Integrated intelligent system for health guardianship and emergency rescue of the elderly

Abstract

This invention discloses an integrated intelligent system for elderly health protection and emergency rescue, relating to the field of intelligent elderly care and health monitoring technology. It includes: a sensor module, a data processing unit, a communication module, a rescue response module, and an AI companion module. This invention integrates multiple types of sensors to comprehensively collect physiological, motion, voice, and location data. Through preprocessing, multimodal fusion, and LSTM model analysis, it accurately determines health status. Through dual-parameter judgment and data verification, it accurately identifies falls and genuine SOS calls, avoiding false triggers. In emergencies, it quickly links to location and alerts guardians and medical institutions. It features three levels of alerts to adapt to different risk scenarios. It incorporates AI emotional companionship to identify and alleviate negative emotions, achieving dual protection of physical and mental well-being. It constructs a full-chain linkage network with automatic device malfunction alerts, providing accurate, timely, and compassionate health protection and emergency rescue support for the elderly.

Description

Technical Field

[0001] This invention relates to the field of intelligent elderly care and health monitoring technology, specifically an integrated intelligent system for elderly health protection and emergency rescue. Background Technology

[0002] As my country's population continues to age rapidly, the number of elderly people who are very old, live alone, or are in empty nests is constantly expanding. Health management and safety assurance for the elderly have become core pain points in social governance and public services. Traditional elderly care models rely on manual patrols and family care, making it difficult to achieve 24-hour coverage and timely response to emergencies such as falls or cardiovascular diseases. These models suffer from prominent problems such as lagging health monitoring, delayed rescue responses, and a lack of emotional support.

[0003] Currently, while the fields of health protection and emergency rescue for the elderly are gradually moving towards intelligentization, existing technologies still have significant limitations: In fall detection, most devices rely solely on a single accelerometer to determine fall events, which is easily affected by everyday movements such as rapid turning and bending, resulting in a high false recognition rate. Furthermore, the lack of secondary verification of the static state after a fall easily leads to false alarms or missed alarms. In health monitoring, existing wearable devices mostly focus on collecting single physiological indicators, failing to achieve multimodal fusion analysis of movement posture, physiological data, and voice information. This makes it impossible to comprehensively assess the health risks of the elderly, and early warning grading and intervention strategies are relatively crude. In emergency rescue, insufficient positioning accuracy and a single communication mode make it difficult for rescuers to quickly obtain the real-time location of the elderly, and the lack of a standardized tiered response mechanism in the rescue process affects rescue efficiency. In addition, most smart terminals only focus on functional implementation, failing to integrate emotional interaction and AI companionship functions, thus failing to meet the psychological care needs of the elderly in emotional states such as loneliness and anxiety.

[0004] Meanwhile, the rapid iteration of technologies such as sensors, artificial intelligence, positioning, and the Internet of Things (IoT) has provided new paths for solving the above problems: sensor technology is developing towards miniaturization, high precision, and multi-functional integration, enabling real-time acquisition of multi-dimensional physiological and motion data; artificial intelligence technology is becoming increasingly mature in the fields of time-series data analysis, speech recognition, and emotion computing, providing personalized health management and emotional support for the elderly; breakthroughs in indoor and outdoor high-precision positioning technology and multi-mode communication technology can ensure the accuracy of location and the reliability of information transmission during emergency rescue; and IoT technology promotes data interoperability among various health devices, laying the foundation for building an integrated service system.

[0005] However, existing technologies have not yet formed a comprehensive solution that integrates accurate fall detection, multimodal health risk warning, tiered emergency rescue, and AI-powered emotional companionship. This makes it difficult to fully meet the diverse needs of the elderly in terms of health management, safety, and emotional care, thus hindering the popularization and upgrading of smart elderly care services. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides an integrated intelligent system for elderly health protection and emergency rescue, solving the problems mentioned in the background section.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the present invention provides the following technical solution: an integrated intelligent system for elderly health protection and emergency rescue, comprising:

[0010] The sensor module integrates physiological index sensors, motion and posture sensors, and voice sensors. The collected analog signals are converted into digital signals by the built-in analog-to-digital converter and then transmitted to the data processing unit.

[0011] Among them, the motion and attitude sensor integrates a three-axis accelerometer and a three-axis gyroscope to collect acceleration and angular velocity values;

[0012] The voice sensor uses a high-sensitivity omnidirectional microphone to collect the elderly’s voice signals in real time and convert the voice signals into digital signals. After the voice signals are converted into digital signals, they are transmitted in two ways: one way is transmitted to the AI ​​companion module and the other way is transmitted to the data processing unit.

[0013] The physiological indicator sensor integrates a photoelectric heart rate sensor, an oscillometric blood pressure sensor, an optical blood oxygen saturation sensor, a high-precision body temperature sensor, and a step count sensor; it is used to collect physiological indicator data corresponding to the elderly's heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature, and number of steps.

[0014] The sensor module also includes a positioning module, which is used to obtain the elderly person's real-time location information;

[0015] The data processing unit is used to calculate the elderly person's movement posture indicators based on the data transmitted by the sensor module, and to confirm whether a fall has occurred; and to identify whether there are abnormal distress calls through distress voice content recognition technology.

[0016] Simultaneously, the data on the elderly's movement posture indicators and physiological indicators are preprocessed, multimodal data are fused, deep learning model analysis is performed, health status is determined and risk warning is given, so as to output the health status characteristic value of the elderly and provide graded warnings.

[0017] The communication module is used to transmit the data collected by the sensor module and analyzed by the data processing unit to the guardian's mobile application and the cloud server of the medical institution / community rescue center;

[0018] The rescue response module is used to generate an emergency trigger signal when a fall occurs, there are abnormal distress calls, or the health condition is severely abnormal; at the same time, it is linked with the positioning module to obtain real-time location information and send it to caregivers and medical institutions / community rescue centers;

[0019] The AI ​​companion module is used to extract emotional features from voice signals and calculate an emotional comprehensive value. The emotional comprehensive value is used to determine the elderly person's emotional state, and then personalized emotional guidance is implemented based on the emotional state.

[0020] As a further aspect of the present invention, the method for confirming a fall event is as follows:

[0021] pass: , calculate the total acceleration A1;

[0022] Simultaneously through: Calculate the attitude rotation angle θ1 within the pre-specified acquisition time interval;

[0023] In the formula, A x A y A z The values ​​are acceleration values ​​collected in real time along three axes by a triaxial accelerometer, where the X-axis represents the left-right direction of the elderly person's body, the Y-axis represents the front-back direction, and the Z-axis represents the up-down direction. x ω y ω z θ represents the angular velocity values ​​acquired in real time along the three axes by the three-axis gyroscope. x θ y θ z These represent the rotation angles of the X, Y, and Z axes per unit time, respectively, with Δt being the data acquisition time interval.

[0024] Among them, the resultant acceleration and the attitude rotation angle are the motion posture indicators of the elderly.

[0025] Next, a dual-parameter determination method using resultant acceleration and attitude rotation angle is adopted, corresponding to preset resultant acceleration threshold A0 and attitude rotation angle threshold θ0;

[0026] When A1≥A0 and θ1≥θ0, and the duration is ≥0.2 seconds and ≤2 seconds, it is initially determined to be a fall event;

[0027] The resultant acceleration and rotation angle are then continuously monitored over the next 3 seconds. If the resultant acceleration A1 drops and stabilizes within the 0.9g-1.1g range over the next 3 seconds, indicating that the elderly person is at rest, a fall event is confirmed, a fall signal is generated, and the fall signal is transmitted to the caregiver.

[0028] As a further aspect of the present invention: if the data recovers to the normal motion state within the next 3 seconds, it is determined to be a false trigger and the subsequent process is not triggered.

[0029] As a further aspect of the present invention: when the data processing unit identifies distress-type voice content using SOS recognition technology:

[0030] When the recognition result of the distress voice content recognition technology is distress voice, the data collected by the current motion and posture sensors are acquired simultaneously;

[0031] If the resultant acceleration A1 is between 0.9g and 1.1g, and the change in the posture rotation angle θ1 is ≤5°, then the elderly person is in a state of stillness or unable to move independently after falling, and a real emergency rescue situation is confirmed.

[0032] As a further aspect of the present invention: if the elderly person is in a normal state of movement or in a stable posture, it is determined to be a misidentification caused by environmental interference.

[0033] As a further aspect of the present invention: heart rate, blood oxygen saturation, body temperature, and number of steps are supported for real-time continuous acquisition, while systolic blood pressure and diastolic blood pressure are acquired periodically and intermittently.

[0034] The collected data are uploaded synchronously with a unified timestamp, forming a time-series physiological data set containing heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature, and steps taken, denoted as S. i Where i = 1, 2, ..., 6, corresponding to heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature, and number of steps taken, respectively;

[0035] The values ​​of systolic and diastolic blood pressure in the physiological data set were directly taken from the measurement results collected periodically and intermittently at the corresponding timestamps;

[0036] As a further aspect of the present invention, the data preprocessing steps are as follows:

[0037] Data preprocessing involves cleaning and standardizing physiological index data and motion and posture data. The specific implementation of each step is as follows:

[0038] Step A1, Data Cleaning:

[0039] The 3σ principle is used to identify and remove outliers. The method is as follows:

[0040] when If the value is not found, it is considered an outlier and is removed. At the same time, the average of the two adjacent valid data values ​​is used to fill the gap.

[0041] Where G represents the variables of each data indicator, and it is referred to as S. i Any one of A1 and θ1;

[0042] If the number of abnormal values ​​exceeds 5 in 10 consecutive data collections, or if 5 or more abnormal values ​​occur consecutively, the sensor is preliminarily determined to be abnormal.

[0043] The data collection process is then restarted and a second verification is performed. If the second verification result still meets the above abnormal conditions, the sensor is ultimately determined to be faulty, and a device fault reminder is sent to the elderly and their caregivers.

[0044] In the comparison formula, G is a single data value collected by the sensor; GP is the average value of 10 consecutive data collections of that data indicator; σ is the standard deviation of 10 consecutive data collections of that data indicator, and i0 is the index of the corresponding data value collected 10 times consecutively for this data indicator;

[0045] Step A2, Standardization Processing:

[0046] Obtain the normal range of values ​​for each data indicator, extract the corresponding maximum and minimum values, and denot them as G. max and G min ;

[0047] Combining the results of data cleaning, and employing the min-max normalization method: This maps all data values ​​of each data indicator to the range of 0-1.

[0048] In the formula, G b These are standardized data values.

[0049] As a further aspect of the present invention, the multimodal data fusion steps are as follows:

[0050] First, the preprocessed physiological index data, motion and posture data are fused to form a unified multidimensional feature dataset, denoted as F. p ;

[0051] Among them, F p The data values ​​for each data indicator are standardized, p=1, 2, ..., 6, 7, 8, where p=1, 2, ..., 6 correspond to the six physiological indicators of the elderly: heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature, and number of steps; and p=7, 8 correspond to the two movement posture indicators of the elderly: resultant acceleration and posture rotation angle.

[0052] Extract the pre-defined weighting coefficients based on health management standards and clinical data for the elderly, and denote them as β. p Furthermore, the sum of all weight coefficients corresponding to the multidimensional feature data is 1.

[0053] Subsequently passed:

[0054] Calculate the comprehensive feature value F after fusing the multidimensional feature datasets.

[0055] As a further aspect of the present invention, the deep learning model analysis steps are as follows:

[0056] Step S1: Training dataset construction:

[0057] Long Short-Term Memory (LSTM) network is used as the deep learning model;

[0058] The training dataset is derived from ≥100,000 clinical health data and daily monitoring data of the elderly, including physiological indicator data and movement posture data of elderly people of different ages and with different underlying medical histories, and each data point is labeled with the corresponding health status label;

[0059] The health status labels are divided into four levels: normal, mildly abnormal, moderately abnormal, and severely abnormal.

[0060] Step S2, Model Input and Adaptation:

[0061] The combined feature values ​​obtained after fusing multimodal data are used to form a time series F. t The data is then input into the LSTM network model according to a pre-defined time window.

[0062] Among them, F t Let be the integrated feature value after fusion at time t, where t = 1, 2, ..., n, and n is the number of data acquisitions;

[0063] During model training, overfitting is avoided through adaptive learning rate adjustment and early stopping mechanism, ensuring the model's generalization ability in elderly health data scenarios;

[0064] Among them, adaptive learning rate adjustment: the initial learning rate is set to 0.001, and the Adam optimizer is used;

[0065] Early stopping mechanism: Training is stopped if the validation set loss does not decrease for 5 consecutive rounds;

[0066] Step S3, Model Output:

[0067] The model outputs the health status characteristic value H of the elderly.

[0068] The value of H ranges from [0,1]. The closer the value of H is to 1, the better the health condition of the elderly person; the closer the value of H is to 0, the worse the health condition.

[0069] Through actual testing, the model has achieved an accuracy rate of ≥95% in determining the health status of the test set, meeting the needs of practical applications. The test set consists of 30,000 health data points of the elderly, independent of the training set.

[0070] As a further solution of the present invention, the steps of health status determination and risk warning are as follows:

[0071] Combine the health status characteristic value H with the data value G corresponding to the standardized processing of each individual physiological index b For classification determination, the method is as follows:

[0072] First, divide the health status of the elderly into four levels: normal, slightly abnormal, moderately abnormal, and severely abnormal. At the same time, extract the corresponding preset health status thresholds H1, H2, H3, and take [0.2, 0.8] as the normal value range of the data value G after standardized processing b For the normal value range;

[0073] The determination criteria are as follows:

[0074] When H≥H3 and the standardized values G of all individual physiological indexes b Are all within the corresponding normal value range of [0.2, 0.8], the health status of the elderly is determined to be normal, that is, there is no health risk and no warning is issued;

[0075] When H2≤H<H3, or the G of a single individual physiological index b Exceeds the corresponding normal value range of [0.2, 0.8], the health status of the elderly is determined to be slightly abnormal, that is, there is a slight health risk, and at the same time, a first-level warning is issued;

[0076] Among them, the first-level warning is a vibration reminder without a buzzer reminder;

[0077] When H1≤H<H2, or the G of 2-3 individual physiological indexes b Exceeds the corresponding normal value range of [0.2, 0.8], the health status of the elderly is determined to be moderately abnormal, that is, there is a medium health risk, and at the same time, a second-level warning is issued;

[0078] Among them, the second-level warning is a vibration + low-volume buzzer reminder, and at the same time, a text message reminder is sent to the guardian;

[0079] When H<H1, or the G of ≥4 individual physiological indexes b Exceeds the [0.2, 0.8] interval, the health status of the elderly is determined to be severely abnormal, that is, there is a serious health risk, and at the same time, a third-level warning is issued

[0080] Among them, the third-level warning is a strong vibration + high-volume buzzer reminder, and at the same time, an emergency reminder is sent to the guardian and the medical institution / community rescue center.

[0081] As a further solution of the present invention, the specific implementation of the AI companion module is as follows:

[0082] Extract the tone (P), speech rate (S), and volume (L) of the elderly person's speech signal as emotional features.

[0083] Simultaneously, the extraction criteria are based on the normal range of tone, speech rate, and volume values ​​pre-recorded and calibrated from the elderly's normal speech signals, and these are sequentially denoted as [P]. min P max ]、[S min S max ]、[L min L max ];

[0084] Subsequently passed:

[0085] Calculate the comprehensive value E of emotional features in the elderly person's speech signal;

[0086] In the formula: γ P γ S γ L The preset weighting coefficients are based on tone, speaking speed, and volume.

[0087] The comprehensive value of emotional characteristics E ranges from [0,1].

[0088] Then, by combining the emotional characteristics of the elderly group with the preset emotional threshold, the emotional state in the elderly’s voice signal was determined.

[0089] When 0.8≤E≤1, it indicates that the emotional state in the elderly person's voice signal is happy;

[0090] When 0.6 ≤ E < 0.8, it indicates that the emotional state in the elderly person's speech signal is calm;

[0091] When 0.3≤E<0.6, it indicates that the emotional state in the elderly person's voice signal is depressed;

[0092] When 0.1 ≤ E < 0.3, it indicates that the emotional state in the elderly person's speech signal is anxiety;

[0093] When 0 ≤ E < 0.1, it indicates that the emotional state in the elderly person's speech signal is anger;

[0094] As a further aspect of the present invention: when the elderly person's emotional state is detected to be depressed, anxious or angry, a corresponding preset emotional guidance script is output.

[0095] At the same time, the comprehensive value of emotional characteristics E is continuously monitored. When E≥0.6, that is, the emotional state returns to calm or happy, the output of emotional guidance words is stopped.

[0096] (III) Beneficial Effects

[0097] This invention provides an integrated intelligent system for elderly health protection and emergency rescue. Compared with existing technologies, it has the following advantages:

[0098] This invention integrates multiple types of sensors to achieve comprehensive acquisition of physiological indicators, movement postures, and voice signals. It employs real-time continuous or periodic intermittent acquisition modes for different physiological indicators, combined with synchronized uploads using a unified timestamp, forming a complete data chain. The data preprocessing stage uses the 3σ principle to remove outliers and complete the data, and after standardization, maps it to a unified range, providing high-quality data support for subsequent analysis. The combination of multimodal data fusion and an LSTM deep learning model accurately determines health status, precisely identifying normal, mild, moderate, and severe abnormal states, making health monitoring more scientific and reliable.

[0099] This invention addresses the high-risk factors faced by the elderly. The system uses two parameters—combined acceleration and posture rotation angle—to determine falls, combined with subsequent 3-second data verification to effectively avoid false triggers. During distress call identification, it integrates motion posture data to distinguish between genuine emergencies and environmental interference, significantly improving identification accuracy. Once a fall, a genuine distress call, or a severe health abnormality is confirmed, the rescue response module immediately links with the positioning module to simultaneously send location information and warning signals to guardians and medical institutions, enabling rapid emergency response, gaining crucial time for rescue, and significantly reducing injuries caused by accidents.

[0100] This invention establishes a three-tiered early warning mechanism based on health status characteristics and the number of abnormalities in individual physiological indicators, with different alert methods corresponding to different warning levels. Mild abnormalities only trigger a vibration alert to avoid disturbing the elderly; moderate abnormalities trigger a text message alert to the guardian; and severe abnormalities simultaneously trigger an emergency response from a medical institution / community rescue center. This ensures no risk is overlooked and allows for precise matching of response measures based on the severity of the risk. This tiered design balances the user experience of the elderly, the guardian's right to know, and the rescue priority of medical institutions, making health management and rescue more targeted.

[0101] This invention breaks through the limitations of traditional health monitoring by integrating an AI companionship module. By extracting emotional features such as tone, speed, and volume from speech, it accurately determines the elderly person's emotional state, including happiness, calmness, depression, anxiety, and anger. For negative emotions, it automatically outputs personalized emotional guidance messages and continuously monitors emotional changes until they stabilize, filling a gap in emotional care for the elderly. This integrated design of "physiological monitoring + emotional care" not only protects physical health but also focuses on mental health, enhancing the elderly's sense of well-being and belonging.

[0102] This invention utilizes a communication module to enable real-time transmission of sensor-collected data and processing results to the guardian's mobile phone and the cloud server of medical institutions / community rescue centers, constructing a fully interconnected network linking the elderly, devices, guardians, and medical institutions. Guardians can monitor the elderly's health and location information in real time, while medical institutions can promptly obtain emergency rescue needs and relevant health data, supporting precise rescue efforts. Simultaneously, automatic device malfunction detection and alerts ensure continuous and stable system operation, comprehensively protecting the daily health and emergency safety of the elderly population. Attached Figure Description

[0103] Figure 1 This is a system block diagram of the integrated intelligent system for elderly health protection and emergency rescue of the present invention.

[0104] Figure 2 This is a flowchart illustrating the integrated intelligent system for elderly health protection and emergency rescue of the present invention. Detailed Implementation

[0105] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0106] Please see Figure 1 and Figure 2 As shown, the embodiments of the present invention provide the following technical solutions:

[0107] As an embodiment of the present invention:

[0108] This invention is an integrated intelligent system for elderly health protection and emergency rescue, comprising:

[0109] The sensor module has built-in motion and attitude sensors. The collected analog signals are converted into digital signals by the built-in analog-to-digital converter and then transmitted to the data processing unit.

[0110] The motion and attitude sensor integrates a three-axis accelerometer and a three-axis gyroscope. The three-axis accelerometer is used to collect acceleration values ​​in real time along the X, Y, and Z axes, and these values ​​are denoted as A. x A y A z ; and the use of a three-axis gyroscope to acquire angular velocity values ​​in real time along the X, Y, and Z axes, denoted as ω respectively. x ω y ω z ;

[0111] Using a six-axis data fusion method, the movement posture indicators of the elderly are monitored in real time;

[0112] In this embodiment, the X-axis represents the left-right direction of the elderly person's body, the Y-axis represents the front-back direction of the elderly person's body, and the Z-axis represents the up-down direction of the elderly person's body.

[0113] The data processing unit is used to combine the data collected by the motion and attitude sensors in the sensor module to confirm whether a fall event has occurred. The confirmation method is as follows:

[0114] pass: , calculate the total acceleration A1;

[0115] In this embodiment, it has been tested and verified.

[0116] When the elderly person is at rest, their A1 value is approximately 1. In this embodiment, when their A1 is between 0.9 and 1.1g, it indicates that the elderly person is at rest, where g represents the acceleration due to gravity.

[0117] When an elderly person is engaged in physical activities such as walking or running, A1 > 1;

[0118] When an elderly person falls, A1 will suddenly increase and decrease. In this embodiment, the sudden increase and decrease refer to the impact at the moment of the fall and the stillness after the fall.

[0119] Simultaneously through:

[0120] Calculate the attitude rotation angle θ1 within the pre-specified acquisition time interval;

[0121] In the formula, θ x θ y θ z These represent the rotation angles of the X, Y, and Z axes per unit time, respectively, with Δt being the data acquisition time interval.

[0122] In this embodiment, it has been tested and verified.

[0123] When an elderly person's body posture changes drastically, such as when they fall and their body tilts, θ1 will exceed the corresponding preset angle threshold within a short period of time.

[0124] Among them, the resultant acceleration and the attitude rotation angle are the motion posture indicators of the elderly.

[0125] Next, a dual-parameter determination method using resultant acceleration and attitude rotation angle is adopted, corresponding to preset resultant acceleration threshold A0 and attitude rotation angle threshold θ0;

[0126] When A1≥A0 and θ1≥θ0, and the duration is ≥0.2 seconds and ≤2 seconds, it is initially determined to be a fall event;

[0127] In this embodiment, the resultant acceleration threshold is A0 = 2.5g, where g is the acceleration due to gravity, a commonly used unit of measurement for acceleration, and 1g = 9.8m / s². 2 Angle threshold: θ0 = 60°;

[0128] The resultant acceleration and posture rotation angle are then continuously monitored over the next 3 seconds. If the resultant acceleration A1 drops back and stabilizes within the 0.9g-1.1g range over the next 3 seconds, indicating that the elderly person is at rest, a fall event is confirmed, a fall signal is generated, and the fall signal is transmitted to the guardian.

[0129] If the data returns to normal motion state within the next 3 seconds, it is determined to be a false trigger and the subsequent process will not be triggered;

[0130] In this embodiment, the dual-parameter determination method can effectively reduce the false recognition rate of falls caused by the elderly turning around quickly, bending over, raising their hands, etc., with a false recognition rate of ≤5%.

[0131] This embodiment collects elderly people's movement posture indicators through six-axis data fusion, employs a dual-parameter determination mechanism of resultant acceleration and posture rotation angle, and combines it with subsequent secondary status verification to accurately identify fall events. This solution effectively distinguishes falls from everyday movements such as turning and bending over, significantly reducing the false recognition rate and achieving accurate and rapid identification of fall events. It provides reliable fall monitoring for elderly people living alone, buys critical time for emergency rescue, protects the lives of the elderly, and improves the stability and reliability of the system's fall recognition.

[0132] As a second embodiment of the present invention:

[0133] In specific implementation, compared with Embodiment 1, the only difference between the technical solution of this embodiment and Embodiment 1 is that in this embodiment, the sensor module also has a built-in voice sensor, and the collected analog signal is converted into a digital signal through the built-in analog-to-digital conversion module and then transmitted to the data processing unit.

[0134] The voice sensor uses a high-sensitivity omnidirectional microphone to collect the elderly person's voice signal in real time and convert the voice signal into a digital signal;

[0135] In this embodiment, a high-sensitivity omnidirectional microphone with a pickup range of 0-5 meters, a pickup frequency of 300-3400Hz (covering the frequency range of human speech), a signal-to-noise ratio of ≥40dB, and the ability to effectively filter environmental noise is selected.

[0136] The data processing unit combines the data collected by the voice sensor in the sensor module and uses distress call voice content recognition technology to identify whether there are any abnormal distress calls.

[0137] When the recognition result of the distress voice content recognition technology is distress voice, the data collected by the current motion and posture sensors are acquired simultaneously;

[0138] If the resultant acceleration A1 is in the static range of 0.9g-1.1g and the attitude rotation angle θ1 does not change significantly, that is, the elderly person is in a static state after falling or unable to move independently, then a real emergency rescue situation is confirmed and the rescue process is triggered.

[0139] In this embodiment, distress calls, such as “Help me,” “Help me,” “I’ve fallen,” etc., are recognized using existing mature technologies. This technology has been commercially implemented in multiple fields such as emergency rescue, public safety, and smart security. For example, it has practical applications in scenarios such as earthquake rubble search and rescue, school anti-bullying systems, outdoor emergency terminals, and smart home security. Therefore, it will not be described in detail.

[0140] If the elderly person is in a normal state of motion or in a stable posture, it is determined to be a false recognition caused by environmental interference, and the rescue process is not triggered; this avoids false triggering of distress signals caused by interference from environmental noise, broadcast sounds, conversations, etc., and improves the accuracy of recognition.

[0141] This embodiment adds a voice sensor to the existing embodiment, combining SOS voice recognition with motion posture data to construct a dual verification mechanism. This effectively filters out environmental noise, conversations, and other interference, preventing accidental triggering of rescue efforts, while also supplementing coverage for scenarios where a fall prevents the system from triggering its own response. This design retains the advantages of fall recognition, strengthens proactive SOS response capabilities, broadens the dimensions of emergency SOS triggering, and further enhances the system's comprehensive coverage and accuracy in emergency situations involving the elderly, providing all-round protection for their safety.

[0142] As an embodiment of the present invention:

[0143] In specific implementation, compared with Embodiment 1 and Embodiment 2, the technical solution of this embodiment is to combine the solutions of Embodiment 1 and Embodiment 2. The only difference between the technical solution of this embodiment and Embodiment 1 and Embodiment 2 is that in this embodiment, the sensor module also has a built-in physiological indicator sensor, and the collected analog signal is converted into a digital signal through the built-in analog-to-digital conversion module and then transmitted to the data processing unit.

[0144] The physiological indicator sensor integrates a photoelectric heart rate sensor, an oscillometric blood pressure sensor, an optical blood oxygen saturation sensor, a high-precision body temperature sensor, and a step count sensor;

[0145] Among them, heart rate, blood oxygen saturation, body temperature, and number of steps can be collected continuously in real time, while systolic blood pressure and diastolic blood pressure are collected intermittently and periodically.

[0146] Each sensor operates independently, and the effectively collected data is synchronously uploaded according to a unified timestamp, forming a time-series physiological data set including heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature, and number of steps, denoted as S. i Where i = 1, 2, ..., 6, corresponding to heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature, and number of steps taken, respectively;

[0147] The values ​​of systolic and diastolic blood pressure in the physiological data set are directly taken from the measurement results collected periodically and intermittently at the corresponding timestamps, ensuring that each indicator is accurately aligned in the time dimension;

[0148] Photoelectric heart rate sensors, oscillometric blood pressure sensors, optical blood oxygen saturation sensors, high-precision body temperature sensors, and step count sensors are existing technologies, so they will not be described in detail.

[0149] The data processing unit receives all digital signals from the sensor module and processes them according to the steps of data preprocessing, multimodal data fusion, deep learning model analysis, health status assessment, and risk warning.

[0150] The data preprocessing steps are as follows:

[0151] Data preprocessing involves cleaning and standardizing physiological index data and motion and posture data. The specific implementation of each step is as follows:

[0152] Step A1, Data Cleaning:

[0153] The 3σ principle is used to identify and remove outliers. The method is as follows:

[0154] when If the value is not found, it is considered an outlier and is removed. At the same time, the average of the two adjacent valid data values ​​is used to fill the gap.

[0155] Where G represents the variables of each data indicator, and it is referred to as S. i Any one of A1 and θ1;

[0156] If the number of abnormal values ​​exceeds 5 in 10 consecutive data collections, or if 5 or more abnormal values ​​occur consecutively, the sensor is preliminarily determined to be abnormal.

[0157] The data collection process is then restarted and a second verification is performed. If the second verification result still meets the above abnormal conditions, the sensor is ultimately determined to be faulty, and a device fault reminder is sent to the elderly and their caregivers.

[0158] In the comparison formula, G is a single data value collected by the sensor; GP is the average value of 10 consecutive data collections of that data indicator; σ is the standard deviation of 10 consecutive data collections of that data indicator, and i0 is the index of the corresponding data value collected 10 times consecutively for this data indicator;

[0159] Step A2, Standardization Processing:

[0160] Obtain the normal range of values ​​for each data indicator, extract the corresponding maximum and minimum values, and denot them as G. max and G min ;

[0161] Combining the results of data cleaning, and employing the min-max normalization method: This maps all data values ​​of each data indicator to the range of 0-1.

[0162] In the formula, G b Data values ​​that have been standardized;

[0163] After standardization, all data are comparable, providing a foundation for subsequent multimodal data fusion;

[0164] The steps for multimodal data fusion are as follows:

[0165] First, the preprocessed physiological index data, motion and posture data are fused to form a unified multidimensional feature dataset, denoted as F. p ;

[0166] Among them, F p The data values ​​for each data indicator are standardized, p=1, 2, ..., 6, 7, 8, where p=1, 2, ..., 6 correspond to the six physiological indicators of the elderly: heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature, and number of steps; and p=7, 8 correspond to the two movement posture indicators of the elderly: resultant acceleration and posture rotation angle.

[0167] Extract the pre-defined weighting coefficients based on health management standards and clinical data for the elderly, and denote them as β. p Furthermore, the sum of all weight coefficients corresponding to the multidimensional feature data is 1.

[0168] In this embodiment, the weighting coefficients are set according to the importance of routine health indicators in geriatric medicine: heart rate and blood oxygen have the highest weight, followed by exercise posture, and step count has the lowest weight. At the same time, the weighting coefficients can be adaptively adjusted according to the user's age and underlying diseases, and can be set to β1=0.2, β2=0.15, β3=0.15, β4=0.2, β5=0.1, β6=0.05, β7=0.08, and β8=0.07.

[0169] Subsequently passed:

[0170] Calculate the comprehensive feature value F after fusing the multidimensional feature datasets;

[0171] The steps for deep learning model analysis are as follows:

[0172] Long Short-Term Memory (LSTM) network is used as the deep learning model;

[0173] The core improvement of this embodiment lies in optimizing the model input format and training adaptation strategy to take into account the temporal characteristics of elderly health data, rather than improving the structure of the LSTM network itself. Furthermore, the basic structure of the LSTM network is a well-known technology and will not be described in detail here.

[0174] Step S1: Training dataset construction:

[0175] The training dataset is derived from ≥100,000 clinical health data and daily monitoring data of the elderly, including physiological indicator data and movement posture data of elderly people of different ages and with different underlying medical histories, and each data point is labeled with the corresponding health status label;

[0176] The health status labels are divided into four levels: normal, mildly abnormal, moderately abnormal, and severely abnormal.

[0177] In this embodiment, data annotation is performed by professional medical personnel to ensure the validity and accuracy of the dataset;

[0178] Step S2, Model Input and Adaptation:

[0179] The combined feature values ​​obtained after fusing multimodal data are used to form a time series F. t The data is then input into the LSTM network model according to a pre-defined time window.

[0180] Where t = 1, 2, ..., n, n is the number of data collections. In this embodiment, the window length is set to 5 minutes.

[0181] During model training, overfitting is avoided through adaptive learning rate adjustment and early stopping mechanism, ensuring the model's generalization ability in elderly health data scenarios;

[0182] Among them, adaptive learning rate adjustment: the initial learning rate is set to 0.001, and the Adam optimizer is used;

[0183] Early stopping mechanism: Training is stopped if the validation set loss does not decrease for 5 consecutive rounds;

[0184] Step S3, Model Output:

[0185] The model outputs the health status characteristic value H of the elderly.

[0186] The value of H ranges from [0,1]. The closer the value of H is to 1, the better the health condition of the elderly person; the closer the value of H is to 0, the worse the health condition.

[0187] The model has been tested and verified to have an accuracy rate of ≥95% in determining health status on the test set, which meets the needs of practical applications. The test set consists of 30,000 health data points of elderly people, which are independent of the training set.

[0188] The steps for health status assessment and risk warning are as follows:

[0189] The health status characteristic value H is combined with the standardized data values ​​G corresponding to each individual physiological indicator. b The classification method is as follows:

[0190] First, the elderly's health status is divided into four levels: normal, mildly abnormal, moderately abnormal, and severely abnormal. Corresponding preset health status thresholds H1, H2, and H3 are extracted, and the range [0.2, 0.8] is used as the standardized data value G. b The normal range of values;

[0191] In this embodiment, H1, H2, and H3 are set to values ​​of 0.8, 0.6, and 0.4, respectively.

[0192] The judgment criteria are as follows:

[0193] When H ≥ 0.8, and the standardized values ​​of all individual physiological indicators G b If all values ​​are within the normal range of [0.2, 0.8], the elderly person's health condition is considered normal, meaning there is no health risk and no warning is issued.

[0194] When 0.6 ≤ H < 0.8, or the G of a single physiological indicator b If the value exceeds the normal range of [0.2, 0.8], the elderly person's health condition is judged as slightly abnormal, that is, there is a slight health risk, and a level one warning is issued.

[0195] The first-level warning is a vibration alert, without a buzzer.

[0196] When 0.4 ≤ H < 0.6, or 2-3 individual physiological indicators G b If the value exceeds the normal range of [0.2, 0.8], the elderly person's health condition is judged as moderately abnormal, that is, there is a moderate health risk, and a level 2 warning is issued.

[0197] The Level 2 warning system includes a vibration alert combined with a low-volume buzzer, and also sends a text message reminder to the guardian.

[0198] When H < 0.4, or ≥ 4 individual physiological indicators G b If the value exceeds the range [0.2, 0.8], the elderly person's health condition is classified as severely abnormal, indicating a serious health risk, and a level three warning is issued.

[0199] The Level 3 warning system involves strong vibration and a loud buzzer, while simultaneously sending an emergency alert to guardians and medical institutions / community rescue centers.

[0200] This embodiment integrates the functions of the previous two embodiments, adding a multi-dimensional physiological indicator sensor to achieve multimodal fusion of physiological and movement posture data. Data quality is ensured through data cleaning and standardization, and a weighted feature fusion and optimized deep learning model are combined to achieve accurate classification and assessment of the elderly's health status. Tiered warnings are provided based on risk levels, upgrading from passive rescue to proactive health management. It also features sensor anomaly detection, significantly improving the system's personalization and professionalism in health management, meeting the diverse health needs of the elderly population.

[0201] As an embodiment of the present invention:

[0202] In specific implementation, compared with Embodiments 1, 2, and 3, the difference between this embodiment and Embodiments 1, 2, and 3 lies only in that this embodiment also includes:

[0203] The communication module is used to transmit the data collected by the sensor module and analyzed by the data processing unit to the guardian's mobile application and the cloud server of the medical institution / community rescue center;

[0204] In this embodiment, the communication module integrates three major communication modules: 4G / 5G mobile communication, Wi-Fi 2.4G / 5G dual-mode, and Bluetooth 5.0. It supports independent operation in multiple modes and automatic seamless switching, adapts to different transmission scenarios such as long distance, short distance at home, and close-range interaction with low-power devices in the vicinity, and has low power consumption characteristics to effectively ensure the device's battery life.

[0205] The rescue response module is used to generate an emergency trigger signal when the data processing unit identifies and confirms that a fall has occurred, there are abnormal distress calls, and the elderly person's health condition is severely abnormal.

[0206] The emergency trigger signal is sent to the guardian and medical institution / community rescue center via the communication module. When the emergency trigger signal is triggered, a location acquisition command is sent to the positioning module to obtain the elderly person's real-time location information. The real-time location information and the emergency trigger signal are then sent to the guardian and medical institution / community rescue center via the communication module.

[0207] In this embodiment, a manual cancellation function is also provided to avoid accidental triggering in emergency situations;

[0208] If no emergency occurs, the elderly person can press the emergency call button on the hardware device for 3 seconds to cancel the emergency rescue process and the system will return to normal operation.

[0209] If no manual cancellation signal is detected within 10 seconds after the rescue process is initiated, the emergency is confirmed to be real, and the subsequent rescue information sending process is initiated.

[0210] Meanwhile, if no feedback is received from the guardian or medical institution / community rescue center within 10 seconds, the system will automatically dial the number of the guardian or medical institution / community rescue center. Once the call is connected, a voice prompt will automatically play: "Your family member is in an emergency. Please check the rescue information and take rescue measures immediately."

[0211] In this embodiment, the emergency trigger signal can be denoted as SOS. q Where q=1, 2, 3, corresponding to a fall, an abnormal distress call, and a severe health condition, respectively; this helps rescuers quickly identify the type of emergency, accurately grasp the core situation on site, and then develop targeted rescue plans and take corresponding first aid measures, thereby improving the efficiency and accuracy of the rescue response.

[0212] This embodiment improves the communication and rescue response mechanism, integrating multi-mode communication modules to achieve seamless switching across multiple scenarios and low-power operation, ensuring stable data transmission and device battery life. It innovatively designs an "emergency trigger - location acquisition - manual cancellation - multi-level notification" process, improving rescue accuracy through real-time location sharing, avoiding false rescues through manual cancellation, and ensuring rapid response from guardians and rescue centers through multi-channel notifications. Emergency events are categorized by type to facilitate targeted handling, building a highly efficient full-chain rescue system, strengthening the system's practical application capabilities, and providing solid protection for emergency rescue of the elderly.

[0213] As a fifth embodiment of the present invention:

[0214] In specific implementation, compared with Embodiment 1, Embodiment 2, Embodiment 3 and Embodiment 4, the technical solution of this embodiment is to combine the solutions of Embodiment 1, Embodiment 2, Embodiment 3 and Embodiment 4. The only difference between the technical solution of this embodiment and Embodiment 1 to Embodiment 4 is that in this embodiment, an AI companion module is also included, and the voice signal is converted into a digital signal and then transmitted in two paths: one path is transmitted to the AI ​​companion module and the other path is transmitted to the data processing unit.

[0215] The AI ​​companion module is used to combine the data collected by the voice sensor in the sensor module to perform emotion analysis and personalized emotional guidance for the elderly.

[0216] The specific implementation is as follows:

[0217] Extract the tone (P), speech rate (S), and volume (L) of the elderly person's speech signal as emotional features.

[0218] In this embodiment, the pitch is obtained by extracting the fundamental frequency of the speech using the autocorrelation method and taking the average value of the fundamental frequency across multiple frames, reflecting the pitch of the speech; the speech rate is the ratio of the number of words to the duration within a statistically valid speech segment, reflecting the speed of speaking; and the volume is obtained by calculating the sound pressure level through the energy of the speech signal and taking the average value across multiple frames, reflecting the loudness of the sound.

[0219] Simultaneously, the extraction criteria are based on the normal range of tone, speech rate, and volume values ​​pre-recorded and calibrated from the elderly's normal speech signals, and these are sequentially denoted as [P]. min P max ]、[S min S max ]、[L min L max ];

[0220] Subsequently passed:

[0221] Calculate the comprehensive value E of emotional features in the elderly person's speech signal;

[0222] In the formula: γ P γ S γ L The preset weighting coefficients are based on tone, speaking speed, and volume.

[0223] In this embodiment, γ P γ S γ L The values ​​were set to 0.4, 0.3, and 0.3 respectively to highlight the influence of intonation on emotional expression.

[0224] The comprehensive value of emotional characteristics E ranges from [0,1].

[0225] Then, by combining the emotional characteristics of the elderly group with the preset emotional threshold, the emotional state in the elderly’s voice signal was determined.

[0226] When 0.8≤E≤1, it indicates that the emotional state in the elderly person's voice signal is happy;

[0227] When 0.6 ≤ E < 0.8, it indicates that the emotional state in the elderly person's speech signal is calm;

[0228] When 0.3≤E<0.6, it indicates that the emotional state in the elderly person's voice signal is depressed;

[0229] When 0.1 ≤ E < 0.3, it indicates that the emotional state in the elderly person's speech signal is anxiety;

[0230] When 0 ≤ E < 0.1, it indicates that the emotional state in the elderly person's speech signal is anger;

[0231] When an elderly person's emotional state is detected as depressed, anxious, or angry, the corresponding preset emotional guidance script is output:

[0232] For those experiencing low moods, say something like, "I know you might be a little unhappy right now, but that's okay. I'm here for you. Tell me what's on your mind."

[0233] To address anxiety, say, "Don't rush, speak slowly. No matter what, there's always a solution. I'm always here for you."

[0234] Regarding expressing anger, say, "Take a deep breath and relax. Tell me what's wrong, I'm listening."

[0235] Continuously monitor the comprehensive value of emotional characteristics E. When E≥0.6, that is, the emotional state returns to calm or happy, then the emotional counseling is terminated and normal interactive functions are restored.

[0236] The AI ​​companion module's emotional feature extraction results, comprehensive emotional value, and emotional state determination information can be transmitted to the guardian's mobile application via the communication module.

[0237] This embodiment adds an AI companionship module, which extracts voice emotion features to build an assessment model and accurately identifies various emotional states of the elderly. For negative emotions, it automatically triggers personalized emotional guidance, providing psychological comfort through warm interaction and filling the emotional gap in the elderly population. This module works in conjunction with the existing rescue and monitoring functions to achieve integrated "physiological protection + emotional care," ensuring the safety of the elderly while meeting their spiritual needs, improving their quality of life and happiness, and enriching the system's humanistic care functions.

[0238] As an embodiment of the present invention:

[0239] In specific implementation, compared with Embodiment 1, Embodiment 2, Embodiment 3, Embodiment 4 and Embodiment 5, the technical solution of this embodiment is to combine the solutions of Embodiment 1 to Embodiment 5.

[0240] This embodiment integrates all the core functions of the previous five embodiments to construct an elderly health protection system that combines fall detection, active SOS, health management, emergency rescue, and AI-powered emotional care. Through multi-module collaborative linkage, it achieves full-scenario coverage from passive emergency response to proactive prevention, and from physiological monitoring to emotional care, comprehensively meeting the diverse safety, health, and emotional needs of the elderly. It provides integrated intelligent protection for elderly people living alone and those of advanced age, significantly improving the safety, health, and well-being of their lives, and possesses extremely high social value and broad application prospects.

[0241] It should be stated that all user data collected in this application was collected with the user's consent and authorization, and the use of user data is legal and compliant, and the use and processing of user data comply with the relevant laws, regulations and standards of the relevant regions.

[0242] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0243] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0244] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0245] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. An integrated intelligent system for elderly health protection and emergency rescue, characterized in that: include: The sensor module integrates physiological index sensors, motion and posture sensors, and voice sensors to collect physiological index data corresponding to the elderly's heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature, and number of steps, as well as the elderly's body acceleration and angular velocity values, and the elderly's voice signals. The analog signals collected by the sensor module are converted into digital signals by the built-in analog-to-digital converter and then transmitted to the data processing unit; the data collected by the voice sensor are transmitted to the AI ​​companion module and the data processing unit respectively. The data processing unit is used to calculate the elderly's movement posture indicators based on the data transmitted by the sensor module, and to confirm whether a fall has occurred; and to identify whether there are abnormal distress calls through distress voice content recognition technology; at the same time, it performs data preprocessing, multimodal data fusion, deep learning model analysis, health status assessment and risk warning on the elderly's movement posture indicators and physiological indicators, and outputs the elderly's health status characteristic values ​​and provides graded warnings. The communication module is used to transmit the data collected by the sensor module and analyzed by the data processing unit to the guardian's mobile application and the cloud server of the medical institution / community rescue center; The rescue response module is used to generate emergency trigger signals when a fall occurs, there are abnormal distress calls, or the health condition is severely abnormal. The AI ​​companion module is used to extract emotional features from voice signals and calculate an emotional comprehensive value. The emotional comprehensive value is used to determine the elderly person's emotional state, and then personalized emotional guidance is implemented based on the emotional state.

2. The integrated intelligent system for elderly health protection and emergency rescue according to claim 1, characterized in that: The methods for confirming a fall incident are as follows: First, the acceleration values ​​collected in real time by the triaxial accelerometer are extracted, and the resultant acceleration is calculated by vector synthesis and denoted as A1. Simultaneously, based on the angular velocity values ​​collected in real time on the three axes by the three-axis gyroscope, the rotation angle of each axis within a pre-specified acquisition time interval is calculated, and then the rotation angles of the three axes are synthesized to obtain the posture rotation angle of the elderly person's body within that time interval, which is recorded as θ1. Among them, the resultant acceleration and the attitude rotation angle are the motion posture indicators of the elderly. Next, a dual-parameter determination method using resultant acceleration and attitude rotation angle is adopted, corresponding to preset resultant acceleration threshold A0 and attitude rotation angle threshold θ0; When A1≥A0 and θ1≥θ0, and the duration is ≥0.2 seconds and ≤2 seconds, it is initially determined to be a fall event; The resultant acceleration and rotation angle are then continuously monitored over the next 3 seconds. If the resultant acceleration A1 drops and stabilizes within the 0.9g-1.1g range over the next 3 seconds, indicating that the elderly person is at rest, a fall event is confirmed, a fall signal is generated, and the fall signal is transmitted to the caregiver.

3. The integrated intelligent system for elderly health protection and emergency rescue according to claim 2, characterized in that: When the data processing unit identifies distress-type voice content using SOS recognition technology: When the recognition result of the distress voice content recognition technology is distress voice, the data collected by the current motion and posture sensors are acquired simultaneously; If the resultant acceleration A1 is in the range of 0.9g-1.1g and the change in attitude rotation angle θ1 is ≤5°, then a real emergency distress call is confirmed. Where g is the gravitational acceleration, which is the commonly used unit of measurement for acceleration.

4. The integrated intelligent system for elderly health protection and emergency rescue according to claim 3, characterized in that: The physiological indicator sensor integrates a photoelectric heart rate sensor, an oscillometric blood pressure sensor, an optical blood oxygen saturation sensor, a high-precision body temperature sensor, and a step count sensor. The collected data is synchronously uploaded according to a unified timestamp, forming a time-series physiological data set containing heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature, and steps, denoted as S. i , where i=1, 2, ..., 6, corresponding to heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature, and number of steps, respectively.

5. The integrated intelligent system for elderly health protection and emergency rescue according to claim 4, characterized in that: The data preprocessing steps are as follows: Data preprocessing involves cleaning and standardizing physiological index data and motion and posture data. The specific implementation of each step is as follows: Step A1, Data Cleaning: The 3σ principle is used to identify and remove outliers. The method is as follows: For a single data value G, the difference between it and the average of 10 consecutive data collections of the corresponding data indicator is calculated. If the absolute value of the difference is greater than 3 times the standard deviation calculated from the 10 consecutive data collections, the data value is determined to be an outlier and is removed. At the same time, the average of the two adjacent valid data values ​​is used to complete the value. Where G represents the variables of each data indicator, and it is referred to as S. i Any one of A1 and θ1; If the number of abnormal values ​​exceeds 5 in 10 consecutive data collections, or if 5 or more abnormal values ​​occur consecutively, the sensor is preliminarily determined to be abnormal. The data collection process is then restarted and a second verification is performed. If the second verification result still meets the above abnormal conditions, the sensor is ultimately determined to be faulty, and a device fault reminder is sent to the elderly and their caregivers. Step A2, Standardization Processing: Obtain the normal value range for each data indicator, and use the maximum and minimum values ​​within this range as benchmarks. Subtract the minimum value of the normal value range from the cleaned value of a single data point, and then divide by the difference between the maximum and minimum values ​​within the normal value range. This determines the standardized data value for each data indicator, and is denoted as G. b .

6. The integrated intelligent system for elderly health protection and emergency rescue according to claim 5, characterized in that: The steps for multimodal data fusion are as follows: First, the preprocessed physiological index data, motion and posture data are fused to form a unified multidimensional feature dataset, denoted as F. p ; Among them, F p The data values ​​for each data indicator are standardized, p=1, 2, ..., 6, 7, 8, where p=1, 2, ..., 6 correspond to the six physiological indicators of the elderly: heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature, and number of steps; and p=7, 8 correspond to the two movement posture indicators of the elderly: resultant acceleration and posture rotation angle. Extract the pre-defined weighting coefficients based on health management standards and clinical data for the elderly, and denote them as β. p Furthermore, the sum of all weight coefficients corresponding to the multidimensional feature data is 1. Subsequently passed: The comprehensive feature value F after the fusion of the multidimensional feature dataset is calculated.

7. The integrated intelligent system for elderly health protection and emergency rescue according to claim 6, characterized in that: The deep learning model analysis employs a Long Short-Term Memory (LSTM) network as the deep learning model, using ≥100,000 clinical health data and daily monitoring data of elderly individuals labeled with health tags as the training set. The comprehensive feature values ​​are then combined to form a time series F. t The data is divided into pre-set time windows and input, and trained using the Adam optimizer and early shutdown mechanism. Finally, the health status characteristic value H of the elderly is output. Where t=1, 2, ..., n, n is the number of times data is collected, and the health status label is divided into four levels: normal, mildly abnormal, moderately abnormal, and severely abnormal.

8. The integrated intelligent system for elderly health protection and emergency rescue according to claim 7, characterized in that: The steps for health status assessment and risk warning are as follows: The health status characteristic value H is combined with the standardized data values ​​G corresponding to each individual physiological indicator. b The classification is determined by extracting the corresponding preset health status thresholds H1, H2, and H3, and taking [0.2, 0.8] as the standardized data value G. b The normal range of values; The judgment criteria are as follows: When H≥H1, and the standardized values ​​of all individual physiological indicators G b If all values ​​are within [0.2, 0.8], the elderly person's health condition is considered normal, and no warning is issued. When H2 ≤ H < H1, or the G of a single physiological indicator b If the value exceeds the normal range of [0.2, 0.8], the elderly person's health condition will be judged as slightly abnormal, and a first-level warning with only vibration reminder will be issued. When H3 ≤ H < H2, or G of 2-3 individual physiological indicators b If the value exceeds the normal range of [0.2, 0.8], the elderly person's health condition will be judged as moderately abnormal, and a level two warning will be issued, including vibration, low-volume buzzer reminder, and SMS reminder sent to the caregiver. When H < H3, or ≥ 4 individual physiological indicators G b If the condition exceeds the range of [0.2, 0.8], the elderly person's health condition will be judged as severely abnormal, and a three-level warning will be issued, including strong vibration, high-volume buzzer reminder, and emergency reminder sent to the guardian and medical institution / community rescue center.

9. The integrated intelligent system for elderly health protection and emergency rescue according to claim 8, characterized in that: The sensor module also includes a positioning module, which is used to obtain the elderly person's real-time location information; When the AI ​​companion module generates an emergency trigger signal, it simultaneously sends a location acquisition command to the positioning module to obtain the elderly person's real-time location information. The real-time location information and the emergency trigger signal are then sent to the guardian and medical institutions / community rescue centers via the communication module.

10. The integrated intelligent system for elderly health protection and emergency rescue according to claim 1, characterized in that: The specific implementation of the AI ​​companion module is as follows: Extract the tone (P), speech rate (S), and volume (L) of the elderly person's speech signal as emotional features. Simultaneously, the extraction criteria are based on the normal range of tone, speech rate, and volume values ​​pre-recorded and calibrated from the elderly's normal speech signals, and these are sequentially denoted as [P]. min P max ]、[S min S max ]、[L min L max ]; Subsequently passed: The comprehensive value E of emotional features in the elderly person's speech signal was calculated. In the formula: γ P γ S γ L The preset weighting coefficients are based on tone, speaking speed, and volume. Based on the comprehensive value of emotional characteristics in the elderly’s voice signals, the elderly’s emotional state is divided into five levels: happy, calm, depressed, anxious, and angry. When an elderly person's emotional state is detected to be depressed, anxious, or angry, the corresponding preset emotional guidance script is output; at the same time, the comprehensive value of emotional characteristics is continuously monitored until the emotional state returns to calm or happy, at which point the output of the emotional guidance script stops.

Images