normal situations without manual intervention; L1 Active Reminder: The system will remind the elderly via voice to avoid issues such as prolonged sitting or missing medication. L2 Family Collaboration: For confirmed low- to medium-risk abnormalities (such as not returning to bed at night), notify the family members to verify; L3 Emergency Assistance: Upon confirmation of a high-risk event (such as a fall), it automatically connects with emergency contacts, property / community centers, and 120 (emergency services), and provides key information (such as location and possible injuries).

[0033] The ecosystem service access includes the following: By connecting with pharmacies, food delivery services, and rehabilitation facilities, seniors or their families can book services with a single click through the system. Introduce online doctors to provide initial remote health consultations for abnormal long-term trends.

[0034] In summary, this method and system for identifying abnormal behavior in elderly people living at home, based on the fusion of millimeter-wave radar and vision, achieves comprehensive behavior monitoring while protecting the privacy of the elderly through multimodal sensor fusion and a hierarchical privacy protection mechanism. The complementary fusion of millimeter-wave radar and visual sensors overcomes privacy concerns associated with purely visual technologies and addresses the limitations of single radar in complex behavior recognition. Furthermore, configurable sensor kits and hardware-level privacy switches allow the elderly to control their monitoring permissions, enabling humanized supervision. Edge intelligence and a three-level verification mechanism significantly improve recognition accuracy and reduce false alarms. A lightweight recognition model is deployed at the local edge to achieve real-time behavior analysis and fall detection. The "three-stage verification" principle, through cross-verification of high-risk events using multi-sensor data, effectively avoids false alarms caused by normal activities (such as sitting or lying down), thus improving system reliability.

[0035] Furthermore, by leveraging cloud-based digital twins and long-term trend analysis, early warnings of health status can be achieved, establishing personalized behavioral profiles for the elderly. By analyzing long-term data such as activity levels, sleep patterns, and gait stability, frailty trends can be identified weeks or months in advance, buying valuable time for early intervention and shifting from passive emergency response to proactive prevention. Through multi-level response and service ecosystem integration, a complete closed-loop elderly care monitoring system can be constructed. Differentiated responses, ranging from unnoticed self-checks to emergency rescues, can be initiated based on the level of event risk. At the same time, resources from communities, medical institutions, and domestic services can be integrated to form a multi-party collaborative health promotion ecosystem, truly realizing an innovative model of technology-enabled elderly care services.

[0036] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0037] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for recognizing abnormal behavior of an elderly person at home based on fusion of millimeter wave radar and vision, characterized by: The method comprises the following operation steps: Step S1: multi-modal data acquisition and privacy protection configuration Deploy a configurable sensor suite in the home environment of the elderly, including millimeter wave radar, visual sensor (infrared + visible light), environmental sensor (door and window, water use, light, ground vibration), and wearable device (such as smart bracelet / chest card), and dynamically enable or shield the sensors according to the user privacy settings to ensure reliable data acquisition while protecting privacy; Step S2: edge real-time behavior perception and preliminary identification Run a high-reliability fall detection algorithm and a local scene understanding engine on the local edge node to process sensor data in real time, identify behaviors such as "fall", "get up", "sit down", and perform multi-sensor cross-validation on high-risk events in combination with the "three verification" principle to control false positives; Step S3: data caching and weak network / offline disaster recovery In the case of unstable or interrupted network, the edge node locally stores the collected data and alarm events after encryption, and automatically synchronizes them to the cloud after the network is restored, ensuring that there is no monitoring gap; Step S4: cloud behavior modeling and long-term trend analysis Establish a "digital twin" behavior profile of the elderly in the cloud, learn their daily routines, behavior styles, and other personalized baselines, and analyze their activity levels, sleep patterns, and gait stability through long-term data analysis to provide early warning of slow-changing abnormalities; Step S5: false alarm feedback and continuous model optimization Receive user feedback on false alarm events through the App interface, and use these feedback data to continuously optimize the edge and cloud recognition models to improve recognition accuracy; Step S6: multi-role collaborative service and closed-loop response Trigger a multi-level response mechanism based on the identification results, from self-checking to emergency rescue, and achieve information synchronization and service closed loop through a multi-role interaction platform (elderly, family / caregiver, community / institution), and access to ecological service resources (such as meal delivery, medical care, rehabilitation, etc.). 2.The home-based elderly abnormal behavior recognition method and system based on fusion of millimeter wave radar and vision according to claim 1, characterized in that: The multi-modal data acquisition in step S1 specifically includes: Collect the motion trajectory, vital signs, and micro-motion information of the elderly through millimeter wave radar to obtain behavior data that is not affected by lighting conditions; Collect infrared and visible light video data through visual sensors, and process them into skeletal key point sequences in real time on the edge to avoid leakage of raw image data; Collect multi-dimensional environmental parameters such as door and window status, water use, light intensity, and ground vibration through the environmental sensor group; Collect physiological parameters such as heart rate and blood oxygen saturation of the elderly through wearable devices, as well as outdoor activity trajectories. 3.The home-based abnormal behavior recognition method and system for the elderly based on fusion of millimeter wave radar and vision according to claim 1, characterized in that: The edge intelligent processing in step S2 specifically includes: Build an abnormal behavior recognition model based on multi-modal data fusion, which dynamically weights the importance of different sensor data using an attention mechanism; The high-reliability fall detection algorithm identifies falls by analyzing multiple features such as body center of mass change rate, posture angle change, and stationary duration, and combining with the set threshold judgment rule.

4. A system for recognizing abnormal behavior of the elderly at home based on fusion of millimeter wave radar and vision, comprising a smart elderly care monitoring system, characterized in that: The smart elderly care monitoring system includes a humanized perception layer, a resilient and intelligent edge layer, a cloud-based wisdom brain and model and service, and an interaction and ecological layer; The humanized perception layer is used to achieve reliable and configurable data acquisition while protecting privacy and user experience; The resilience and intelligent edge layer is used to achieve local high-reliability decision-making, and the system core function is not dead in a weak network / disconnected network environment. The cloud-based wisdom brain and model are used to achieve personalized analysis, long-term trend mining, and system self-evolution. The service, interaction, and ecological layer is used to build a multi-party collaborative and actively positive health promotion closed loop.

5. The system for recognizing abnormal behavior of the old people at home based on fusion of millimeter wave radar and vision according to claim 4, characterized in that: The humanized perception layer includes a configurable sensing suite and a dynamic perception and privacy mechanism. The configurable sensing suite is a hierarchical scheme, including a basic version, an enhanced version, and wearable device supplements (such as smart bands / chest cards). The dynamic perception and privacy mechanism includes user-definable "privacy time / space", hardware-level privacy switches, and data anonymization processing.

6. The system for recognizing abnormal behavior of the old people at home based on fusion of millimeter wave radar and vision according to claim 4, characterized in that: The resilience and intelligent edge layer includes a high-reliability fall detection algorithm, a local scene understanding engine, a "three-verification" alarm principle, and edge data caching and synchronization. The high-reliability fall detection algorithm is used to independently run a carefully optimized, low-computational fall detection model (based on skeletal points or radar trajectories) on the edge, with the primary goal of controlling false positives, and preferring to miss a report rather than frequently misreporting. The local scene understanding engine is used to understand "getting up", "sitting down on the sofa", "walking in the kitchen", and other normal patterns, reducing false positives for these behaviors. The "three-verification" alarm principle is used for suspected high-risk events (such as falls) that are not immediately reported to the cloud. The edge node cross- verifies multi-sensor data (for example, radar detects rapid falling + environmental microphone (optional, requires authorization) captures impact sound / groan + subsequent long period of no movement), and confirms before triggering a high-level alarm. The edge data caching and synchronization is used to store all data and alarm events locally in encrypted form during network interruptions, and automatically synchronize when the network is restored, leaving no monitoring gaps.

7. The system for recognizing abnormal behavior of old people at home based on fusion of millimeter wave radar and vision according to claim 4, characterized in that: The cloud-based wisdom brain and model include establishing a "digital twin" behavior profile, early warning of slow-changing abnormalities, a false positive feedback learning cycle, and multi-resident learning and knowledge transfer. The establishment of a "digital twin" behavior profile is used to learn the daily routine of the elderly, as well as their behavior style, such as walking speed, activity range, and dominant hand, forming a unique baseline. The early warning of slow-changing abnormalities analyzes long-term data such as activity level, sleep regularity, and gait stability to provide a "weakening trend" weeks or months in advance, providing valuable time for intervention. The false positive feedback learning cycle provides a simple App feedback interface, allowing family members or the elderly to mark "false positives". The system collects these feedbacks to continuously optimize the edge and cloud models, making them more accurate over time. The multi-resident learning and knowledge transfer is used to learn common patterns from massive user data under strict anonymity and aggregation, improving the initial accuracy of the system for new users.

8. The system for recognizing abnormal behavior of old people at home based on fusion of millimeter wave radar and vision according to claim 4, characterized in that: The service, interaction, and ecological layer includes a multi-role interaction platform and a service closed loop. The multi-role interaction platform includes the elderly end (extremely simple interaction), the family / caregiver end (App / mini program), and the community / institution end (management background). The service closed loop includes a multi-level response mechanism and ecological service access.

Images