Fall early warning system for self-adaptive monitoring of gaits of old people

Abstract

The invention relates to the technical field of intelligent health monitoring, in particular to a fall early warning system for self-adaptive monitoring of gaits of old people, which comprises the following steps: S1, data acquisition; s2, data processing; s3, data judgment: calculating a gait parameter deviation degree based on a dynamic baseline, and combining action type recognition and a secondary confirmation mechanism; s4, outputting a result: executing privacy protection type hierarchical linkage early warning according to the risk level, and receiving feedback data to reversely optimize the system model; the task boundary of each link is defined, so that the method has strong engineering realizability, system integration and landing are facilitated, dynamic personalized gait baseline construction and feedback optimization are brought into a core process, the limitation of traditional'fixed threshold 'monitoring is broken through, continuous adjustment can be performed along with the change of the physical state of the elderly, the reliability of long-term monitoring is improved, and the method is suitable for popularization and application. The whole link from data acquisition to model optimization is covered, the integrity of the monitoring process is guaranteed, the core target of'falling early warning of the home elderly 'is focused, and functional redundancy is avoided.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent health monitoring, and in particular to a fall warning system for adaptive gait monitoring in the elderly. Background Technology

[0002] As home-based elderly care is the mainstream model, the risk of falls among elderly people living alone is a core pain point in home-based elderly care services. Falls can cause injuries of varying degrees to the elderly, such as bruises, fractures, and head injuries, and in severe cases, they can even be life-threatening. Falls are closely related to the decline in the elderly's physiological functions (such as weakened muscle strength and poor coordination), the complexity of the home environment, and sudden physical conditions.

[0003] In existing technologies, fall detection systems for elderly people living at home primarily rely on image recognition and sensor data collection for monitoring. Some solutions employ multi-feature fusion algorithms to improve recognition accuracy. For example, invention patent application CN119848700A discloses a fall detection system and method for elderly people living at home based on feature fusion. This system combines gait parameters, environmental parameters, and cane usage parameters for risk assessment, attempting to address the insufficient recognition rate caused by single-feature monitoring. However, these systems still have significant shortcomings in practical applications and are difficult to meet the actual needs of home settings.

[0004] On the one hand, the complexity of the home environment leads to a persistently high risk of false alarms in the system. Common problems in home settings include sudden changes in lighting (such as backlighting on balconies or lights being on at night), furniture obstruction (such as sofas or dining tables obscuring key body parts), and clothing interference (such as heavy winter clothing obscuring joint points). Existing systems' posture estimation algorithms are mostly general designs and have not been optimized for these unique home-related interference factors. This makes the system prone to errors in identifying key points when recognizing the elderly's gait and posture due to environmental interference, leading to misjudging normal actions such as "bending over to pick up an object" or "squatting to rest" as signs of a fall or actual fall behavior. Frequent false alarms not only reduce the system's reliability but also cause fatigue among family members and community staff, causing them to ignore genuine risk warnings. Furthermore, existing systems lack adaptability to individual gait differences among the elderly, failing to consider special gait types such as the "flustered gait" of Parkinson's patients or the "limping gait" of arthritis patients, relying solely on a general gait baseline for risk assessment, further exacerbating the false alarm problem.

[0005] On the other hand, existing systems have significant shortcomings in functional completeness, and their early warning response mechanisms and privacy protection designs are ill-suited to the needs of home scenarios. Regarding early warning response, most existing systems only push warning information to a single receiver or lack clear hierarchical linkage logic. When children are busy and fail to check in time, early warning responses are easily delayed, making it impossible to provide timely assistance to elderly people who have fallen. Furthermore, the lack of a secondary confirmation mechanism after a fall makes it difficult to effectively distinguish between a genuine fall and a false trigger, further reducing the effectiveness of the early warning response. In terms of privacy protection, most systems relying on image recognition use video parameter acquisition and image recognition schemes that do not perform targeted anonymization processing on the collected video data, posing a risk of privacy leaks during the storage and transmission of raw images. Additionally, the lack of a privacy mode switching function means that video monitoring continues even when the elderly enter private areas such as bathrooms, severely violating the elderly's privacy protection needs and resulting in low system acceptance.

[0006] Therefore, there is an urgent need for an elderly gait adaptive monitoring and fall warning system that can adapt to the complex home environment, reduce the risk of false alarms, and improve the early warning response mechanism and privacy protection design, so as to solve the shortcomings of existing technologies and effectively improve the accuracy, reliability and practicality of fall warning for elderly people living at home. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a fall warning system with adaptive gait monitoring for the elderly.

[0008] The fall warning method for adaptive gait monitoring in the elderly according to the present invention includes the following steps: S1. Data Acquisition: Simultaneously collect core gait data, environmental interference data, and scene status data of elderly people living at home, and obtain a denoised gait data sequence after preprocessing; S2. Data processing: Perform environmental interference calibration on the gait data sequence, and construct and dynamically update a personalized gait baseline that adapts to the individual characteristics of the elderly. S3. Data Judgment: Based on the dynamic baseline, the deviation of gait parameters is calculated, and combined with the action type recognition and secondary confirmation mechanism, the fall risk level is determined in a graded manner. S4. Output Results: Implement privacy-protected tiered alerts based on risk levels, and receive feedback data to optimize the system model. Clearly define the task boundaries of each step to ensure the method has strong engineering feasibility, facilitates system integration and implementation, and incorporates the construction of dynamic personalized gait baselines and feedback optimization into the core process. This breaks through the limitations of traditional "fixed threshold" monitoring, allowing for continuous adjustment based on changes in the elderly's physical condition, improving the reliability of long-term monitoring, and covering the entire chain from data collection to model optimization. This ensures the integrity of the monitoring process while focusing on the core objective of "fall warning for elderly people living at home" and avoids functional redundancy.

[0009] Preferably, the specific implementation of data acquisition in step 1 includes: Core gait data: Step speed (v), cadence (f), stride length (s), body center of gravity offset angle (θ), and joint range of motion (α) are collected at 15Hz using millimeter-wave radar, forming the original gait data sequence D=[v1,f1,s1,θ1,α1;v2,f2,s2,θ2,α2;...;v n ,f n ,s n ,θ n ,α n ], where n is the amount of data collected; Environmental interference data: Light intensity L, ground humidity H, and distance to obstructions D are collected via environmental sensors. o This forms an interference data sequence E=[L1,H1,D] o1 ;L2,H2,D o2 ;...;L m H m D om ], where m is the number of data collections; Scene state data: The location and activity status of the elderly are collected through infrared sensors, forming state data S=[Area1,State1;Area2,State2;...;Area...]. k State k ], where k is the number of state records, Area represents the type of area where the elderly person is located, specifically a public area (living room, bedroom) or a private area (toilet, bathroom), and State represents the current activity state of the elderly person, specifically still, walking, bending over, squatting; Data preprocessing: The "Kalman filter + interference threshold removal" algorithm is used to remove abnormal data points caused by sudden changes in lighting, slippery ground, and occlusion in the original gait data D based on the interference data sequence E, resulting in a denoised gait data sequence D'. Core gait data, interference data, and scene state data are collected simultaneously. These three types of data corroborate each other, solving the judgment bias problem caused by insufficient data dimensions from a single sensor. Through the "Kalman filter + interference threshold removal" algorithm, abnormal data points caused by environmental interference such as sudden changes in lighting, slippery ground, and occlusion are accurately removed, improving the accuracy of subsequent data processing.

[0010] Preferably, the specific implementation of the data processing in step 2 includes: Environmental interference calibration: Based on the interference data sequence E, an environmental impact factor matrix M (M1+M2+M3=1, where M1 is the light impact factor, M2 is the humidity impact factor, and M3 is the shading impact factor) is constructed. The gait data sequence D' is then calibrated using the formula D''=D'×(1-ω×M) to obtain the environmentally adapted gait data D''. Here, ω is the environmental impact weight (0.1-0.3), which is dynamically adjusted according to the light intensity L and the ground humidity H. Personalized gait baseline construction: Using a 10-day initial calibration period, gait data D'' after environmental adaptation was collected. Combined with individual elderly patient information (age, height, disease type, gait characteristics), a clustering algorithm was used to classify gait types. The mean and standard deviation of each parameter were calculated to construct the initial personalized gait baseline B0=[v0,σᵥ;f0,σ_f;s0,σ s [;θ0,σθ;α0,σα];where v0 is the calibrated mean gait speed, f0 is the calibrated mean cadence, s0 is the calibrated mean stride length, θ0 is the calibrated mean body center of gravity offset angle, α0 is the calibrated mean joint range of motion, σᵥ is the standard deviation of gait speed, σ_f is the standard deviation of cadence, σ s σ_θ represents the standard deviation of stride length, σ_θ represents the standard deviation of the body center of gravity offset angle, and σ_α represents the standard deviation of joint range of motion. Personalized gait baseline dynamic update: During the routine monitoring phase, the initial baseline B0 is iteratively updated using a 7-day sliding window algorithm to obtain the dynamic baseline B. t (t represents the number of monitoring days); when changes in the elderly's physical condition are detected, the window size is shortened to 3 days; by constructing an environmental influencing factor matrix and calibrating gait data, interference from complex home environmental factors such as light, humidity, and shading is effectively offset, improving the stability of the method in different scenarios. Based on the individual information of the elderly (age, disease, gait characteristics), a dedicated baseline is constructed to solve the problem of insufficient adaptation of traditional general baselines to elderly people with special gait conditions such as Parkinson's disease and arthritis, thereby improving the accuracy of early warning. A dual-mode design of regular updates with a 7-day sliding window and rapid updates with a 3-day window is adopted to respond promptly to changes in the elderly's physical condition such as postoperative rehabilitation and disease progression, ensuring the timeliness of the baseline.

[0011] Preferably, the specific implementation of the data judgment in step 3 includes: Gait parameter deviation calculation: Define deviation δ x =|x_real-x t | / σ t (x_real is the real-time gait parameter value, x) t The mean and σ of the corresponding parameters in the dynamic personalized gait baseline t (where the standard deviations of the corresponding parameters in the dynamic personalized gait baseline are used to obtain the deviation set δ=[δᵥ,δ_f,δ s ,δθ,δα]; Action type recognition: Using a lightweight CNN+LSTM model, based on scene state data S and environment-adapted gait data D''_real, we can distinguish between "normal walking", "standing still", "bending over to pick up an object", "squatting and resting", and "falling due to imbalance". Fall risk level assessment: Mild risk is determined by the deviation δ of any one parameter. x ∈[1,1.5) and the action type is "normal walking" with a duration of ≥8 minutes; moderate risk is any one parameter deviation δ x ∈[1.5,2) or 2 or more parameters deviation δ x ∈[1,1.5) and the action type is "walking" or "bending over", with a duration of ≥5 minutes; severe risk is any one parameter deviation δ x ≥2 or the action type is identified as "unbalanced fall"; A secondary verification mechanism is implemented: When a moderate / severe risk level is determined, posture data is supplemented by a privacy-desensitized visual module in public areas, while millimeter-wave radar depth scanning is used in private areas. The risk authenticity is verified by combining changes in the human body's aspect ratio, eliminating false positives. By combining the dual judgment dimensions of "gait parameter deviation calculation" and "action type recognition," the one-sidedness of a single dimension judgment is avoided, significantly improving the accuracy of fall risk level classification. A secondary verification process is set up for moderate / severe risks, with visual contour verification in public areas and millimeter-wave radar depth scanning in private areas, effectively eliminating false positives and reducing unnecessary warning interference. A pruned and quantized CNN+LSTM model is used to adapt to low-computing-power edge devices at home, with a single-frame processing latency of ≤100ms, ensuring the timeliness of risk assessment and reserving sufficient time for emergency response.

[0012] Preferably, the specific implementation of the result output in step 4 includes: Privacy protection processing: During the data collection phase, the visual module is automatically turned off in private areas, and only millimeter-wave radar monitoring is retained; during the transmission phase, all gait and posture data are processed using the AES encryption algorithm, and visual data retains only limb contour features, removing facial and private area information; Tiered and coordinated early warning system: Mild fall risk is only pushed to the child's mobile app, with text reminders and a brief explanation of gait parameter deviations; moderate fall risk is simultaneously pushed to the child's mobile app and the community grid worker's terminal, with the child receiving text and voice reminders and the grid worker receiving information on the risk area and duration and initiating a follow-up phone call; severe fall risk is pushed in conjunction with the child's mobile app, the community grid worker's terminal, and the emergency platform of the nearest community hospital, with the push information including the risk level, time of occurrence, and precise location, triggering the community grid worker to conduct a home visit and the community hospital to be on standby for emergency response; Reverse optimization of the dynamic personalized gait baseline: Receive feedback data (real risk / false alarm) from children and community grid workers, and transmit this feedback data back to the data processing stage for use in the dynamic personalized gait baseline B. t Further optimization reduces the false alarm rate. A privacy protection system is built from three aspects: collection (visual access disabled in private areas), transmission (AES encryption), and storage (short-term retention + encrypted storage), which fully meets the privacy needs of home scenarios. A gradient response strategy of "mild reminder - moderate follow-up - severe emergency" is designed based on risk level to avoid resource waste or emergency delay caused by "one-size-fits-all" warnings. By receiving feedback data from children and grid workers, the gait baseline is optimized in reverse to achieve a closed-loop iteration of "warning-feedback-optimization" and continuously reduce the false alarm rate of the system in long-term operation.

[0013] A fall warning system for adaptive gait monitoring in the elderly includes a multi-source data acquisition module, an environmental adaptive processing module, a personalized risk assessment module, and a graded warning output module. The modules work together through an encrypted data transmission link. The multi-source data acquisition module is used to simultaneously collect core gait data, interference data, and scene status data of elderly people living at home, and obtain a denoised gait data sequence after preprocessing. The environmental adaptive processing module is used to calibrate the gait data sequence for environmental interference, and to construct and dynamically update a personalized gait baseline that adapts to the individual characteristics of the elderly. The personalized risk assessment module is used to calculate the deviation of gait parameters based on the dynamic personalized gait baseline, and combine the action type recognition and secondary confirmation mechanism to classify and determine the fall risk level. The tiered early warning output module is used to perform privacy protection processing, implement tiered linkage early warning according to the fall risk level, and receive feedback data to optimize the dynamic personalized gait baseline. Each module works together through an encrypted link to achieve end-to-end implementation of "collection-processing-judgment-early warning". The modular architecture design allows the software and hardware to be upgraded independently. Subsequent addition of new sensor types and optimization of algorithm models do not require reconstruction of the entire system, reducing upgrade costs.

[0014] Preferably, the multi-source data acquisition module is deployed in high-fall-prone areas of the home (bathroom entrance, bedroom bedside, living room corridor), employing a combined acquisition scheme of "millimeter-wave radar + environmental sensor + privacy-protecting visual module". It supports drill-free wall-mounted installation without damaging home decoration; it is precisely deployed in high-fall-prone areas of the home and supports drill-free wall-mounted installation, adapting to different decoration styles and requiring no professional construction, thus improving user installation convenience. The combination of "millimeter-wave radar + environmental sensor + privacy-protecting visual module" not only ensures the accuracy of gait data acquisition but also avoids privacy leakage through visual desensitization design. All sensors trigger acquisition synchronously, and the data is pre-processed and then aggregated for transmission, avoiding redundant data occupying bandwidth and improving module operating efficiency.

[0015] Preferably, the environment adaptive processing module incorporates a clustering algorithm unit and a personalized gait baseline update unit. The clustering algorithm unit supports the classification of normal gait, panicked gait of Parkinson's patients, and limping gait of arthritis patients, and can add custom gait types according to changes in the elderly's physical condition. The clustering algorithm unit can accurately classify typical types such as normal gait, panicked gait of Parkinson's patients, and limping gait of arthritis patients, expanding the scope of the system's applicable population. It supports manual / automatic addition of custom gait types, adapting to personalized changes in the elderly's physical condition and improving the module's long-term adaptability. The clustering algorithm unit and the baseline update unit work together to realize an automated process of "gait type classification - baseline construction - dynamic update" without manual intervention.

[0016] Preferably, the personalized risk assessment module integrates a lightweight CNN+LSTM model. This model undergoes pruning and quantization to adapt to low-computing-power edge devices at home, with a single-frame processing latency of ≤100ms. The lightweight CNN+LSTM model can be directly deployed on home edge devices such as Raspberry Pi, without relying on cloud computing power, reducing network dependence and usage costs. While ensuring a single-frame processing latency of ≤100ms, the model has high accuracy in action type recognition, balancing the real-time performance and accuracy of risk assessment. The model undergoes standardized pruning and quantization processing, allowing for rapid portability to edge computing terminals of different brands, improving the module's hardware adaptability.

[0017] Preferably, the hierarchical early warning output module supports custom configuration of early warning recipients, allowing the addition of receiving terminals such as community elderly care service centers and relatives, and setting the early warning priority for each receiving terminal; it also supports setting the priority of early warning messages to ensure that emergency terminals such as community hospitals receive high-risk early warnings first, avoiding delays in critical information, and enabling interface docking with community elderly care systems and hospital emergency platforms, breaking through the limitations of single terminals and achieving cross-platform collaborative early warning.

[0018] Compared with existing technologies, the beneficial effects of this invention are as follows: it clarifies the task boundaries of each link, making the method highly engineering-feasible and easy to integrate and implement into the system. It incorporates the construction of dynamic and personalized gait baselines and feedback optimization into the core process, breaking through the limitations of traditional "fixed threshold" monitoring. It can be continuously adjusted according to changes in the elderly's physical condition, improving the reliability of long-term monitoring. It covers the entire link from data collection to model optimization, ensuring the integrity of the monitoring process while focusing on the core objective of "fall warning for elderly people living at home" and avoiding functional redundancy. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0020] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. The present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0021] The terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. For example, the first feature difference value and the second feature difference value mentioned below are different numerical values. It should be understood that such numerical values ​​can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those steps or modules explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. The division of modules in the embodiments of this application is merely a logical division. In actual applications, there may be other division methods. For example, multiple modules may be combined into or integrated into another system, or some features may be ignored or not performed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interface, and the indirect coupling or communication connection between modules may be electrical or other similar forms. None of these are limited in the embodiments of this application. Furthermore, the modules or sub-modules described as separate components may or may not be physically separate, may or may not be physical modules, or may be distributed among multiple circuit modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of this application.

[0022] To address the false alarm problem and low system acceptance issues mentioned in the background section, like Figure 1 As shown, this application proposes a fall early warning method based on adaptive gait monitoring in the elderly, comprising the following steps: S1. Data Acquisition: Simultaneously collect core gait data, environmental interference data, and scene status data of elderly people living at home, and obtain a denoised gait data sequence after preprocessing; S2. Data processing: Perform environmental interference calibration on the gait data sequence, and construct and dynamically update a personalized gait baseline that adapts to the individual characteristics of the elderly. S3. Data Judgment: Based on the dynamic baseline, the deviation of gait parameters is calculated, and combined with the action type recognition and secondary confirmation mechanism, the fall risk level is determined in a graded manner. S4. Output Results: Implement privacy-protected tiered alerts based on risk levels, and receive feedback data to optimize the system model in reverse. The specific implementation of the data processing in step 2 includes: Environmental interference calibration: Based on the interference data sequence E, an environmental impact factor matrix M (M1+M2+M3=1, where M1 is the light impact factor, M2 is the humidity impact factor, and M3 is the shading impact factor) is constructed. The gait data sequence D' is then calibrated using the formula D''=D'×(1-ω×M) to obtain the environmentally adapted gait data D''. Here, ω is the environmental impact weight (0.1-0.3), which is dynamically adjusted according to the light intensity L and the ground humidity H. Personalized gait baseline construction: Using a 10-day initial calibration period, gait data D'' after environmental adaptation was collected. Combined with individual elderly patient information (age, height, disease type, gait characteristics), a clustering algorithm was used to classify gait types. The mean and standard deviation of each parameter were calculated to construct the initial personalized gait baseline B0=[v0,σᵥ;f0,σ_f;s0,σ s [;θ0,σθ;α0,σα];where v0 is the calibrated mean gait speed, f0 is the calibrated mean cadence, s0 is the calibrated mean stride length, θ0 is the calibrated mean body center of gravity offset angle, α0 is the calibrated mean joint range of motion, σᵥ is the standard deviation of gait speed, σ_f is the standard deviation of cadence, σ s σ_θ represents the standard deviation of stride length, σ_θ represents the standard deviation of the body center of gravity offset angle, and σ_α represents the standard deviation of joint range of motion. Personalized gait baseline dynamic update: During the routine monitoring phase, the initial baseline B0 is iteratively updated using a 7-day sliding window algorithm to obtain the dynamic baseline B. t (t represents the number of monitoring days); when a change in the elderly person's physical condition is detected, the window size is shortened to 3 days; The specific implementation of the data judgment in step 3 includes: Gait parameter deviation calculation: Define deviation δ x =|x_real-x t | / σ t (x_real is the real-time gait parameter value, x) t The mean and σ of the corresponding parameters in the dynamic personalized gait baseline t (where the standard deviations of the corresponding parameters in the dynamic personalized gait baseline are used to obtain the deviation set δ=[δᵥ,δ_f,δ s ,δθ,δα]; Action type recognition: Using a lightweight CNN+LSTM model, based on scene state data S and environment-adapted gait data D''_real, we can distinguish between "normal walking", "standing still", "bending over to pick up an object", "squatting and resting", and "falling due to imbalance". Fall risk level assessment: Mild risk is determined by the deviation δ of any one parameter. x ∈[1,1.5) and the action type is "normal walking" with a duration of ≥8 minutes; moderate risk is any one parameter deviation δ x ∈[1.5,2) or 2 or more parameters deviation δ x ∈[1,1.5) and the action type is "walking" or "bending over", with a duration of ≥5 minutes; severe risk is any one parameter deviation δ x ≥2 or the action type is identified as "unbalanced fall"; Secondary confirmation mechanism: When the risk is determined to be moderate / severe, posture data is collected in public areas through a privacy-desensitized visual module, while private areas are switched to millimeter-wave radar depth scanning. The risk authenticity is verified by combining changes in the human body aspect ratio, and false positives are eliminated. The specific implementation of the result output in step 4 includes: Privacy protection processing: During the data collection phase, the visual module is automatically turned off in private areas, and only millimeter-wave radar monitoring is retained; during the transmission phase, all gait and posture data are processed using the AES encryption algorithm, and visual data retains only limb contour features, removing facial and private area information; Tiered and coordinated early warning system: Mild fall risk is only pushed to the child's mobile app, with text reminders and a brief explanation of gait parameter deviations; moderate fall risk is simultaneously pushed to the child's mobile app and the community grid worker's terminal, with the child receiving text and voice reminders and the grid worker receiving information on the risk area and duration and initiating a follow-up phone call; severe fall risk is pushed in conjunction with the child's mobile app, the community grid worker's terminal, and the emergency platform of the nearest community hospital, with the push information including the risk level, time of occurrence, and precise location, triggering the community grid worker to conduct a home visit and the community hospital to be on standby for emergency response; Reverse optimization of the dynamic personalized gait baseline: Receive feedback data (real risk / false alarm) from children and community grid workers, and transmit this feedback data back to the data processing stage for use in the dynamic personalized gait baseline B. t Further optimization will reduce the subsequent false alarm rate.

[0023] In this embodiment, by constructing and dynamically updating a personalized gait baseline, the differences in age, height, disease type, and gait characteristics of different elderly people are adapted, thus solving the problem of low early warning accuracy of traditional general baselines. Based on multi-source environmental sensor data, an influence factor matrix is ​​constructed, and gait data is calibrated for environmental adaptation. This effectively reduces false alarms caused by environmental interference such as light, humidity, and occlusion, and improves the monitoring stability in complex home scenarios. By employing a multi-layered privacy protection mechanism that combines "disabling the visual module in private areas + encrypted data transmission + desensitization of visual data," the system balances monitoring needs with the privacy and safety of the elderly, thereby increasing user acceptance. By combining gait parameter deviation and movement type recognition with a secondary confirmation mechanism, the accuracy of fall risk assessment is greatly improved; the tiered linkage early warning system enables a gradient response of "mild reminder - moderate follow-up - severe emergency", taking into account both timeliness and reasonable resource allocation. By using feedback data to optimize personalized gait baselines, the false alarm rate is continuously reduced, improving the long-term reliability and adaptability of the system and adapting to the dynamic changes in the elderly's physical condition.

[0024] After introducing the embodiments of the fall warning method based on adaptive gait monitoring for the elderly proposed in this application, the embodiments of the fall warning system based on adaptive gait monitoring for the elderly proposed in this application are described below, such as... Figure 2 As shown, the fall warning system for adaptive gait monitoring in the elderly includes: a multi-source data acquisition module, an environmental adaptive processing module, a personalized risk assessment module, and a graded warning output module. Each module works collaboratively through an encrypted data transmission link. The multi-source data acquisition module is used to simultaneously collect core gait data, interference data, and scene status data of elderly people living at home, and obtain a denoised gait data sequence after preprocessing. The environmental adaptive processing module is used to calibrate the gait data sequence for environmental interference, and to construct and dynamically update a personalized gait baseline that adapts to the individual characteristics of the elderly. The personalized risk assessment module is used to calculate the deviation of gait parameters based on the dynamic personalized gait baseline, and combine the action type recognition and secondary confirmation mechanism to classify and determine the fall risk level. The graded early warning output module is used to perform privacy protection processing, implement graded linkage early warning according to the fall risk level, and receive feedback data to optimize the dynamic personalized gait baseline in reverse. The multi-source data acquisition module is deployed in high-risk areas of the home (bathroom entrance, bedroom bedside, living room corridor), and adopts a combination acquisition scheme of "millimeter-wave radar + environmental sensor + privacy-protecting vision module". It supports hole-free wall-mounted installation without damaging home decoration. The environment adaptive processing module has a built-in clustering algorithm unit and a personalized gait baseline update unit. The clustering algorithm unit supports the classification of normal gait, panicked gait of Parkinson's patients, and limping gait of arthritis patients, and can add custom gait types according to changes in the elderly's physical condition. The personalized risk assessment module integrates a lightweight CNN+LSTM model, which is adapted to low-computing-power edge devices at home after pruning and quantization, with a single-frame processing latency of ≤100ms. The hierarchical early warning output module supports custom configuration of early warning recipients, allowing the addition of receiving terminals such as community elderly care service centers and relatives, and setting the early warning priority for each receiving terminal.

[0025] In this embodiment, the modules work together through a standardized encrypted transmission link, which can be adapted to sensors and terminal devices of different brands. The functions can be flexibly expanded in the future, reducing the system upgrade cost. The multi-source data acquisition module adopts a drill-free installation design, which is suitable for complex home environments. The installation process is simple and does not require professional construction, thus improving the user's convenience. The environmental adaptive processing module supports the classification of multiple pathological gait types and can accurately adapt to the gait characteristics of patients with common elderly diseases such as Parkinson's disease and arthritis, thus expanding the system's applicable population. The lightweight model of the personalized risk assessment module is adapted to low-computing-power home devices, without relying on cloud computing power, reducing network dependence and usage costs, while ensuring the privacy and security of local data processing. The tiered early warning output module supports customization of the receiving objects and priorities, and can be flexibly configured according to the different elderly care needs of different families (such as elderly people living alone or elderly people in community care), thereby improving the personalization and practicality of the early warning scheme.

[0026] After introducing the embodiments of the fall warning system for adaptive gait monitoring of the elderly proposed in this application, the following describes a computer-readable storage medium storing a computer program that, when executed by a processor, implements a fall warning method for adaptive gait monitoring of the elderly as described in any of the above embodiments.

[0027] The main functions achieved by this invention are: 1. Personalized dynamic baseline mechanism: Breaking through the limitations of general baselines, it constructs a unique gait baseline by combining the individual information of the elderly, and dynamically updates it through a 7-day / 3-day dual-mode sliding window to adapt to changes in physical condition.

[0028] 2. Multi-source environmental anti-interference calibration: Construct an environmental impact factor matrix, calibrate gait data through quantitative formulas, and reduce false alarms caused by home environment interference by combining preprocessing algorithms.

[0029] 3. Privacy and Tiered Early Warning Synergy: Create a full-link privacy protection system of "regional control - encrypted transmission - data anonymization", and design multi-terminal tiered early warning systems of "light - medium - heavy" to balance privacy and timeliness.

[0030] 4. Closed-loop iterative optimization system: The dynamic baseline is optimized by reverse optimization of user feedback data, forming a closed loop of "collection-judgment-early warning-optimization" to continuously reduce the false alarm rate in long-term operation.

[0031] It is important to clarify that the computer-readable storage medium in this application includes both permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0032] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0033] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the methods, apparatuses, and devices described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0034] In the embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection between devices or modules through some interfaces, and may be electrical, mechanical, or other forms.

[0035] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0036] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0037] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0038] The computer program product includes one or more computer instructions. When the computer program is loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0039] Although embodiments of this application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles of this application.

Claims

1. A fall warning method based on adaptive gait monitoring in the elderly, characterized in that, Includes the following steps: S1. Data Acquisition: Simultaneously collect core gait data, environmental interference data, and scene status data of elderly people living at home, and obtain a denoised gait data sequence after preprocessing; S2. Data processing: Perform environmental interference calibration on the gait data sequence, and construct and dynamically update a personalized gait baseline that adapts to the individual characteristics of the elderly. S3. Data Judgment: Based on the dynamic baseline, the deviation of gait parameters is calculated, and combined with the action type recognition and secondary confirmation mechanism, the fall risk level is determined in a graded manner. S4. Output Results: Implement privacy-protected tiered alerts based on risk levels, and receive feedback data to optimize the system model.

2. The fall warning method for adaptive gait monitoring in the elderly as described in claim 1, characterized in that, The specific implementation of data acquisition in step 1 includes: Core gait data: Step speed (v), cadence (f), stride length (s), body center of gravity offset angle (θ), and joint range of motion (α) are collected at 15Hz using millimeter-wave radar, forming the original gait data sequence D=[v1,f1,s1,θ1,α1;v2,f2,s2,θ2,α2;...;v n ,f n ,s n ,θ n ,α n ], where n is the amount of data collected; Environmental interference data: Light intensity L, ground humidity H, and distance to obstructions D are collected via environmental sensors. o This forms an interference data sequence E=[L1,H1,D] o1 ;L2,H2,D o2 ;...;L m H m D om ], where m is the number of data collections; Scene state data: The location and activity status of the elderly are collected through infrared sensors, forming state data S=[Area1,State1;Area2,State2;...;Area...]. k State k ], where k is the number of state records, Area represents the type of area where the elderly person is located, specifically a public area (living room, bedroom) or a private area (toilet, bathroom), and State represents the current activity state of the elderly person, specifically still, walking, bending over, squatting; Data preprocessing: The "Kalman filter + interference threshold removal" algorithm is used to remove abnormal data points caused by sudden changes in illumination, slippery ground, and occlusion in the original gait data D based on the interference data sequence E, so as to obtain the denoised gait data sequence D'.

3. The fall warning method for adaptive gait monitoring in the elderly as described in claim 1, characterized in that, The specific implementation of the data processing described in step 2 includes: Environmental interference calibration: Based on the interference data sequence E, an environmental impact factor matrix M (M1+M2+M3=1, where M1 is the light impact factor, M2 is the humidity impact factor, and M3 is the shading impact factor) is constructed. The gait data sequence D' is then calibrated using the formula D''=D'×(1-ω×M) to obtain the environmentally adapted gait data D''. Here, ω is the environmental impact weight (0.1-0.3), which is dynamically adjusted according to the light intensity L and the ground humidity H. Personalized gait baseline construction: Using a 10-day initial calibration period, gait data D'' after environmental adaptation was collected. Combined with individual elderly patient information (age, height, disease type, gait characteristics), a clustering algorithm was used to classify gait types. The mean and standard deviation of each parameter were calculated to construct the initial personalized gait baseline B0=[v0,σᵥ;f0,σ_f;s0,σ s [;θ0,σθ;α0,σα];where v0 is the calibrated mean gait speed, f0 is the calibrated mean cadence, s0 is the calibrated mean stride length, θ0 is the calibrated mean angle of body center of gravity offset, α0 is the calibrated mean range of motion, σᵥ is the standard deviation of gait speed, σ_f is the standard deviation of cadence, σ s σ_θ represents the standard deviation of stride length, σ_θ represents the standard deviation of the body center of gravity offset angle, and σ_α represents the standard deviation of joint range of motion. Personalized gait baseline dynamic update: During the routine monitoring phase, the initial baseline B0 is iteratively updated using a 7-day sliding window algorithm to obtain the dynamic baseline B. t (t represents the number of monitoring days); when a change in the elderly person's physical condition is detected, the window size is shortened to 3 days.

4. The fall warning method for adaptive gait monitoring in the elderly as described in claim 1, characterized in that, The specific implementation of the data judgment in step 3 includes: Gait parameter deviation calculation: Define deviation δ x =|x_real-x t | / σ t (x_real is the real-time gait parameter value, x) t The mean and σ of the corresponding parameters in the dynamic personalized gait baseline t (where the standard deviations of the corresponding parameters in the dynamic personalized gait baseline are used to obtain the deviation set δ=[δᵥ,δ_f,δ s ,δθ,δα]; Action type recognition: Using a lightweight CNN+LSTM model, based on scene state data S and environment-adapted gait data D''_real, we can distinguish between "normal walking", "standing still", "bending over to pick up an object", "squatting and resting", and "falling due to imbalance". Fall risk level assessment: Mild risk is determined by the deviation δ of any one parameter. x ∈[1,1.5) and the action type is "normal walking" with a duration of ≥8 minutes; moderate risk is any one parameter deviation δ x ∈[1.5,2) or 2 or more parameters deviation δ x ∈[1,1.5) and the action type is "walking" or "bending over", with a duration of ≥5 minutes; severe risk is any one parameter deviation δ x ≥2 or the action type is identified as "unbalanced fall"; Secondary confirmation mechanism: When the risk is determined to be moderate / severe, posture data is collected in public areas through a privacy-desensitized visual module, while private areas are switched to millimeter-wave radar depth scanning. The risk authenticity is verified by combining changes in the aspect ratio of the human body, and false positives are eliminated.

5. The fall warning method for adaptive gait monitoring in the elderly as described in claim 1, characterized in that, The specific implementation of the result output in step 4 includes: Privacy protection processing: During the data collection phase, the visual module is automatically turned off in private areas, and only millimeter-wave radar monitoring is retained; during the transmission phase, all gait and posture data are processed using the AES encryption algorithm, and the visual data retains only the limb outline features, removing facial and private area information; Tiered and coordinated early warning system: Mild fall risk is only pushed to the child's mobile app, with text reminders and a brief explanation of gait parameter deviations; moderate fall risk is simultaneously pushed to the child's mobile app and the community grid worker's terminal, with the child receiving text and voice reminders and the grid worker receiving information on the risk area and duration and initiating a follow-up phone call; severe fall risk is pushed in conjunction with the child's mobile app, the community grid worker's terminal, and the emergency platform of the nearest community hospital, with the push information including the risk level, time of occurrence, and precise location, triggering the community grid worker to conduct a home visit and the community hospital to be on standby for emergency response; Reverse optimization of the dynamic personalized gait baseline: Receive feedback data (real risk / false alarm) from children and community grid workers, and transmit this feedback data back to the data processing stage for use in the dynamic personalized gait baseline B. t Further optimization will reduce the subsequent false alarm rate.

6. A fall warning system for adaptive gait monitoring in the elderly, used to implement the method of any one of claims 1-5, characterized in that, It includes a multi-source data acquisition module, an environmental adaptive processing module, a personalized risk assessment module, and a graded early warning output module. Each module works collaboratively through an encrypted data transmission link. The multi-source data acquisition module is used to simultaneously collect core gait data, interference data, and scene status data of elderly people living at home, and obtain a denoised gait data sequence after preprocessing. The environmental adaptive processing module is used to calibrate the gait data sequence for environmental interference, and to construct and dynamically update a personalized gait baseline that adapts to the individual characteristics of the elderly. The personalized risk assessment module is used to calculate the deviation of gait parameters based on the dynamic personalized gait baseline, and combine the action type recognition and secondary confirmation mechanism to classify and determine the fall risk level. The graded early warning output module is used to perform privacy protection processing, implement graded linkage early warning according to the fall risk level, and receive feedback data to optimize the dynamic personalized gait baseline.

7. The fall warning system for adaptive gait monitoring in the elderly as described in claim 6, characterized in that, The multi-source data acquisition module is deployed in high-risk areas of the home (bathroom entrance, bedroom bedside, living room corridor), and adopts a combination acquisition scheme of "millimeter-wave radar + environmental sensor + privacy-protecting vision module". It supports hole-free wall-mounted installation without damaging home decoration.

8. The fall warning system for adaptive gait monitoring in the elderly as described in claim 6, characterized in that, The environment adaptive processing module has a built-in clustering algorithm unit and a personalized gait baseline update unit. The clustering algorithm unit supports the classification of normal gait, panicked gait of Parkinson's patients, and limping gait of arthritis patients, and can add custom gait types according to changes in the elderly's physical condition.

9. The fall warning system for adaptive gait monitoring in the elderly as described in claim 6, characterized in that, The personalized risk assessment module integrates a lightweight CNN+LSTM model, which is adapted to low-computing-power edge devices at home after pruning and quantization, with a single-frame processing latency of ≤100ms.

10. The fall warning system for adaptive gait monitoring in the elderly as described in claim 6, characterized in that, The hierarchical early warning output module supports custom configuration of early warning recipients, allowing the addition of receiving terminals such as community elderly care service centers and relatives, and setting the early warning priority for each receiving terminal.

Images