A system and method for safe monitoring of a solitary elderly person

Abstract

This invention relates to the field of elderly care technology, specifically to a non-intrusive safety monitoring system and method for elderly people living alone. The technical solution includes: a data acquisition module for non-intrusive collection of multi-dimensional physiological data and home environment data of elderly people living alone; a data transmission module for real-time data transmission; a data preprocessing module for preprocessing the data to obtain standardized multi-dimensional data; a feature extraction module for extracting key feature data; a data fusion module for assigning trust levels to the key feature data and calculating fused feature data through a multi-source data fusion algorithm; and a status recognition module for identifying the elderly person's current physical state and risk level, automatically triggering a tiered early warning mechanism when a dangerous situation is detected. This invention combines the elderly person's historical health data and daily activity habits, effectively solving the technical problems of misjudgment and missed judgments caused by single-data monitoring, and can accurately identify the elderly person's real-time physical state and risk level.

Description

Technical Field

[0001] This invention relates to the field of elderly monitoring technology, specifically to a non-intrusive safety monitoring system and method for elderly people living alone. Background Technology

[0002] With the continuous advancement of urbanization, the number of elderly people living alone is expanding year by year, and their home safety and emotional needs have become urgent social issues that need to be addressed. Currently, elderly people living alone mainly face two core dilemmas: First, insufficient safety risk prevention capabilities. When elderly people experience dangerous situations at home, such as falls or sudden cardiovascular and cerebrovascular diseases (such as myocardial infarction or stroke), they often miss the best treatment opportunity due to lack of discovery and assistance. Existing monitoring methods mostly rely on the elderly's active operation (such as emergency call devices and manual alarm devices), which cannot effectively trigger alarms when the elderly lose mobility or consciousness after a fall, resulting in significant limitations in application. Second, loneliness and social isolation are prominent issues. Elderly people living alone lack daily companionship and emotional exchange, and prolonged solitude can easily lead to psychological problems such as depression and anxiety. Existing monitoring systems mostly focus on the single function of safety monitoring and do not take into account the need for emotional companionship services, failing to meet the diverse needs of elderly people living alone.

[0003] Existing monitoring technologies for elderly people living alone generally use video monitoring. Although this can achieve visual monitoring, it poses a serious risk of privacy leakage and cannot achieve real-time collection and accurate analysis of physiological data. It is difficult to identify abnormal physical conditions of the elderly in advance and cannot provide early warning of dangerous situations. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a non-intrusive safety monitoring system for elderly people living alone. This system acquires multi-dimensional physiological data and home environment data of the elderly through a non-intrusive data collection method. Combined with a data fusion algorithm, it accurately identifies the elderly person's real-time physical condition and risk level, enabling real-time early warning and rapid response to dangerous situations such as falls and sudden illnesses. The non-intrusive safety monitoring system for elderly people living alone includes: The data acquisition module is used to collect multi-dimensional physiological data and home environment data of elderly people living alone without them noticing.

[0005] The data transmission module is used to transmit multi-dimensional physiological data and home environment data to the data fusion processing module in real time. At the same time, the data fusion processing results, status recognition results, and danger warning information are transmitted to the cloud server and the terminal interaction module.

[0006] The data preprocessing module is used to preprocess the collected multidimensional physiological data and home environment data to obtain standardized multidimensional data.

[0007] The feature extraction module is used to extract key feature data from standardized multi-dimensional data.

[0008] The data fusion module is used to assign trust levels to key feature data and calculate fused feature data F through a multi-source data fusion algorithm.

[0009] The status recognition module is used to identify the elderly person's current physical condition and risk level based on fused feature data. When a dangerous situation is identified, a graded early warning mechanism is automatically triggered.

[0010] The terminal interaction module is used to enable efficient interaction among relevant personnel.

[0011] The cloud server is used to store the elderly’s multi-dimensional physiological data, home environment data, historical health data, physical condition and early warning records.

[0012] Preferably, the elderly living alone safety monitoring system also includes an emotional companionship module, which is used to provide personalized emotional companionship services to the elderly by combining the elderly's real-time physical condition and daily activity habits.

[0013] The preferred emotional companionship module includes: a voice companionship unit, a family interaction unit, and an interest-based push unit; the voice companionship unit supports elderly people's voice communication and can proactively remind them to pay attention to their health based on their physical condition; the family interaction unit automatically sends reminder signals to family members and caregivers when it detects that an elderly person has been alone for a long time; it can also regularly push information about the elderly person according to the needs of family members; the interest-based push unit regularly pushes content related to the elderly person's interests and hobbies.

[0014] Preferably, the data acquisition module includes: a physiological data acquisition unit, an environmental data acquisition unit, and a supplementary data acquisition unit. The physiological data acquisition unit is used for non-intrusive collection of multi-dimensional physiological data from elderly people living alone; the environmental data acquisition unit is used to collect home environment data for elderly people living alone; and the supplementary data acquisition unit is used to collect home gas concentration and smoke concentration.

[0015] Preferred methods include outlier removal, noise filtering, and data completion.

[0016] Preferred features include: Outlier removal: Precisely removes invalid and abnormal data caused by sensor malfunctions and environmental interference. Noise filtering: Performs noise filtering on the collected raw data to remove invalid noise caused by environmental interference and retain valid data. Data completion: For missing data during sensor acquisition, interpolation is used to complete the data and obtain standardized multi-dimensional data.

[0017] Preferred key feature data includes: physiological feature data and environmental feature data.

[0018] Preferably, the method for obtaining key feature data includes: receiving preprocessed standardized multi-dimensional data; unifying the data format into a standardized matrix, performing analysis and calculation on the standardized multi-dimensional data, and then combining with the core requirements of monitoring the elderly living alone, screening out key feature data strongly related to the dangerous state, and then statistically aggregating them to obtain the final key feature set. The standardized multi-dimensional data includes physiological feature data and environmental feature data. Physiological threshold features (such as blood oxygen threshold, body posture angle, etc.) can be analyzed and calculated by different methods.

[0019] Standardization calculation method: , where x’ is the standardized key feature data, x is the preprocessed standardized multi-dimensional data, μ is the mean of the standardized multi-dimensional data, and σ is the standard deviation of the standardized multi-dimensional data.

[0020] Preferably, the method for processing physiological feature data includes: 1. Selecting the db4 wavelet basis in combination with the fluctuation characteristics of physiological data; 2. Performing 3-layer wavelet decomposition on the standardized continuous physiological data ft, and the decomposition formula: ft = A3 + D3 + D2 + D1, where A3 is the 3-layer low-frequency component, and D3, D2, D1 are the high-frequency components of each layer. 3. Calculating the energy value E of the high-frequency component D3, , where i is the data point number of the 3-layer high-frequency component data, D 3（i） is the i-th data point of the 3-layer high-frequency component, and I is the data length, that is, the total number of data points of the 3-layer high-frequency component; setting an energy threshold E0, when E > E0, it is determined as key feature data.

[0021] Preferably, the method for processing environmental feature data and physiological threshold features includes: 1. Constructing a feature matrix Xm×n from the preprocessed standardized home environment data and physiological threshold data, where m is the number of data collection samples and n is the feature dimension; 2. Calculating the covariance matrix C of the feature matrix X. 3. Solving the eigenvalues λ of the covariance matrix C. 4. Setting a cumulative contribution rate threshold, and selecting the first k principal components with a cumulative contribution rate of the eigenvalues ≥ 85% as the core feature vectors. 5. Mapping the original feature matrix X onto the core feature vectors to obtain the principal component feature matrix Ym×k, and screening out the features strongly related to falls and physiological abnormalities as key feature data.

[0022] Preferably, the calculation method of the covariance matrix C: , where is the mean vector of the feature matrix, and X T is the transposed matrix of the feature matrix X.

[0023] Preferably, the calculation method of the fused feature data F includes: 1. According to the defined set of fusion recognition targets 1. H1, H2, and H3 are fused to identify the target. 2. Based on weight allocation logic, a corresponding trust level of H1, H2, and H3 is assigned to each key feature data. 3. The trust levels of different targets are calculated pairwise, and all products are summed to obtain K. 4. The comprehensive trust value M(H) of all key feature data is calculated. j The H with the highest overall trust value is the fused feature data.

[0024] Preferred: Overall Trust Value , where j is the target number of the fusion identification, k is the total number of feature dimensions in the principal component feature matrix, k' is the feature dimension number in the principal component feature matrix, k'=1,2,…k; l is the fusion depth order of the feature dimension conflict coefficient K in the principal component feature matrix, and L is the total number of fusion depth orders of the conflict coefficient K, l=1,2,…,L.

[0025] Preferred methods for calculating fused feature data include: , where N is the total number of dimensions in the final key feature set; N' is the dimension number, N'=1, 2, ...N; l is the fusion depth order of the feature dimension conflict coefficient K, and L is the total number of fusion depth orders of the conflict coefficient K, l=1, 2, ..., L; X is the standard value of the final key feature. N’ It is the feature dimension data value numbered N', and X0 is the step coefficient.

[0026] Preferred configuration: The status recognition module includes: Body status recognition: Based on fused feature data and combined with a preset body status recognition model, it accurately identifies the elderly person's current body status. Risk factor assessment: Used to classify and assess the elderly person's current risk factor by combining the elderly person's real-time body status, home environment, and historical health data.

[0027] The preferred method for identifying body states includes: Step 1: Matching the fused feature data with the preset threshold database of various body state features in the model; Step 2: The random forest algorithm is responsible for preliminary classification, calculating the matching degree between the current data and the four states based on the physiological and environmental features in the fused feature data; Step 3: The neural network algorithm is responsible for correction and optimization, correcting the preliminary identification results of the random forest by combining the elderly's historical health data; Step 4: Outputting the preliminary identification results.

[0028] Preferred: Risk Factor Where F represents the fused feature data, F1 represents the state correction coefficient, F2 represents the environmental risk coefficient, and F3 represents the historical correction coefficient.

[0029] The tiered early warning system includes local early warning, terminal early warning, and third-party early warning. The specific implementation methods are as follows: Low risk coefficient (0-3 points): No early warning signal is triggered. The elderly person's real-time physical status data is simply uploaded to the cloud server and the terminal interaction module for family members and guardians to view at any time.

[0030] Medium risk level (4-6 points): Triggers a local mild warning, which is triggered by a gentle prompt tone through a home terminal (such as a smart speaker) to remind the elderly to rest, drink water, or seek medical attention. At the same time, the warning information and the elderly's real-time physical status data are sent to the terminal devices of family members and guardians to remind them to pay close attention to the elderly's condition.

[0031] High risk level (7-10 points): Triggers a local severe alert, issuing an urgent alarm sound and flashing lights through the home terminal; at the same time, the alert information, the elderly's real-time physical status data, and on-site environmental data are quickly sent to the terminal devices of family members and guardians, and simultaneously pushed to the community health service center and emergency rescue agencies to ensure that rescuers can respond quickly and provide on-site rescue, while retaining on-site data to provide accurate reference for rescue work.

[0032] This invention also provides a method for non-intrusive safety monitoring of elderly people living alone. Based on the aforementioned non-intrusive safety monitoring system for elderly people living alone, this method includes the following steps, which can realize the integrated operation of non-intrusive monitoring of the physical status of elderly people living alone, data fusion analysis, status identification, danger warning and emotional companionship. The specific steps are as follows: S1. Simultaneously collect multi-dimensional physiological data and home environment data of elderly people living alone; S2. Real-time transmission of the collected multi-dimensional physiological data and home environment data; S3. Preprocess the multidimensional physiological data and home environment data to obtain standardized multidimensional data; S4. Extract key feature data from the preprocessed standardized multi-dimensional data; S5. Merge multi-dimensional data or key feature data to obtain fused feature data F; S6. By integrating feature data with a pre-set physical condition recognition model, the current physical condition of the elderly is identified. At the same time, a risk factor assessment algorithm is used to classify and assess the current risk factor of the elderly based on their real-time physical condition, environmental conditions, and historical health data. S7. Automatically trigger the corresponding graded early warning mechanism based on the risk factor classification results.

[0033] The technical effects and advantages of this invention are as follows: 1. Achieve seamless monitoring, enhance user experience and strengthen privacy protection: Adopting a non-contact, distributed seamless data collection module, the data collection process does not require the elderly to operate actively or wear any devices, which can effectively avoid interfering with the elderly's daily life; at the same time, it avoids the discomfort of wearing wearable devices and the privacy leakage issues of video monitoring, fully adapts to the living habits of elderly people living alone, and significantly improves the elderly's acceptance of the system.

[0034] 2. Multi-source data fusion significantly improves the accuracy of status recognition and early warning: By integrating the physiological data of the elderly and home environment data, and using a professional multi-source data fusion algorithm, combined with the elderly's historical health data and daily activity habits, it effectively solves the technical problems of misjudgment and omission in single data monitoring. It can accurately identify the elderly's real-time physical status and risk factor, and realize early warning and graded warning of dangerous situations such as falls and sudden illnesses, effectively reducing the incidence and mortality of dangerous accidents.

[0035] 3. Balancing safety monitoring and emotional support to completely solve core pain points: This system not only realizes safety monitoring and danger warning for elderly people living alone, but also has a dedicated emotional support module. Through functions such as voice companionship, family interaction, and interest-based push notifications, it effectively alleviates the loneliness and social isolation of the elderly, fully meets the dual needs of elderly people living alone for seamless safety monitoring and emotional support, and significantly improves their quality of life and happiness.

[0036] 4. Multi-terminal linkage to build a comprehensive protection system: Construct a multi-terminal interactive system involving family members, communities, emergency agencies, and home residents. In the event of a dangerous situation, it can achieve multi-faceted coordinated response to ensure that rescue work can be carried out quickly. At the same time, it allows family members to remotely monitor the elderly’s condition and communities to carry out precise assistance work, forming a comprehensive and multi-level protection network for elderly people living alone. Attached Figure Description

[0037] Figure 1 This is a structural block diagram of a non-intrusive safety monitoring system for elderly people living alone, as proposed in this invention.

[0038] Figure 2 This is a flowchart of a physiological characteristic data processing method in a non-intrusive safety monitoring system for elderly people living alone, as proposed in this invention.

[0039] Figure 3 This is a flowchart of a method for processing environmental feature data and physiological threshold features in a non-intrusive safety monitoring system for elderly people living alone, as proposed in this invention.

[0040] Figure 4 This is a flowchart of a method for calculating integrated feature data in a non-intrusive safety monitoring system for elderly people living alone, as proposed in this invention.

[0041] Figure 5This is a flowchart of a method for recognizing body status in a non-intrusive safety monitoring system for elderly people living alone, as proposed in this invention.

[0042] Figure 6 This is a flowchart illustrating a method for non-intrusive safety monitoring of elderly people living alone, as proposed in this invention. Detailed Implementation

[0043] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the invention, and should not be construed as limiting the invention. Rather, embodiments of the invention include all variations, modifications, and equivalents falling within the spirit and scope of the appended claims.

[0044] Example 1 refer to Figure 1 This embodiment proposes a non-intrusive safety monitoring system for elderly people living alone. It acquires multi-dimensional physiological and home environment data of the elderly through non-intrusive data collection, and combines this with a data fusion algorithm to accurately identify the elderly person's real-time physical condition and risk level. This enables real-time early warning and rapid response to dangerous situations such as falls and sudden illnesses. Simultaneously, it integrates with an emotional support module to effectively alleviate loneliness and social isolation among the elderly, comprehensively meeting the dual core needs of elderly people living alone for non-intrusive safety monitoring and emotional support, thereby improving their home safety and quality of life. The non-intrusive safety monitoring system for elderly people living alone includes: The data acquisition module is used to collect multi-dimensional physiological and home environment data of elderly people living alone without any contact. The collection process requires no active operation from the elderly, effectively avoiding interference with their daily lives and maximizing their privacy. This module consists of multiple contactless acquisition units, deployed in a distributed manner and installed in key areas of the elderly person's home (such as bedrooms, living rooms, bathrooms, kitchens, etc.). It can also be an integrated detection device (tracking data on the elderly person's movements), such as a wristband. The data acquisition module can include: a physiological data acquisition unit, an environmental data acquisition unit, and a supplementary data acquisition unit. The physiological data acquisition unit is used to collect multi-dimensional physiological data of elderly people living alone without any contact. Specifically, it can use millimeter-wave radar sensors or non-contact heart rate and blood oxygen sensors to collect core physiological data such as heart rate, blood oxygen saturation, respiratory rate, body temperature, range of physical activity, and changes in body posture. The collection frequency can be flexibly set according to actual needs (default is 1 time / minute, automatically increasing to 1 time / 10 seconds after a danger warning is triggered), ensuring real-time data collection in dangerous situations. Environmental Data Acquisition Unit: This unit collects home environment data for elderly people living alone. It can utilize temperature and humidity sensors, human presence sensors, and sound sensors to accurately collect environmental data such as temperature and humidity parameters, the elderly person's activity trajectory, ambient noise levels, and abnormal sounds (e.g., falls, groans of pain). This data provides auxiliary support for assessing the elderly person's physical condition and identifying potential dangers. Supplementary Data Acquisition Unit: Optional gas and smoke sensors can be added to collect home gas and smoke concentrations. This effectively mitigates secondary safety hazards caused by the elderly forgetting to turn off the gas or accidental fires, further enhancing the system's safety protection capabilities.

[0045] The data acquisition module features a low-power design, operating via either battery power or household circuit power, offering flexible and convenient deployment. All sensors employ a non-contact acquisition design, preserving the privacy of the elderly and perfectly adapting to the daily habits of seniors living alone, thus increasing their acceptance of the system. For example, for seniors living alone in a two-bedroom apartment in the city, the non-contact data acquisition module's acquisition units can be distributed across key areas of the home to ensure comprehensive and thorough data collection. Millimeter-wave radar sensors, non-contact heart rate and blood oxygen sensors, and human presence sensors are deployed in the bedroom to monitor the senior's physiological state and activity during sleep. Millimeter-wave radar sensors, sound sensors, and temperature and humidity sensors are deployed in the living room to monitor the senior's daily activities and environmental parameters. Millimeter-wave radar sensors and sound sensors (primarily monitoring for fall impacts) are deployed in the bathroom to prevent falls. Gas sensors and temperature and humidity sensors are deployed in the kitchen to monitor for gas leaks and ensure kitchen safety.

[0046] The data transmission module transmits multi-dimensional physiological data and home environment data in real time to the data fusion processing module. Simultaneously, it transmits the data fusion processing results, status identification results, and danger warning information to the cloud server and terminal interaction module, achieving efficient two-way data transmission. The data transmission module can employ a dual transmission mode of "wireless transmission + wired backup," effectively ensuring the stability, real-time performance, and integrity of data transmission: Wireless transmission: Utilizing low-power wireless transmission protocols such as WiFi, Bluetooth BLE, and LoRa, it adapts to complex home environments, providing comprehensive coverage of the elderly person's entire living space with low power consumption, thus not increasing system operating costs. Wired backup: Employing Ethernet transmission, the system automatically switches to wired transmission mode when wireless transmission experiences signal interruptions or interference, effectively preventing data loss and ensuring uninterrupted monitoring. Furthermore, the data transmission module incorporates a dedicated data encryption unit, employing AES encryption algorithm to encrypt all transmitted data throughout the entire process, effectively preventing data leakage and tampering, comprehensively protecting the elderly person's privacy and health information security, and complying with relevant data security regulations.

[0047] The data preprocessing module is used to preprocess the collected multi-dimensional physiological data and home environment data to obtain standardized multi-dimensional data. Preprocessing can include outlier removal, noise filtering, and data completion. It accurately removes invalid and abnormal data caused by sensor malfunctions or environmental interference, and uses interpolation to complete missing data, ensuring the accuracy, completeness, and standardization of the data. Outlier removal: Accurately removes invalid and abnormal data caused by sensor malfunctions or environmental interference. For example, a heart rate exceeding 180 beats / minute without corresponding postural changes is considered invalid data and removed to prevent abnormal data from affecting subsequent analysis results. Noise filtering: Performs noise filtering on the collected raw data (such as environmental sound data collected by sound sensors), removing invalid noise caused by environmental interference (such as everyday television sounds and external noise), and retaining valid data (such as sounds of falls or groans). Data completion: For missing data during sensor acquisition, interpolation is used to complete the data to obtain standardized multi-dimensional data, ensuring data completeness and avoiding bias in fusion analysis caused by missing data.

[0048] The feature extraction module is used to extract key feature data from standardized multi-dimensional data. Key feature data can specifically include: physiological feature data (such as heart rate fluctuation amplitude, blood oxygen saturation threshold, abnormal respiratory rate features, etc.) and environmental feature data (such as features of interrupted activity trajectory in the elderly, abnormal sound features, and sudden changes in body posture, etc.). The overall process can include: receiving preprocessed standardized multi-dimensional data (including physiological data: heart rate, blood oxygen saturation, respiratory rate, etc.; environmental data: activity trajectory, sound intensity, body posture angle, etc.); the data format is unified into a standardized matrix to eliminate the influence of dimensions, and the calculation method is standardized. In this dataset, x' represents standardized key feature data, x represents preprocessed standardized multidimensional data, μ represents the mean of the standardized multidimensional data, and σ represents the standard deviation of the standardized multidimensional data. This ensures that different types of data can be collaboratively extracted for features, although other standardization methods are not excluded. The standardized multidimensional data is analyzed and calculated, and then, combined with the core needs of monitoring elderly people living alone (focusing on identifying falls and sudden illnesses), key feature data strongly correlated with dangerous states are selected. These are then statistically aggregated to obtain the final key feature set. The standardized multidimensional data includes physiological and environmental feature data, and physiological threshold features (blood oxygen threshold, body posture, etc.) can be analyzed and calculated using different methods. (Reference) Figure 2 Physiological characteristic data processing methods may include: 1. Combining the fluctuation characteristics of physiological data, selecting the db4 wavelet basis (adapting to the low-frequency stability and high-frequency abrupt change characteristics of human physiological signals); 2. Performing 3-level wavelet decomposition on the standardized continuous physiological data ft, with the decomposition formula: ft = A3 + D3 + D2 + D1, where A3 represents the 3 levels of low-frequency components (corresponding to the stable part of the physiological signal, such as the normal heart rate baseline), and D3, D2, and D1 represent the high-frequency components of each level (corresponding to the fluctuating and abrupt parts of the physiological signal, such as sudden increases in heart rate and rapid breathing); 3. Calculating the energy value E (fluctuation amplitude characteristic) of the high-frequency component D3. Where i is the data point number of the 3-layer high-frequency component, and D 3（i） Let I be the i-th data point of the 3-layer high-frequency component, where I is the data length (i.e., the total number of data points in the 3-layer high-frequency component). An energy threshold E0 is set; when E > E0, it is considered key feature data. For example, after preprocessing, 10 minutes of standardized heart rate data for an elderly person (unit: beats / minute, standardized range [0, 1]) is obtained. Through 3-layer decomposition using the db4 wavelet basis, the 3-layer high-frequency component D3 is obtained, and its energy value E = 135 is calculated, which is greater than the threshold E0 = 120. Therefore, the feature "abnormal heart rate fluctuation amplitude" is extracted. Simultaneously, the respiratory rate data undergoes the same processing, resulting in a high-frequency component energy value E = 98, corresponding to abnormal fluctuations in the standardized respiratory rate data. Combined with the original respiratory rate threshold (<90 beats / minute), the feature "abnormal respiratory rate" is extracted as key feature data. (Reference) Figure 3, the environmental feature data and the physiological threshold feature processing method may include: 1. Construct a feature matrix Xm×n from the preprocessed standardized home environment data and physiological threshold data, where m is the number of data collection samples and n is the feature dimension (in this embodiment, it is 5 dimensions: blood oxygen saturation, body posture angle, sound intensity, activity trajectory duration, temperature and humidity); 2. Calculate the covariance matrix C of the feature matrix X. The calculation method is: , where is the mean vector of the feature matrix, and X T is the transpose matrix of the feature matrix X, which is used to measure the correlation of each dimension data and eliminate redundancy. 3. Solve the eigenvalues λ and the corresponding eigenvectors α of the covariance matrix C. The larger the eigenvalue, the stronger the discrimination ability of the corresponding eigenvector. 4. Set the cumulative contribution rate threshold (set to 85% in this system), and select the first k principal components with the cumulative contribution rate of eigenvalues ≥ 85% as the core eigenvectors. 5. Map the original feature matrix X onto the core eigenvectors to obtain the principal component feature matrix Ym×k, where k < n. Select the features strongly correlated with falls and physiological abnormalities as the key feature data. For example, construct a 5-dimensional feature matrix X 30×5(30 samples, 5 dimensions: blood oxygen saturation, body posture angle, sound intensity, activity trajectory duration, temperature and humidity), after calculating the covariance matrix C, 5 eigenvalues ​​were obtained: λ1=3.2 (body posture angle), λ2=1.8 (sound intensity), λ3=0.9 (blood oxygen saturation), λ4=0.3 (activity trajectory duration), λ5=0.1 (temperature and humidity); the cumulative contribution rate was calculated as: 3.2+1.8+0.9 / 3.2+1.8+0.9+0.3+0.1=88.6%≥85%, and the first 3 principal components were selected. Further screening of features strongly correlated with dangerous states: body angle (principal component 1), sound intensity (principal component 2), and blood oxygen saturation (principal component 3). Combined with preset thresholds (body angle > 45°, sound intensity ≥ 80 dB, blood oxygen saturation < 90%), “body abrupt change features” (body angle = 52° > 45°), “abnormal sound features” (sound intensity = 86 dB ≥ 80 dB), and “abnormal blood oxygen saturation threshold features” (blood oxygen = 88% < 90%) were extracted. These, along with the “abnormal heart rate fluctuation amplitude” and “abnormal respiratory rate” features extracted by wavelet transform, were combined to form the final key feature set for subsequent fusion analysis. By combining two algorithms, wavelet transform is used to accurately capture instantaneous abnormal fluctuations in physiological data (adapted to monitoring sudden illnesses), while principal component analysis is used to remove redundancy in environmental data and extract core distinguishing features (adapted to fall detection). The extracted key features are consistent with the feature parameters of data fusion and state recognition, ensuring the coherence and feasibility of the technical solution. At the same time, the algorithm parameters (such as wavelet basis db4, decomposition level 3, and cumulative contribution rate 85%) are all determined through training with a large number of elderly people living alone, and can be fine-tuned according to the health conditions of different elderly people to meet the personalized adaptation needs of the system.

[0049] The data fusion module assigns trust levels (weights) to key feature data, calculating the fused feature data F through a multi-source data fusion algorithm. This addresses the problem of misjudgment based on a single feature and aligns with the weight allocation logic (physiological features have higher weight than environmental features, and anomalies are weighted more heavily). (Reference) Figure 4 Specific calculation methods may include: 1. fusing and identifying the set of targets according to the definition. The data is divided into three categories: H1 (high risk, corresponding to a risk coefficient of 7-10 points), H2 (medium risk, corresponding to a risk coefficient of 4-6 points), and H3 (normal / low risk, corresponding to a risk coefficient of 0-3 points). 2. Based on the weight allocation logic (physiological features 60%-70% weight, environmental features 30%-40%), a trust level corresponding to H1, H2, and H3 is assigned to each key feature data (total ≤ 1). Abnormal features have a high trust level for H1 / H2, and normal features have a high trust level for H3. 3. The trust level products of different identification targets are calculated pairwise, and then all products are summed to obtain K. The specific steps are as follows: The conflict coefficient calculation of multiple key feature data can be carried out in steps, merging pairwise and gradually accumulating the conflict amount. The final result is consistent and more concise, which is in line with the actual calculation logic of engineering. For example, if the key feature data is (A, B, C, D, E), first calculate the conflict coefficient K{AB} of the first two pieces of evidence (A, B). Then, merge A and B and calculate the conflict coefficient K{(AB)C} with the third key feature data C. Add these values ​​sequentially to obtain the total conflict coefficient K of the five key feature data. If K > 0.5 (high conflict, such as "physiologically normal but environmentally abnormal"), use a conflict correction factor α (set to 0.6) to correct the confidence value and avoid fusion bias. 4. Calculate the comprehensive confidence value M(H) of all key feature data. j ),formula: Where j is the target number for fusion identification, which can be 1, 2, or 3. k is the total number of feature dimensions in the principal component feature matrix, k' is the feature dimension number in the principal component feature matrix, k' = 1, 2, ..., k; l is the fusion depth order of the conflict coefficient K of the feature dimension in the principal component feature matrix, and L is the total number of fusion depth orders of the conflict coefficient K, l = 1, 2, ..., L. For example, the core dimension is 5 and L is 4, which will not be elaborated here. H, with the highest comprehensive trust value, is the fused feature data. This calculation method solves the pain points of misjudgment and omission of single data. It can collaboratively fuse and analyze multi-dimensional monitoring data (physiological features + environmental features), avoiding the limitations of relying on only one data (such as a single heart rate or a single activity trajectory) to judge the elderly's status, greatly improving the accuracy of status recognition and danger warning, and adapting to the monitoring needs of elderly people living alone in multiple scenarios such as falls and sudden illnesses. Quantifiable evidence trust levels for personalized monitoring: Based on the elderly person's historical health data and the importance of features (e.g., physiological features have higher weight than environmental features), a corresponding trust level can be assigned to each monitoring feature. Trust level parameters can be fine-tuned according to the health status of different elderly individuals (e.g., those with hypertension or diabetes) to adapt to the system's personalized monitoring needs. It can identify contradictions between different pieces of evidence (e.g., "normal physiological data but abnormal environmental data"). When the conflict coefficient is too high, it optimizes through correction factors to avoid bias in the fusion results, ensuring the stability and reliability of the fusion results even in complex home environments (e.g., slight sensor interference, distinguishing between normal and abnormal solitude for the elderly). The fusion feature data calculation method may also include: , where N is the total number of dimensions in the final key feature set, is the dimension of all feature dimensions, N' is the dimension number, N'=1,2,…N; l is the fusion depth order of the feature dimension conflict coefficient K, and L is the total number of fusion depth orders of the conflict coefficient K, l=1,2,…,L. This is the standard value for the final key feature. It can be set manually or it can be the median of the corresponding dimension of health data. Details will not be elaborated here. N’ N' is the feature dimension data value, which can be obtained through detection. X0 is the step coefficient, which can be obtained empirically, generally between 2 and 5, and will not be elaborated here. This method can perform statistical calculations on various values, reduce interference from normal data, evaluate the deviation of the standard data range, and quickly amplify values ​​with large deviations, avoiding data distortion caused by data accumulation. It can quickly amplify data with large deviations.

[0050] The status recognition module identifies the elderly person's current physical condition and risk level based on fused feature data. When a dangerous situation is detected, a tiered early warning mechanism is automatically triggered to ensure that dangerous situations are detected and dealt with promptly, maximizing the safety of the elderly person's life. Specific functions are as follows: Body Status Recognition: Based on fused feature data and a pre-set body status recognition model, the system accurately identifies the current body status of the elderly, specifically categorized as normal, mild abnormal (e.g., fatigue, dizziness, slightly low blood oxygen), moderate abnormal (e.g., abnormal heart rate, shortness of breath), and severe abnormal (e.g., falls, sudden illness, loss of consciousness). The body status recognition model is constructed using a combination of random forest and neural network algorithms. The model has been trained and converged using 1000 sets of cloud-based sample data of elderly people living alone (including normal and various levels of abnormal scenarios) with a loss function <0.01, and can be directly adapted to multi-dimensional fused feature data. (Reference) Figure 5The specific process includes: First, matching the fused feature data with a pre-set "threshold library of various body state features" in the model (the threshold library is determined by sample training and can be fine-tuned according to individual differences among the elderly); Second, the random forest algorithm is responsible for initial classification, calculating the matching degree between the current data and four states: "normal, mildly abnormal, moderately abnormal, and severely abnormal," based on physiological features (heart rate fluctuations, abnormal blood oxygenation, etc.) and environmental features (abrupt changes in body posture, activity trajectory, etc.) in the fused feature data; Third, the neural network algorithm is responsible for correction and optimization, combining the elderly's historical health data to refine the initial identification results of the random forest. The process involves several steps: 1) Correcting and eliminating "misjudgments in normal scenarios" (e.g., interrupting an elderly person's activity trajectory during a nap, correcting it to a normal state based on historical activity patterns); 2) Outputting preliminary identification results (one of four states), while simultaneously comparing the fused feature data with the corresponding risk coefficient range (pre-bound: fusion result <0.3 corresponds to normal, 0.3-0.6 corresponds to mild / moderate abnormality, ≥0.6 corresponds to severe abnormality), verifying the consistency of the identification results; if inconsistent, re-identifying using historical data, and finally outputting a precise and unique physical state result, which is simultaneously transmitted to the state identification and risk warning module to provide a basis for risk coefficient assessment.

[0051] Risk assessment: This involves combining the elderly person's real-time physical condition, home environment, and historical health data to classify and assess their current risk level. In this system, F represents the fused feature data, F1 is the state correction coefficient, taken from preset values: normal = 0, mild abnormality = 0.4, moderate abnormality = 0.7, severe abnormality = 1.0; F2 is the environmental risk coefficient, with preset risk items and weights: sudden physical changes (0.4), abnormal sounds (0.3), interruption of activity trajectory (0.2), abnormal temperature / humidity / gas (0.1). Each risk item is counted as 1 if triggered, and 0 if not triggered. The weighted sum is used to obtain the environmental risk coefficient (0-1 range); F3 is the historical correction coefficient: calculated based on the cloud-based historical abnormal data of the elderly, formula: F3 = number of abnormalities in the past 3 months ÷ total number of monitoring days in the past 3 months (normalized to 0-1, H=0 if there are no abnormalities, the higher the frequency of abnormalities, the larger F3). The assessment range is 0-10 points, where 0-3 points are low risk coefficient (corresponding to normal state), 4-6 points are medium risk coefficient (corresponding to mild / moderate abnormal state), and 7-10 points are high risk coefficient (corresponding to severe abnormality / dangerous situation). Tiered early warning: Based on the risk level classification results, corresponding tiered early warnings are triggered. Early warning methods include local early warning, terminal early warning, and third-party early warning. The specific implementation methods are as follows: Low risk factor (0-3 points): No warning signals are triggered. The elderly person’s real-time physical status data is only uploaded to the cloud server and terminal interaction module for family members and guardians to view at any time. Medium risk level (4-6 points): Triggers a local mild warning, which is triggered by a gentle prompt tone through a home terminal (such as a smart speaker) to remind the elderly to rest, drink water, or seek medical attention. At the same time, the warning information and the elderly's real-time physical status data are sent to the terminal devices of family members and guardians to remind them to pay close attention to the elderly's condition. High risk level (7-10 points): Triggers a local severe alert, issuing an urgent alarm sound and flashing lights through the home terminal; at the same time, the alert information, the elderly's real-time physical status data, and on-site environmental data are quickly sent to the terminal devices of family members and guardians, and simultaneously pushed to the community health service center and emergency rescue agencies to ensure that rescuers can respond quickly and provide on-site rescue, while retaining on-site data to provide accurate reference for rescue work.

[0052] The body condition recognition model is constructed using machine learning algorithms such as random forest and neural networks. It is trained and optimized using a large amount of elderly health data and abnormal condition data to continuously improve the accuracy of body condition recognition, effectively reduce false positives and false negatives, and ensure the reliability of the recognition results. For example, when the system collects data showing an elderly person's heart rate of 110 beats / minute (heart rate fluctuations exceeding the normal range), blood oxygen saturation of 88% (below the preset threshold), and respiratory rate of 10 breaths / minute (below the preset threshold), and the elderly person's activity trajectory is interrupted for more than 30 minutes, the fused feature data shows that the elderly person is in a moderately abnormal state, with a risk factor assessment of 7 points (high risk). The system determines this as a suspected sudden hypertension complication and immediately triggers a severe warning mechanism.

[0053] The emotional companionship module, based on the elderly person's real-time physical condition and daily activity habits, provides personalized emotional companionship services to effectively alleviate loneliness and social isolation, and improve the quality of life for seniors at home. Specifically, it includes: a voice companionship unit: a built-in intelligent voice interaction system that supports voice wake-up, daily chat, and interest-based communication (such as listening to opera, telling stories, reading news, etc.). It can also proactively remind the elderly person to pay attention to their health based on their physical condition (e.g., mild abnormalities), providing emotional comfort and health tips. A family interaction unit: when the system detects that the elderly person has been alone for an extended period (e.g., no communication with the outside world for 24 consecutive hours, or a monotonous activity pattern), it automatically sends reminder signals to family members and caregivers, guiding them to engage in video or voice calls with the elderly person. It can also periodically push snippets of the elderly person's daily activities and physical condition reports according to the family's needs, achieving family interaction and allowing family members to stay informed about the elderly person's situation. An interest-based push unit: based on the elderly person's interests and hobbies, it regularly pushes content such as opera, music, health knowledge, and information on community activities for the elderly, guiding them to participate in social activities, enriching their daily lives, and alleviating loneliness.

[0054] The terminal interaction module enables efficient interaction between family members, guardians, community workers, emergency rescue personnel, and the system. It supports relevant personnel in viewing the elderly person's physical condition, receiving early warning information, and implementing remote intervention. Specifically, it includes multiple terminal devices adapted to different user needs: Family Terminal: Utilizing convenient formats such as mobile apps and WeChat mini-programs, it allows family members to view the elderly person's real-time physiological data, historical health reports, physical condition recognition results, and danger warning information. They can also remotely send voice and video call requests to the elderly person and receive emotional support reminders pushed by the system. Community Terminal: Deployed in community health service centers and community neighborhood committees, it allows community workers to view the overall monitoring status of all elderly people living alone within their jurisdiction, receive real-time high-risk warning information, and promptly arrange for staff to conduct door-to-door checks and provide assistance, improving the efficiency of supporting elderly people living alone in the community. Emergency Terminal: Deployed in emergency rescue organizations, it receives high-risk warning information and the elderly person's on-site environment and physiological data, facilitating rescue personnel to quickly grasp the on-site situation, dispatch rescue forces for on-site rescue, and shorten rescue response time. Home terminal: Adopting forms that are easy for the elderly to operate, such as smart speakers and touch screens, it is deployed in key areas of the elderly's home, supports voice interaction, allows the elderly to check their own physical status, receive local warning prompts and emotional companionship content, and is adapted to the elderly's operating habits.

[0055] The cloud server stores all system-related data for the elderly, including physiological data, environmental data, historical health data, body condition recognition results, early warning records, and emotional support records. Employing distributed cloud storage technology, it ensures data security, reliability, and scalability. It also supports data querying, statistics, and analysis, providing accurate data support for families, communities, and emergency response agencies. Furthermore, the cloud server is used for continuous training and optimization of the body condition recognition model and data fusion algorithm, constantly improving the system's monitoring accuracy and early warning precision, ensuring continuous performance enhancement.

[0056] Example 2 refer to Figure 6 The present invention also provides a method for non-intrusive safety monitoring of elderly people living alone. Based on the above-mentioned non-intrusive safety monitoring system for elderly people living alone, this method includes the following steps, which can realize the integrated operation of non-intrusive monitoring of the physical status of elderly people living alone, data fusion analysis, status identification, danger warning and emotional companionship. The specific steps are as follows: S1. Simultaneously collect multi-dimensional physiological data and home environment data of elderly people living alone; among which, physiological data includes heart rate, blood oxygen saturation, respiratory rate, body temperature, range of physical activity, and changes in body posture; environmental data includes home temperature and humidity, elderly person's activity trajectory, environmental noise, and abnormal sounds; the collection process does not require active operation by the elderly person, and the collection frequency is automatically executed according to the system's preset parameters, ensuring that the data collection is seamless and real-time.

[0057] S2. The collected multi-dimensional physiological data and home environment data are transmitted in real time. AES encryption can be performed, and a dual transmission method of "wireless transmission + wired backup" is adopted to transmit the encrypted data to the data fusion processing module in real time, while simultaneously uploading the data to the cloud server for storage. Through encryption and dual transmission modes, the security, real-time performance and integrity of data transmission are ensured, effectively preventing data leakage and loss.

[0058] S3. Preprocess the multi-dimensional physiological data and home environment data to obtain standardized multi-dimensional data. The decrypted physiological and environmental data undergo outlier removal, noise filtering, and data completion processing to accurately remove invalid data and complete missing data, resulting in standardized multi-dimensional data. This lays the foundation for subsequent feature extraction and fusion analysis.

[0059] S4. Extract key feature data from the preprocessed standardized multidimensional data. Key feature data includes key physiological features and environmental features. Physiological features include heart rate fluctuation amplitude, blood oxygen saturation threshold, and abnormal respiratory rate features. Environmental features include features of interrupted activity trajectory of the elderly, abnormal sound features, and sudden changes in body posture.

[0060] S5. Fuse multi-dimensional data or key feature data to obtain fused feature data F.

[0061] S6. By integrating feature data with a pre-set body condition recognition model, the system identifies the elderly person's current body condition (normal, mild, moderate, or severe). Simultaneously, a risk factor assessment algorithm is used, combining the elderly person's real-time body condition, environmental conditions, and historical health data, to classify and assess the elderly person's current risk factor (assessment range 0-10 points), obtaining a clear risk factor classification result, which provides a basis for subsequent classification and early warning.

[0062] S7. Automatically trigger the corresponding graded early warning mechanism based on the risk factor classification results to achieve precise response and emergency linkage in dangerous situations.

[0063] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0064] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A non-intrusive safety monitoring system for elderly people living alone, characterized in that, The unobtrusive safety monitoring system for elderly people living alone includes: The data acquisition module is used to collect multi-dimensional physiological data and home environment data of elderly people living alone without being noticed. The data transmission module is used for real-time data transmission; The data preprocessing module is used to preprocess the collected multidimensional physiological data and home environment data to obtain standardized multidimensional data; The feature extraction module is used to extract key feature data from standardized multi-dimensional data. The data fusion module is used to calculate fused feature data from key feature data using a multi-source data fusion algorithm. The status recognition module is used to identify the elderly person's current physical condition and risk level based on fused feature data. When a dangerous situation is identified, a graded early warning mechanism is automatically triggered. The terminal interaction module is used to enable efficient interaction with relevant personnel; Cloud servers are used to store data.

2. The non-intrusive safety monitoring system for elderly people living alone according to claim 1, characterized in that, It also includes an emotional companionship module, which is used to provide personalized emotional companionship services for the elderly by combining their real-time physical condition and daily activity habits.

3. The non-intrusive safety monitoring system for elderly people living alone according to claim 2, characterized in that, The emotional companionship module includes: a voice companionship unit, a family interaction unit, and an interest-based push unit. The voice companionship unit supports elderly people's voice communication and can proactively remind them to pay attention to their health based on their physical condition. The family interaction unit automatically sends reminder signals to family members and caregivers when it detects that an elderly person has been alone for a long time. It also regularly pushes information about the elderly person based on the needs of family members. The interest-based push unit pushes content based on the elderly person's interests and hobbies on a regular basis.

4. The non-intrusive safety monitoring system for elderly people living alone according to claim 1, characterized in that, The data acquisition module includes: a physiological data acquisition unit, an environmental data acquisition unit, and a supplementary data acquisition unit; the physiological data acquisition unit is used to collect multi-dimensional physiological data of elderly people living alone without being noticed; the environmental data acquisition unit is used to collect home environment data of elderly people living alone; and the supplementary data acquisition unit is used to collect home gas concentration and smoke concentration.

5. The non-intrusive safety monitoring system for elderly people living alone according to claim 1, characterized in that, Standardized multidimensional data includes physiological characteristic data, environmental characteristic data, and physiological threshold features.

6. The non-intrusive safety monitoring system for elderly people living alone according to claim 5, characterized in that, The physiological feature data processing method includes:

1. Combining the fluctuation characteristics of physiological data, the db4 wavelet basis is selected; 2. The standardized continuous physiological data ft is decomposed into 3-level wavelet decomposition, with the decomposition formula: ft=A3+D3+D2+D1, where A3 is the 3-level low-frequency component, and D3, D2, and D1 are the high-frequency components of each level; 3. The energy value E of the high-frequency component D3 is calculated; an energy threshold E0 is set, and when E>E0, it is determined to be key feature data.

7. The non-intrusive safety monitoring system for elderly people living alone according to claim 5, characterized in that, The processing method for environmental feature data and physiological threshold features includes:

1. Constructing a feature matrix X from the preprocessed standardized home environmental data and physiological threshold data, where m is the number of data collection samples and n is the feature dimension; 2. Calculating the covariance matrix C of the feature matrix X; 3. Solving the eigenvalues λ of the covariance matrix C; 4. Setting the cumulative contribution rate threshold and selecting the top k principal components with a cumulative contribution rate of eigenvalues ≥ 85% as the core feature vectors; 5. Mapping the original feature matrix X onto the core feature vectors to obtain the principal component feature matrix Ym×k, where k < n, and screening out the features strongly related to falls and physiological abnormalities as the key feature data. m×n , where m is the number of data collection samples, n is the feature dimension; 2. Calculating the covariance matrix C of the feature matrix X; 3. Solving the eigenvalues λ of the covariance matrix C; 4. Setting the cumulative contribution rate threshold and selecting the top k principal components with a cumulative contribution rate of eigenvalues ≥ 85% as the core feature vectors; 5. Mapping the original feature matrix X onto the core feature vectors to obtain the principal component feature matrix Ym×k, where k < n, and screening out the features strongly related to falls and physiological abnormalities as the key feature data.

8. The non-intrusive safety monitoring system for elderly people living alone according to claim 1, characterized in that, The methods for calculating fused feature data include:

1. fusing the set of recognition targets according to the definition.

1. H1, H2, and H3 are fused to identify the target; 2. Based on weight allocation logic, a corresponding trust level of H1, H2, and H3 is assigned to each key feature data; 3. The trust levels of different identification targets are calculated pairwise, and then all products are summed to obtain K; 4. The comprehensive trust value M (H1, H2, and H3) of all key feature data is calculated. j M (H) has the highest overall trust value. j This refers to fused feature data.

9. The non-intrusive safety monitoring system for elderly people living alone according to claim 1, characterized in that, The status recognition module includes: physical status recognition and risk factor assessment; physical status recognition is based on fused feature data and combined with a preset physical status recognition model to accurately identify the current physical status of the elderly; risk factor assessment is used to classify and assess the current risk factor of the elderly by combining the elderly's real-time physical status, home environment, and historical health data.

10. A method for non-intrusive safety monitoring of elderly people living alone, characterized in that, The method for monitoring the safety of elderly people living alone without them noticing includes: S1. Simultaneously collect multi-dimensional physiological data and home environment data of elderly people living alone; S2. Real-time transmission of the collected multi-dimensional physiological data and home environment data; S3. Preprocess the multidimensional physiological data and home environment data to obtain standardized multidimensional data; S4. Extract key feature data from the preprocessed standardized multi-dimensional data; S5. Merge multi-dimensional data or key feature data to obtain fused feature data; S6. By integrating feature data with a pre-set physical condition recognition model, the current physical condition of the elderly is identified. At the same time, a risk factor assessment algorithm is used to classify and assess the current risk factor of the elderly based on their real-time physical condition, environmental conditions, and historical health data. S7. Automatically trigger the corresponding graded early warning mechanism based on the risk factor classification results.

Images