An old person fall early warning method and system based on multi-dimension detection

Abstract

This invention discloses a fall warning method and system for the elderly based on multi-dimensional detection, relating to the field of fall warning for the elderly. The method includes: collecting multi-dimensional data in a low-power manner using an elderly wristband, a regular camera, and environmental sensors; the edge layer automatically switching between daytime assessment and nighttime response modes based on time and activity intensity; the daytime mode uses dynamic risk coefficient calculation to screen high-risk targets and performs focused camera tracking analysis, while the nighttime mode is triggered by mattress sensors for immediate response; finally, based on the assessment results, a tiered task is generated and dispatched to the nearest caregiver for execution via Bluetooth positioning, while the model is optimized through feedback. This method effectively reduces the system's dependence on high-precision hardware and computing power, is low-cost and highly reliable, significantly reduces false alarms and protects privacy; it achieves a shift from passive alarm to active intervention, blocking fall risks at the source through human-machine collaboration, and significantly improving warning effectiveness.

Description

Technical Field

[0001] This invention relates to the field of fall warning for the elderly, and in particular to a fall warning method and system for the elderly based on multi-dimensional detection. Background Technology

[0002] Faced with the increasingly serious problem of an aging population, the traditional family-based elder care model can no longer meet the growing demand for elder care. More and more elderly people, especially those living alone or in empty nests, are choosing to enter nursing homes and other facilities to seek more professional care services. However, the biggest problem in nursing homes is falls among the elderly. Therefore, effectively preventing and reducing falls among the elderly and ensuring their safety and health has become a focal point of social concern.

[0003] Existing fall warning methods for the elderly generally involve deploying high-precision cameras, sensors, and radar throughout nursing homes for 24 / 7 monitoring. These devices, combined with high computing power, perform gait analysis, posture analysis, and arm swing balance detection for each elderly person, providing monitoring and early warning based on real-time dynamic data and static health data.

[0004] However, this early warning method, due to the extensive use of high-precision cameras and even radar, results in high costs and the 24 / 7 operation of various facilities is prone to damage. Furthermore, the system's real-time dynamic calculations for each elderly person around the clock pose challenges to server hardware costs and network bandwidth quality, potentially leading to high server power consumption and, in the long run, server crashes. In addition, the 24 / 7 monitoring of each elderly person with various high-precision devices not only infringes on their privacy but also causes false alarms due to algorithm errors, leading to aversion to the early warning method and monitoring. Finally, existing methods rely solely on dynamic posture algorithms, only effective when an elderly person is about to fall. Therefore, they not only fail to prevent falls at their source but also lack effective preventative measures when a fall is imminent, resulting in ineffective prevention and only allowing for post-fall remedial action, leading to poor early warning results. To address these issues, this invention provides a multi-dimensional detection-based fall early warning method and system for the elderly to solve these problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a fall warning method and system for the elderly based on multi-dimensional detection. It solves the problems of existing warning methods, which suffer from high costs due to the extensive use of sophisticated cameras and even radar, and the vulnerability of these constantly operating facilities; the challenges posed by the system's real-time dynamic calculations for each elderly person, which strain server hardware costs and network bandwidth, leading to high power consumption and potential server crashes in the long run; the privacy violations caused by monitoring each elderly person around the clock with various sophisticated devices, and the potential for false alarms due to algorithm errors, resulting in aversion to the warning method and monitoring; and finally, existing methods rely solely on dynamic posture algorithms, only effective when a fall is imminent. Therefore, they fail to prevent falls at their source and lack effective preventative measures when a fall is imminent, resulting in ineffective prevention and only allowing for reactive, post-fall remedial measures, leading to poor warning performance.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method and system for early warning of falls in the elderly based on multi-dimensional detection, comprising:

[0007] S1. Multi-dimensional data perception based on low-power terminal layer:

[0008] This step involves the elderly wearing a Bluetooth beacon terminal wristband, the caregiver wearing a corresponding Bluetooth terminal, and terminal devices deployed in the environment, including ordinary cameras, pressure sensors, and IoT and Bluetooth positioning base stations, working together to continuously collect data on the elderly’s behavior, physiology, location, and environmental status.

[0009] S2. Scene pattern self-recognition and switching based on edge layer judgment:

[0010] The edge layer system switches between two system modes based on the system time and the intensity of elderly activity fed back by the camera terminal. The two system modes are a daytime high-complexity judgment mode and a nighttime low-latency response mode.

[0011] S3. Edge layer early warning processing methods based on different modes:

[0012] In the daytime high-complexity analysis mode, the system instructs ordinary cameras to perform low-resolution periodic scanning of the monitored area, quickly identify all elderly people in the area through a lightweight algorithm, and evaluate the elderly people using a weighted dynamic risk calculation model to screen out the target with the highest global risk coefficient. Then, the camera focuses on and continuously tracks the target at high resolution, concentrating limited high-computing resources on these high-risk targets for key analysis, while maintaining basic monitoring of low-risk targets.

[0013] In the low-latency response mode at night, when the mattress pressure sensor detects data fluctuations, the system immediately instructs the ordinary camera to switch to night mode and perform continuous monitoring.

[0014] S4. Steps for dispatching tasks based on hierarchical response:

[0015] In particular, based on the key judgment results of the daytime high-complexity judgment mode in S3, the system automatically generates response instructions of different levels; for high-risk warnings, the system dispatches the intervention task to the nearest caregiver terminal in real time based on the indoor Bluetooth positioning information; in the nighttime low-latency response mode, when the system detects the highest priority task generated after the elderly are ready to get out of bed, it will immediately send the details of the task to the caregiver on duty.

[0016] S5. Execution Feedback and Model Optimization:

[0017] After performing the task, the caregiver sends the results back to the edge layer via the terminal.

[0018] Preferably, the specific process steps in S1 are as follows:

[0019] S11. Identity entry and dynamic health record binding:

[0020] The wristband records and archives the elderly person's detailed information and records the elderly person's recent health information into the wristband, creating a real-time updated personal dynamic risk profile for each elderly person. At the same time, the Bluetooth device periodically broadcasts its unique Bluetooth identifier.

[0021] S12. Multi-dimensional perception:

[0022] The terminal device also includes a regular surveillance camera and a pressure sensor. The regular surveillance camera is deployed in key areas and performs loop monitoring in a low frame rate mode, only outputting video streams without the need for built-in complex algorithms. The pressure sensor is placed in the wheelchair or bed to confirm the elderly person's standing up when using the relevant equipment by detecting changes in pressure data.

[0023] S13. Summary of Environmental Data:

[0024] The terminal device also includes an IoT and Bluetooth positioning base station, which is used to ensure the connectivity between the edge layer system and the terminal layer device, ensure the low cost and connectivity of the system, and transmit multi-dimensional environmental data and Bluetooth positioning information to the edge layer for processing in real time.

[0025] Preferably, in step S1, the human body in the ordinary camera image is always an anonymous pixel block. The system uses Bluetooth positioning technology to spatially match the real-time location of the elderly person in the image with the reported location of the Bluetooth bracelet ID, thereby associating the behavior with the bracelet, and then associating the bracelet information with the elderly person's digital profile. In step S11, the elderly person's digital profile image is collected by the caregiver and uploaded to the edge layer system for updating. The current digital profile is then comprehensively converted and evaluated into a corresponding health level coefficient and archived.

[0026] Preferably, the specific process steps in S2 are as follows:

[0027] S21. Pattern Partitioning and Initialization:

[0028] The system defines and has two built-in working modes:

[0029] Daytime High Complexity Assessment Mode: In this mode, the system enables a dynamic risk coefficient-based approach. The computing power focus and analysis process involves multi-level and multi-dimensional data analysis to achieve low-power, high-precision early warning and optimize computing resources.

[0030] Nighttime low-latency response mode: In this mode, the system responds instantly to key behavioral signals to achieve proactive intervention with minimal latency, preventing falls at the source;

[0031] S22. Timing-based primary mode switching:

[0032] The edge layer system has a built-in real-time clock module. The system first performs a preliminary switch between two default modes based on the current time.

[0033] S23. Pattern fine-tuning based on the activity intensity of the elderly:

[0034] To adapt to changes in the elderly's daily routines during different seasons or on special days, the system introduces an elderly activity intensity value A as a feedback variable to dynamically fine-tune time-based mode switching. The elderly activity intensity value A is calculated periodically by the edge layer system, and its calculation formula is as follows:

[0035] ,in

[0036] This refers to all cameras that are "active" within the current period's field of view.

[0037] This represents the total number of elderly people currently in the hospital who are wearing wristbands;

[0038] This represents the number of monitoring areas that the system has determined to be "active" within the current period.

[0039] This represents the total number of areas within the hospital that are monitored by cameras.

[0040] and These are the weighting coefficients, and This is used to adjust the contribution of "personnel activity ratio" and "regional activity ratio" in the overall judgment;

[0041] S24. Adaptive switching judgment:

[0042] The system will compare the calculated activity intensity value A with the preset threshold. and The comparisons are made, and the final pattern decision is made according to the following rules:

[0043] If the system is currently in daytime mode, and Continue to exceed If the elderly group has entered a resting state, the system will automatically switch to night mode.

[0044] If the system is currently in night mode, and Continue to exceed If the elderly population is found to be generally active, the system will automatically switch to daytime mode.

[0045] If the above conditions are not met, the system will maintain the current mode based on the initial time judgment.

[0046] Preferably, in step S3, the process of the daytime high-complexity analysis mode is as follows:

[0047] S31. Initial state: All ordinary cameras are in low-resolution patrol mode for global scanning and ordinary monitoring;

[0048] S32. Trigger: A regular camera detects a human-shaped target within its visual area in patrol mode;

[0049] S33. Identity Association: The system reads the list of all Bluetooth wristband information within the physical area covered by the camera at this time, and binds each wristband information to the corresponding visual target through spatial position overlap;

[0050] S34. Risk Calculation: The system quickly queries the health coefficient of each elderly person and substitutes it into the risk coefficient formula to calculate the current comprehensive risk coefficient of each elderly person. ;

[0051] S35. Preliminary Risk Screening and Ranking: The marginal layer system assesses the comprehensive risk coefficient of all elderly individuals in S34. Sort by comprehensive risk coefficient The elderly individuals were marked as Level L1 (attention level) and will be continuously monitored.

[0052] S36. Further risk screening and decision-making: The edge layer issues instructions to switch ordinary cameras to high frame rate and high resolution mode, and starts a complex fall detection algorithm to perform refined behavior analysis only on L1 attention level targets to confirm their status. When an elderly person is determined to be engaging in high-risk activities in a high-risk area, and the fall detection algorithm also detects an anomaly, the elderly person is marked as L2 warning level; when the camera determines that an elderly person has suddenly fallen at a special time, the elderly person is marked as L3 alarm level.

[0053] Preferably, in step S34, the comprehensive risk coefficient The calculation formula is:

[0054] ,in,

[0055] The basic risk is a fixed value derived from the elderly person's static records, which is preset by the administrator based on factors such as age, medical history, and medication.

[0056] The abnormality score is calculated to determine the degree of deviation between the current behavior and the individual's historical normal behavior patterns.

[0057] This serves as the baseline for behavioral risk, a predefined constant representing the degree of danger inherent in the behavior itself. As the environmental risk baseline, a predefined constant is used to assess the inherent risk at the current location;

[0058] This is the environmental coefficient, a multiplier adjusted based on real-time environmental conditions.

[0059] Personnel escort attenuation, negative gain term. Where d is the straight-line distance to the nearest caregiver in meters, and k is the attenuation coefficient. The closer the caregiver is, the larger this value is, and the lower the total risk R is.

[0060] This is the clustering attenuation factor, a negative gain term. , where n is the number of other elderly people within 3 meters of the target elderly person, and m is the attenuation coefficient for each person.

[0061] Preferably, in step S3, under the low-latency response mode at night, distributed pressure sensors deployed under the mattress continuously monitor the elderly person's bed-keeping status. When a significant change in pressure distribution is detected, it is determined to be a "primary bed-leaving trigger event," and the event signal is immediately reported to the edge layer system. After receiving the primary trigger signal, the edge layer system instructs the ordinary camera facing the bed to switch to low-light mode at night and activates a regional dynamic detection algorithm to scan the area around the bed to visually verify whether there are any signs of the elderly person getting up. After the system receives the visual verification from the camera, the system abandons the complex dynamic risk calculation model and directly generates a highest priority task defined as "nighttime bed-leaving assistance."

[0062] Preferably, in step S4, during the daytime mode, for the L2 warning level in S36, the system generates an assistance task and sends it to the nearest caregiver displayed via Bluetooth signal, eliminating the risk of fall with the caregiver's assistance; for the L3 alarm level, the system immediately generates an emergency rescue task and sends it to the nearest caregiver displayed via Bluetooth signal, and automatically triggers the highest level response such as notifying family members and unlocking the door; during the nighttime mode, for the highest priority task, the system forwards the task to the dedicated terminal of the night shift caregiver and the terminal of the patrol caregiver closest to the room, triggering sound, light and vibration reminders to ensure that the information is received immediately.

[0063] Preferably, the system includes:

[0064] Terminal layer module: The terminal layer module includes a Bluetooth bracelet for the elderly, a caregiver terminal, a regular camera, a pressure sensor, and an IoT and Bluetooth positioning base station, which are used for the collection of raw data and the execution of instructions, and are deployed in nursing homes;

[0065] Edge layer module: The edge layer module is deployed on the local server of the nursing home. It is used to receive data from the terminal layer module, as well as most of the real-time calculation, decision-making and scheduling. It is the core of ensuring low latency and privacy security of the system. Its core is the dynamic risk assessment and scheduling engine.

[0066] Cloud module: The cloud module is responsible for macro-level analysis, optimization and remote interaction. It does not participate in real-time decision-making and ensures that the local system continues to operate normally even if the external network is interrupted.

[0067] The technical effects and advantages of this invention are as follows:

[0068] 1. This multi-dimensional detection-based fall warning method and system for the elderly significantly reduces system deployment and maintenance costs while improving reliability. The invention innovatively employs a combination of low-cost, low-power terminal devices such as commercially available ordinary cameras, Bluetooth beacons, and pressure sensors, replacing the expensive high-precision radar, smart cameras, and biosensors used in existing technologies. Through a "cloud-edge-device" architecture and a "dynamic computing power focusing" method, core computing power requirements are concentrated on local edge servers. Daily processing only requires handling low-resolution video streams, greatly reducing reliance on server hardware, network bandwidth, and cloud computing power. This results in lower overall system power consumption, less hardware damage, and controllable costs while maintaining high reliability, significantly enhancing the product's market competitiveness and feasibility for widespread adoption.

[0069] 2. This multi-dimensional detection-based fall warning method and system for the elderly significantly improves the accuracy and response speed of warnings while greatly protecting the privacy of the elderly. The invention utilizes a "scenario-adaptive" mechanism to intelligently switch between two logics: a daytime "global scan - key analysis" approach and a nighttime "instant trigger - rapid response" approach, optimizing system resource utilization. During the day, cross-validation using a multi-dimensional weighted risk model effectively filters false alarms caused by daily activities. At night, the collaboration between mattress pressure sensors and visual verification achieves extremely low latency from the elderly leaving the bed to the caregiver's response, preventing falls at the source. Simultaneously, the visual algorithm consistently processes anonymized "human-shaped pixel blocks," associating identities through Bluetooth IDs, completely avoiding privacy violations associated with facial recognition and making it more acceptable to the elderly.

[0070] 3. This multi-dimensional detection-based fall warning method and system for the elderly represents a paradigm shift from "passive alarm" to "active intervention," constructing a closed-loop optimization system of human-machine collaboration. The ultimate goal of this invention is not merely to more accurately detect falls, but to ultimately "prevent" falls from occurring. The system uses Bluetooth positioning technology to precisely dispatch warning tasks to the nearest caregiver's terminal, achieving a seamless connection between "risk detection – task allocation – personnel execution," making humans a crucial link in preventing risks. Especially in night mode, intervention and assistance can be provided before a risk occurs (when the elderly person has just gotten out of bed). Furthermore, by collecting feedback from caregivers, the system continuously calibrates and optimizes the risk model parameters, forming a self-learning closed loop that becomes increasingly accurate with use, enabling the entire system's warning capabilities to continuously evolve. Attached Figure Description

[0071] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0072] Figure 1 This is a system architecture diagram of the present invention;

[0073] Figure 2 This is a schematic diagram of the algorithm recognition process of the present invention;

[0074] Figure 3 This is the daytime dynamic early warning logic diagram of the present invention; Detailed Implementation

[0075] 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.

[0076] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0077] This invention discloses a method and system for early warning of falls in the elderly based on multi-dimensional detection, according to the appendix. Figure 1 As shown, it includes:

[0078] S1. Multi-dimensional data perception based on low-power terminal layer:

[0079] This step involves the elderly wearing a Bluetooth beacon terminal wristband, the caregiver wearing a corresponding Bluetooth terminal, and terminal devices deployed in the environment, including ordinary cameras, pressure sensors, and IoT and Bluetooth positioning base stations, working together to continuously collect data on the elderly’s behavior, physiology, location, and environmental status in a low-power and low-cost manner.

[0080] S2. Scene pattern self-recognition and switching based on edge layer judgment:

[0081] The edge layer system switches between two system modes based on the system time and the intensity of elderly activity fed back by the camera terminal. These two system modes are a daytime high-complexity judgment mode and a nighttime low-latency response mode, which are divided according to the different activities of the elderly at different times of the day.

[0082] S3. Edge layer early warning processing methods based on different modes:

[0083] In the daytime high-complexity analysis mode, the system instructs ordinary cameras to perform low-resolution periodic scans of the monitored area, quickly identify all elderly people in the area through a lightweight algorithm, and evaluate the elderly people using a weighted dynamic risk calculation model to screen out the target with the highest global risk coefficient. Then, the camera focuses on and continuously tracks the target at high resolution, concentrating limited high-computing power resources on these high-risk targets for key analysis, while maintaining basic monitoring of low-risk targets.

[0084] In the low-latency response mode at night, when the mattress pressure sensor detects data fluctuations, the system immediately instructs the ordinary camera to switch to night mode and perform continuous monitoring.

[0085] S4. Steps for dispatching tasks based on hierarchical response:

[0086] Based on the key assessment results of the daytime high-complexity assessment mode in S3, the system automatically generates response instructions of different levels. For high-risk warnings, the system uses indoor Bluetooth positioning information to send intervention tasks to the nearest caregiver terminal in real time, thereby achieving seamless connection and rapid interruption from "system alarm" to "personnel execution". In the nighttime low-latency response mode, when the system detects the highest priority task generated after the elderly are about to get out of bed, it will immediately send the details of the task to the on-duty caregiver to ensure that the risk of falls when the elderly get up at night is zero.

[0087] S5. Execution Feedback and Model Optimization:

[0088] After performing their tasks, caregivers provide feedback via their terminals. The edge layer system collects this feedback data to continuously calibrate and optimize the parameters of the individual dynamic risk model in S3, forming a closed-loop learning system that becomes more accurate with use.

[0089] Furthermore, the specific process steps in S1 are as follows:

[0090] S11. Identity entry and dynamic health record binding:

[0091] The bracelet records and archives the elderly person's detailed information and records the elderly person's recent health information into the bracelet, creating a real-time updated personal dynamic risk profile for each elderly person. At the same time, the Bluetooth device periodically broadcasts its unique Bluetooth identifier.

[0092] S12. Multi-dimensional perception:

[0093] The terminal equipment also includes ordinary surveillance cameras and pressure sensors. The ordinary surveillance cameras are deployed in key areas and perform loop monitoring in low frame rate mode, only outputting video streams without the need for built-in complex algorithms. The pressure sensors are placed in wheelchairs and beds to confirm when the elderly get up when using the relevant equipment by detecting changes in pressure data.

[0094] S13. Summary of Environmental Data:

[0095] Terminal devices also include IoT and Bluetooth positioning base stations. IoT and Bluetooth positioning base stations are used to ensure the connectivity between the edge layer system and the terminal layer devices, ensure the low cost and connectivity of the system, and transmit multi-dimensional environmental data and Bluetooth positioning information to the edge layer for processing in real time.

[0096] Furthermore, in S1, the human body in the ordinary camera footage is always an anonymous pixel block. The camera never needs to know "who that face is." It only needs to identify "a human-shaped target at the stairwell." The system uses the Bluetooth signal strength in the target area to determine the strongest Bluetooth ID in the area, thus associating the "human-shaped target" with the "bracelet ID" and retrieving the elderly person's information. This fundamentally eliminates privacy leaks and significantly reduces computing power consumption. The system uses Bluetooth positioning technology to spatially match the elderly person's real-time location in the footage with the reported location of the Bluetooth bracelet ID, thereby associating behavior with the bracelet and then linking the bracelet information to the elderly person's digital profile. In S11, the elderly person's digital profile image is collected by the caregiver and uploaded to the edge layer system for updating. The current digital profile is then comprehensively converted and evaluated into a corresponding health level coefficient and archived.

[0097] Furthermore, the specific process steps in S2 are as follows:

[0098] S21. Pattern Partitioning and Initialization:

[0099] The system defines and has two built-in working modes:

[0100] Daytime High Complexity Assessment Mode: The default working hours are from 06:00 to 20:00; in this mode, the system enables dynamic risk coefficient-based assessment. The computing power focus and analysis process involves multi-level and multi-dimensional data analysis to achieve low-power, high-precision early warning and optimize computing resources.

[0101] Nighttime low-latency response mode: The default working hours are from 20:00 to 06:00 the next day; in this mode, the system will trigger an immediate response to key behavioral signals to achieve proactive intervention with minimal latency and prevent falls from the source.

[0102] S22. Timing-based primary mode switching:

[0103] The edge layer system has a built-in real-time clock module. The system first performs a preliminary switch between two default modes based on the current time.

[0104] S23. Pattern fine-tuning based on the activity intensity of the elderly:

[0105] To adapt to changes in the elderly's daily routines during different seasons or on special days, the system introduces the elderly's activity intensity value A as a feedback variable to dynamically fine-tune the time-based mode switching. The elderly's activity intensity value A is calculated periodically by the edge layer system, and its calculation formula is as follows:

[0106] ,in

[0107] This refers to the number of elderly people identified as being in an "active state," such as walking or engaging in vigorous movement, within the current period's camera field of view.

[0108] This represents the total number of elderly people currently in the hospital who are wearing wristbands;

[0109] This represents the number of monitored areas that the system has determined to be "active" within the current period. Additionally, an area is considered "active" if, within that period, the number of elderly people identified as active in that area is greater than or equal to 1.

[0110] This represents the total number of areas within the hospital that are monitored by cameras.

[0111] and These are the weighting coefficients, and This is used to adjust the contribution of "personnel activity ratio" and "regional activity ratio" in the overall judgment;

[0112] S24. Adaptive switching judgment:

[0113] The system will compare the calculated activity intensity value A with the preset threshold. and The comparisons are made, and the final pattern decision is made according to the following rules:

[0114] If the system is currently in daytime mode, and Continue to exceed If the elderly group has entered a resting state, the system will automatically switch to night mode.

[0115] If the system is currently in night mode, and Continue to exceed If the elderly population is found to be generally active, the system will automatically switch to daytime mode.

[0116] If the above conditions are not met, the system will maintain the current mode based on the initial time judgment.

[0117] According to the appendix Figure 2 and Figure 3 As shown, the process in S3 under the daytime high-complexity analysis mode, which is specifically disclosed, is as follows:

[0118] S31. Initial state: All ordinary cameras are in low-resolution patrol mode for global scanning and general monitoring;

[0119] S32. Trigger: A regular camera detects a human-shaped target within its visual area in patrol mode;

[0120] S33. Identity Association: The system reads the list of all Bluetooth wristband information within the physical area covered by the camera at this time, and binds each wristband information to the corresponding visual target through spatial position overlap;

[0121] S34. Risk Calculation: The system quickly queries the health coefficient of each elderly person and substitutes it into the risk coefficient formula to calculate the current comprehensive risk coefficient of each elderly person. ;

[0122] S35. Preliminary Risk Screening and Ranking: The marginal layer system assesses the comprehensive risk coefficient of all elderly individuals in S34. Sort by comprehensive risk coefficient The elderly individuals were marked as Level L1 (attention level) and will be continuously monitored.

[0123] S36. Further Risk Screening and Decision-Making: The edge layer issues instructions to switch ordinary cameras to high frame rate and high resolution mode, and activates complex fall detection algorithms (such as the ST-GCN algorithm based on pose estimation). Only L1 level of interest targets are subjected to refined behavioral analysis to confirm their status. When an elderly person is determined to be engaging in high-risk activities in a high-risk area, and the fall detection algorithm also detects an anomaly, the elderly person is marked as L2 warning level. When the camera determines that an elderly person has suddenly fallen at a special time, the elderly person is marked as L3 alarm level.

[0124] In existing technologies, high-precision AI analysis of all camera video streams is typically performed around the clock and across the entire screen for real-time early warning, resulting in significant waste of computing power. This method creatively designs a dynamic computing power allocation mode that shifts from "global low-power inspection to local high-risk focusing." 90% of the time, the algorithm processes low-resolution video streams and Bluetooth data packets, resulting in extremely low power consumption. Only when the risk factor exceeds a threshold does the system act like a searchlight, instructing a specific camera to perform high-precision analysis of a specific area within a specific timeframe. This not only saves computing power but also directly reduces hardware costs and network bandwidth requirements.

[0125] In step S34, the comprehensive risk coefficient is calculated. The calculation formula is:

[0126] ,in,

[0127] The basic risk is a fixed value derived from the elderly person's static records, preset by the administrator based on factors such as age, medical history, and medication, and its range is from 0.0 to 3.0;

[0128] The abnormality score is calculated as the degree of deviation between the current behavior and the individual's historical normal behavior pattern, ranging from 0.0 to 2.0.

[0129] As the behavioral risk baseline, it is a predefined constant representing the degree of danger of the behavior itself, where sitting still = 0.1, walking = 0.8, going up and down stairs = 1.5, and bending over = 1.2;

[0130] As the environmental risk baseline, a predefined constant is used to determine the inherent risk at the current location, where room = 0.5, corridor = 1.0, toilet = 1.8, and stairwell = 2.0.

[0131] This is the environmental coefficient, a multiplier adjusted according to real-time environmental conditions, where nighttime = 1.5, wet ground = 2.0, and normal daytime = 1.0;

[0132] Personnel escort attenuation, negative gain term. Where d is the straight-line distance to the nearest caregiver in meters, and k is the attenuation coefficient, usually set to 1.0. The closer the caregiver is, the larger this value is, and the lower the total risk R is.

[0133] This is the clustering attenuation factor, a negative gain term. , where n is the number of other elderly people within 3 meters of the target elderly person, and m is the attenuation coefficient per person, usually set to 0.3.

[0134] According to the appendix Figure 2As shown, in step S3, specifically in the low-latency response mode at night, distributed pressure sensors deployed under the mattress continuously monitor the elderly person's bed-keeping status. When a significant change in pressure distribution is detected, it is determined to be a "primary bed-leaving trigger event," and the event signal is immediately reported to the edge layer system. After receiving the primary trigger signal, the edge layer system instructs the ordinary camera facing the bed to switch to low-light mode at night and activates the regional dynamic detection algorithm to scan the area around the bed to visually verify whether there are any signs of the elderly person getting up. After the system receives the visual verification from the camera, the system abandons the complex dynamic risk calculation model and directly generates a highest priority task defined as "nighttime bed-leaving assistance."

[0135] According to the appendix Figure 2 As shown, in step S3, specifically in the low-latency response mode at night, distributed pressure sensors deployed under the mattress continuously monitor the elderly person's bed-keeping status. When a significant change in pressure distribution is detected, it is determined to be a "primary bed-leaving trigger event," and the event signal is immediately reported to the edge layer system. After receiving the primary trigger signal, the edge layer system instructs the ordinary camera facing the bed to switch to low-light mode at night and activates the regional dynamic detection algorithm to scan the area around the bed to visually verify whether there are any signs of the elderly person getting up. After the system receives the visual verification from the camera, the system abandons the complex dynamic risk calculation model and directly generates a highest priority task defined as "nighttime bed-leaving assistance."

[0136] According to the appendix Figures 2 to 3 As shown, it is particularly important to emphasize that in step S4, during the daytime mode, for the L2 warning level in S36, the system generates an assistance task and dispatches it to the nearest caregiver displayed via Bluetooth signal, eliminating the risk of falls with the caregiver's assistance; for the L3 alarm level, the system immediately generates an emergency rescue task and dispatches it to the nearest caregiver displayed via Bluetooth signal, and automatically triggers the highest level response, such as notifying family members and unlocking the room door; during the nighttime mode, for the highest priority task, the system forwards and pushes the task to the dedicated terminal of the night shift caregiver and the terminal of the patrol caregiver closest to the room, triggering sound, light, and vibration alerts to ensure that the information is received immediately.

[0137] According to the appendix Figure 1 As shown, it is important to emphasize that the system includes:

[0138] Terminal layer module: The terminal layer module includes a Bluetooth bracelet for the elderly, a caregiver terminal, a regular camera, a pressure sensor, and an IoT and Bluetooth positioning base station, which are used for the collection of raw data and the execution of instructions. Deployed in nursing homes, the terminal layer module completely eliminates expensive equipment such as high-precision radar and smart cameras, and only uses ordinary commercial-grade hardware, which greatly reduces deployment and maintenance costs.

[0139] Edge layer module: The edge layer module is deployed on the local server of the nursing home. It is used to receive data from the terminal layer module, as well as the majority of real-time computing, decision-making and scheduling. It is the core of ensuring low latency and privacy security of the system. Its core is the dynamic risk assessment and scheduling engine.

[0140] Cloud module: The cloud module is responsible for macro-level analysis, optimization and remote interaction. It does not participate in real-time decision-making, ensuring that the local system continues to operate normally even if the external network is interrupted. It notifies family members when an alarm occurs, enhancing the emotional care value of the system.

[0141] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A fall warning method for the elderly based on multi-dimensional detection, characterized in that, include: S1. Multi-dimensional data perception based on low-power terminal layer: This step involves the elderly wearing a Bluetooth beacon terminal wristband, the caregiver wearing a corresponding Bluetooth terminal, and terminal devices deployed in the environment, including ordinary cameras, pressure sensors, and IoT and Bluetooth positioning base stations, working together to continuously collect data on the elderly’s behavior, physiology, location, and environmental status. S2. Scene pattern self-recognition and switching based on edge layer judgment: The edge layer system switches between two system modes based on the system time and the intensity of elderly activity fed back by the camera terminal. The two system modes are a daytime high-complexity judgment mode and a nighttime low-latency response mode. The specific process steps in S2 are as follows: S21. Pattern Partitioning and Initialization: The system defines and has two built-in working modes: Daytime High Complexity Assessment Mode: In this mode, the system enables a dynamic risk coefficient-based approach. The computing power focus and analysis process involves multi-level and multi-dimensional data analysis to achieve low-power, high-precision early warning and optimize computing resources. Nighttime low-latency response mode: In this mode, the system responds instantly to key behavioral signals to achieve proactive intervention with minimal latency, preventing falls at the source; S22. Timing-based primary mode switching: The edge layer system has a built-in real-time clock module. The system first performs a preliminary switch between two default modes based on the current time. S23. Pattern fine-tuning based on the activity intensity of the elderly: To adapt to changes in the elderly's daily routines during different seasons or on special days, the system introduces an elderly activity intensity value A as a feedback variable to dynamically fine-tune time-based mode switching. The elderly activity intensity value A is calculated periodically by the edge layer system, and its calculation formula is as follows: wherein the number of elderly people in "active state" recognized in the field of view of all cameras for the current period; Total number of elderly currently in the hospital and wearing a bracelet; the number of monitored regions that the system determines to be "active" for the current cycle; Total number of areas to be monitored by cameras deployed within the hospital; and These are the weighting coefficients, and This is used to adjust the contribution of "personnel activity ratio" and "regional activity ratio" in the overall judgment; S24. Adaptive switching judgment: The system compares the calculated activity intensity value A with a pre-set threshold value and and follows the following rules for the final pattern decision: If the system is currently in daytime mode, and Continue to exceed If the elderly group has entered a resting state, the system will automatically switch to night mode. If the system is currently in night mode, and Continue to exceed If the elderly population is found to be generally active, the system will automatically switch to daytime mode. If the above conditions are not met, the system will maintain the current mode based on the initial time judgment. S3. Edge layer early warning processing methods based on different modes: In the daytime high-complexity analysis mode, the system instructs ordinary cameras to perform low-resolution periodic scanning of the monitored area, quickly identify all elderly people in the area through a lightweight algorithm, and evaluate the elderly people using a weighted dynamic risk calculation model to screen out the target with the highest global risk coefficient. Then, the camera focuses on and continuously tracks the target at high resolution, concentrating limited high-computing resources on these high-risk targets for key analysis, while maintaining basic monitoring of low-risk targets. In the low-latency response mode at night, when the mattress pressure sensor detects data fluctuations, the system immediately instructs the ordinary camera to switch to night mode and perform continuous monitoring. S4. Steps for dispatching tasks based on hierarchical response: In particular, based on the key judgment results of the daytime high-complexity judgment mode in S3, the system automatically generates response instructions of different levels; for high-risk warnings, the system dispatches the intervention task to the nearest caregiver terminal in real time based on the indoor Bluetooth positioning information; in the nighttime low-latency response mode, when the system detects the highest priority task generated after the elderly are ready to get out of bed, it will immediately send the details of the task to the caregiver on duty. S5. Execution Feedback and Model Optimization: After performing the task, the caregiver sends the results back to the edge layer via the terminal. 2.The fall early warning method for the elderly based on multi-dimension detection according to claim 1, characterized in that, The specific process steps in S1 are as follows: S11. Identity entry and dynamic health record binding: The wristband records and archives the elderly person's detailed information and records the elderly person's recent health information into the wristband, creating a real-time updated personal dynamic risk profile for each elderly person. At the same time, the Bluetooth device periodically broadcasts its unique Bluetooth identifier. S12. Multi-dimensional perception: The terminal device also includes a regular surveillance camera and a pressure sensor. The regular surveillance camera is deployed in key areas and performs loop monitoring in a low frame rate mode, only outputting video streams without the need for built-in complex algorithms. The pressure sensor is placed in the wheelchair or bed to confirm the elderly person's standing up when using the relevant equipment by detecting changes in pressure data. S13. Summary of Environmental Data: The terminal device also includes an IoT and Bluetooth positioning base station, which is used to ensure the connectivity between the edge layer system and the terminal layer device, ensure the low cost and connectivity of the system, and transmit multi-dimensional environmental data and Bluetooth positioning information to the edge layer for processing in real time. 3.The old people fall early warning method based on multi-dimension detection according to claim 2, characterized in that, In step S1, the human body in the ordinary camera image is always an anonymous pixel block. The system uses Bluetooth positioning technology to spatially match the real-time location of the elderly in the image with the reported location of the Bluetooth bracelet ID, thereby associating the behavior with the bracelet, and then associating the elderly's digital profile with the bracelet information. In step S11, the elderly's digital profile image is collected by the caregiver and uploaded to the edge layer system for updating. The current digital profile is then comprehensively converted and evaluated into a corresponding health level coefficient and archived. 4.The fall early warning method for the elderly based on multi-dimension detection of claim 1, wherein, In S3, the process of the daytime high-complexity analysis mode is as follows: S31. Initial state: All ordinary cameras are in low-resolution patrol mode for global scanning and ordinary monitoring; S32. Trigger: A regular camera detects a human-shaped target within its visual area in patrol mode; S33. Identity Association: The system reads the list of all Bluetooth wristband information within the physical area covered by the camera at this time, and binds each wristband information to the corresponding visual target through spatial position overlap; S34. Risk Calculation: The system quickly queries the health coefficient of each elderly person and substitutes it into the risk coefficient formula to calculate the current comprehensive risk coefficient of each elderly person. ; S35. Preliminary Risk Screening and Ranking: The marginal layer system assesses the comprehensive risk coefficient of all elderly individuals in S34. Sort by comprehensive risk coefficient The elderly individuals were marked as Level L1 (attention level) and will be continuously monitored. S36. Further risk screening and decision-making: The edge layer issues instructions to switch the ordinary camera to a high frame rate and high resolution mode, and starts a complex fall detection algorithm to perform refined behavior analysis only on L1 attention level targets to confirm their status. When it is determined that an elderly person is engaging in high-risk activities in a high-risk area and the fall detection algorithm also detects an abnormality, the elderly person is marked as L2 warning level. When the camera determines that an elderly person has suddenly fallen at a specific time, the elderly person is marked as an L3 alarm level. 5.The fall early warning method for the elderly based on multi-dimension detection of claim 4, characterized in that, In the step S34, the comprehensive risk coefficient The calculation formula is: wherein, The basic risk is a fixed value from the static profile of the elderly, which is preset by the administrator according to factors such as age, medical history, medication, etc. abnormality degree, the deviation degree of the current behavior from the historical normal behavior pattern of the person is calculated; a behavioral risk base, a predefined constant for the degree of danger of the behavior itself; an environmental risk base, a constant predefined for the inherent risk of the current location; is an environmental coefficient, a multiplier adjusted according to real-time environmental conditions; Personnel escort attenuation, negative gain term, among which d is the straight-line distance to the nearest caregiver in meters, and k is the attenuation coefficient. The closer the caregiver is, the larger this value is, and the lower the total risk R is. Here, is the clustering attenuation factor, a negative gain term, where... , where n is the number of other elderly people within 3 meters of the target elderly person, and m is the attenuation coefficient for each person. 6.The old people fall early warning method based on multi-dimension detection according to claim 5, characterized in that, In step S3, under the low-latency response mode at night, distributed pressure sensors deployed under the mattress continuously monitor the elderly person's bedside status. When a significant change in pressure distribution is detected, it is determined as a "primary bed-leaving trigger event," and the event signal is immediately reported to the edge layer system. After receiving the primary trigger signal, the edge layer system instructs the ordinary camera facing the bed to switch to low-light mode at night and activates a regional dynamic detection algorithm to scan the area around the bed to visually verify whether there are any signs of the elderly person getting up. After the system receives the visual verification from the camera, the system abandons the complex dynamic risk calculation model and directly generates a highest priority task defined as "nighttime bed-leaving assistance." 7.The old people fall early warning method based on multi-dimension detection according to claim 6, characterized in that, In step S4, during the daytime, for the L2 warning level in S36, the system generates an assistance task and sends it to the nearest caregiver indicated by the Bluetooth signal, eliminating the risk of falls with the caregiver's assistance; for the L3 alarm level, the system immediately generates an emergency rescue task and sends it to the nearest caregiver indicated by the Bluetooth signal, and automatically triggers the highest level response, such as notifying family members and unlocking the door; during the nighttime, for the highest priority task, the system forwards the task to the dedicated terminal of the night shift caregiver and the terminal of the patrol caregiver closest to the room where the highest priority task is located, triggering sound, light, and vibration alerts to ensure that the information is received immediately.

8. The system of the elderly fall early warning method based on multi-dimension detection according to any one of claims 1-7, characterized in that, The system includes: Terminal layer module: The terminal layer module includes a Bluetooth bracelet for the elderly, a caregiver terminal, a regular camera, a pressure sensor, and an IoT and Bluetooth positioning base station, which are used for the collection of raw data and the execution of instructions, and are deployed in nursing homes; Edge layer module: The edge layer module is deployed on the local server of the nursing home. It is used to receive data from the terminal layer module, as well as most of the real-time calculation, decision-making and scheduling. It is the core of ensuring low latency and privacy security of the system. Its core is the dynamic risk assessment and scheduling engine. Cloud module: The cloud module is responsible for macro-level analysis, optimization and remote interaction. It does not participate in real-time decision-making and ensures that the local system continues to operate normally even if the external network is interrupted.

Images