Health monitoring method and monitoring system for elderly people living alone

Abstract

The invention relates to a health monitoring method and detection system for elderly people living alone. The method mainly comprises the steps that data are collected through sensors, various types of sensors can be installed at different places in a room, and based on the collected data, the following recognition is carried out: activity recognition based on audio data, position recognition based on an infrared sensor, posture recognition based on a physiological function sensor and gait recognition based on the physiological function sensor. The invention relates to identification of heart rate and respiratory rate based on a thin film piezoelectric sensor. The method can continuously monitor the health condition of the elderly.

Description

Technical Field

[0001] The present invention relates to the field of health detection, and in particular to a health monitoring method and a monitoring system for elderly people living alone. Background Art

[0002] The elderly face a variety of health problems, and the demand for daily health monitoring is increasing. There are many existing devices for testing the elderly, but none of them is convenient for continuously monitoring the activities of the elderly. Summary of the invention

[0003] Based on this, a method for monitoring the health of elderly people living alone is provided, which is convenient for monitoring the elderly people continuously.

[0004] A method for monitoring the health of an elderly person living alone, comprising:

[0005] Collecting data through sensors,

[0006] Based on the collected data, the following recognition is performed: Activity recognition based on audio data.

[0007] In one of the embodiments, the activity recognition based on audio data is specifically implemented by applying the following modules: an audio event segmentation module, a feature extraction module and a classifier module.

[0008] In one embodiment,

[0009] The audio event segmentation module is used to achieve the following objectives: segmenting audio data into specific behavior events,

[0010] The feature extraction module is used to achieve the following objectives: generate a Log-Mel spectrogram based on audio data,

[0011] The classifier module is used to achieve the following purpose: output classification results based on the Log-Mel spectrum.

[0012] In one embodiment,

[0013] Specifically, the model used by the classifier module includes four blocks, each block contains a convolution layer and a pooling layer. After 4 blocks, a 512-dimensional embedding vector is extracted, followed by average pooling and reshaping operations, and then a new 512-dimensional embedding vector is obtained. Then the new 512-dimensional embedding vector is input into the attention mechanism module. The attention mechanism module consists of a multi-head self-attention mechanism and a feedforward neural network with residual connections and layer normalization. The attention mechanism module is followed by a flattening processing layer, a fully connected layer, and a softmax classification layer.

[0014] In one embodiment,

[0015] The sensors are of various types, and the results are obtained by comprehensive analysis of data obtained by the various sensors.

[0016] In one embodiment, the position recognition based on the infrared sensor is also included, which specifically includes:

[0017] Get the infrared sensor data,

[0018] Eliminate outliers due to infrared sensors being falsely triggered by sunlight,

[0019] Fill in the missing data,

[0020] The data obtained by the infrared sensor is processed using a clustering algorithm to identify the location of elderly people living alone.

[0021] In one embodiment, gesture recognition based on physiological function sensors is also included, specifically including:

[0022] Get data from physiological function sensors,

[0023] Recognize the posture of elderly people living alone based on data from physiological function sensors.

[0024] In one embodiment, gait recognition based on physiological function sensors is also included, specifically including:

[0025] Get data from physiological function sensors,

[0026] Identify the gait of elderly people living alone based on data from physiological function sensors.

[0027] In one embodiment, the recognition of heart rate and breathing rate based on a thin film piezoelectric sensor is also included, specifically including:

[0028] Get data from the thin-film piezoelectric sensor mounted on the bed,

[0029] The heart rate and breathing rate of elderly people living alone are identified based on data from thin film piezoelectric sensors.

[0030] A health monitoring system for elderly people living alone for implementing a health monitoring method for elderly people living alone comprises a data acquisition module and a data analysis module. The data acquisition module is used to collect data through sensors, and the data analysis module is used to identify and analyze based on the collected data.

[0031] The beneficial effects of this application are:

[0032] The present application proposes a non-invasive, continuous and comprehensive health monitoring method and system designed for the elderly. It is conducive to ensuring all-weather monitoring without interfering with their daily lives. The system combines a variety of smart sensors, such as audio sensors, infrared sensors, thin film piezoelectric sensors and physiological function sensors. It can identify human activities and provide intelligent sleep monitoring and physiological parameter evaluation. By evaluating 30 days of data from 10 subjects over 65 years old, the method of the present application showed high accuracy in activity recognition, verifying its effectiveness in actual home care environments. The method of the present application ensures effective health monitoring without disturbing the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a schematic diagram of a method for monitoring the health of elderly people living alone according to an embodiment of the present application.

[0034] Figure 2 A schematic diagram of a classifier module according to an embodiment of the present application. DETAILED DESCRIPTION

[0035] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0036] like Figure 1 As shown, an embodiment of the present application provides a health monitoring method for elderly people living alone, including: collecting data through sensors, various types of sensors can be installed in different places indoors, and performing the following identifications based on the collected data: activity recognition based on audio data, position recognition based on infrared sensors, posture recognition based on physiological function sensors, gait recognition based on physiological function sensors, and heart rate and breathing rate recognition based on thin film piezoelectric sensors.

[0037] Specifically, the audio data is data obtained by an audio sensor. Audio sensors can be installed at different locations in the user's home to obtain audio data.

[0038] In one of the embodiments, the activity recognition based on audio data is specifically implemented by applying the following modules: an audio event segmentation module, a feature extraction module and a classifier module.

[0039] Specifically, this application assumes that there is an audio sensor in the area of ​​interest, and the goal is to identify human activities from the uninterrupted audio stream in the environment of the area. The specific activities that are attempted to be identified are labeled A_1, A_2, A_3, ..., A_s, where the variable s represents the total number of activity categories in the predefined set S. The framework of the proposed activity recognition system based on audio data consists of three main parts: audio event segmentation module, feature extraction module and classifier module.

[0040] Specifically, the audio event separation module is used to achieve the following purposes: segment the audio data into specific behavioral events. In order to segment the audio data into specific behavioral events, each file is accurately marked and divided using Praat software. The purpose is to automatically identify the activity behaviors of elderly people living alone and classify the audio data in real time. Therefore, during the labeling process, these audio data are divided into 13 different categories: washing clothes, snoring, coughing, listening to music, watching TV, talking, footsteps, cooking, cutting vegetables, dishes, doing housework, noise and environmental sounds. Environmental sounds refer to the sounds produced when the elderly are inactive, including inherent sensor noise, animal calls, car horns, etc.

[0041] The feature extraction module is used to achieve the following purpose: generating a Log-Mel spectrogram based on audio data. Existing methods can be applied here to achieve this.

[0042] The classifier module is used to achieve the following purpose: output classification results based on the Log-Mel spectrum.

[0043] On the basis of the above, if Figure 2 As shown, the model used in the above classifier module includes four blocks, each block contains a convolution layer and a pooling layer. After 4 blocks, a 512-dimensional embedding vector is extracted, followed by average pooling and reshaping operations, and then a new 512-dimensional embedding vector is obtained. Then the new 512-dimensional embedding vector is input into the attention mechanism module. The attention mechanism module consists of a multi-head self-attention mechanism and a feedforward neural network with residual connections and layer normalization. The attention mechanism module is followed by a flattening processing layer, two fully connected layers, and a softmax classification layer.

[0044] Specifically, the extraction of a 512-dimensional embedding vector after the above 4 blocks is specifically implemented by the following method: This application uses the original architecture of the VGGish model, that is, the 4 modules of this application, of which there are 1 to 8 layers, and utilizes the pre-trained weights of the original architecture. After removing the fully connected layer, a 512-dimensional VGGish embedding vector is obtained.

[0045] Specifically, the advantages of the above-mentioned classifier module of the present application are:

[0046] Improved feature extraction capability: The above classifier module of the present application adopts a deeper network structure and a more complex convolutional layer design to enhance the ability to extract local features. This enables the model to better capture complex image or audio features and improves classification accuracy.

[0047] Strong adaptability: By using pre-trained weights and transfer learning, the above-mentioned classifier module of the present application can adapt to different tasks and data sets, so that the model can also achieve good performance on small data sets.

[0048] Attention mechanism: The introduction of the attention mechanism enables the model to automatically focus on important features, thereby improving the classification effect. This mechanism helps to improve the interpretability of the model and make the output more targeted.

[0049] In one of the embodiments, on the basis of the above, it also includes location recognition based on infrared sensors, specifically including: acquiring data from the infrared sensor, eliminating abnormal values ​​obtained due to the infrared sensor being mistakenly triggered by sunlight, supplementing missing data, and using a clustering algorithm to process the data obtained by the infrared sensor to identify the location of the elderly living alone.

[0050] Specifically, in order to address potential error events, a data correction process for the infrared sensors was performed. This correction process includes two specific steps: outlier elimination and missing data processing. First, outliers that occur when the sensor on the balcony is falsely triggered by sunlight are eliminated. The specific approach is:

[0051] Data monitoring: Continuously monitor the data output by the sensor to identify spikes or mutations that are obviously abnormal. Threshold setting: Set a reasonable threshold and mark the sensor reading as an outlier when it exceeds the threshold. Outlier identification: Automatically detect and mark abnormal readings that appear repeatedly within a specific time period. Data cleaning: Delete or replace the marked outliers to ensure that the data set contains only reliable readings.

[0052] By identifying and removing these outliers, we ensure that the sensor data accurately represents the real events detected in the monitoring area.

[0053] Furthermore, the present application also handles the situation of sensor data loss, which may occur when the elderly are indoors and the sensors are not activated. In this case, an event indicating activation is introduced in the living room. An event indicating activation refers to a signal or state introduced during the sensor monitoring process to fill in missing data or mark a specific event. In the application of infrared sensors, it mainly has the functions of missing data supplementation, status identification, and ensuring data integrity. This inserted event is placed between the previous and subsequent activation events to maintain the temporal continuity of the data. By implementing these optimization measures, the accuracy and integrity of the data collected by the infrared sensor are significantly improved.

[0054] After the above preprocessing steps, the infrared sensor data is processed using a clustering algorithm, that is, adjacent similar values ​​are iteratively grouped until similar groups can no longer be merged, thereby identifying the daily activity patterns of the elderly.

[0055] On the basis of the above, in one of the embodiments, it also includes posture recognition based on physiological function sensors, specifically including: obtaining data from the physiological function sensors, and identifying the posture of elderly people living alone based on the data from the physiological function sensors. The physiological function sensor here is a device for monitoring and collecting physiological data of the human body, and is a prior art. The physiological function sensor can detect the posture, movement and physiological parameters of an individual, such as heart rate, breathing rate and activity status, in real time. The physiological function sensor helps to monitor an individual by identifying specific biological signals. The physiological function sensor here is a sensor of the prior art. The posture recognition of elderly people living alone based on the data from the physiological function sensors can also be achieved by applying the existing technology. Here, the prior art is integrated into the solution of the present application to identify and track the posture of the elderly.

[0056] In one of the embodiments, gait recognition based on physiological function sensors is also included, specifically including: acquiring data from the physiological function sensors, and recognizing the gait of elderly people living alone based on the data from the physiological function sensors.

[0057] Specifically, gait analysis in the elderly is particularly valuable due to the challenges they face, such as falls and frailty. Frail individuals often experience difficulty walking, which may further increase the risk of falls. By performing gait analysis, it becomes feasible to identify objective parameters and patterns associated with these pathological conditions. This can allow for early detection of gait abnormalities and provide the opportunity for timely intervention to mitigate the risk of falls and related events.

[0058] This application uses physiological function sensors to detect the posture of individuals, calculate parameters such as step frequency and balance ratio, and perform comprehensive gait analysis. For the gait pattern of human walking, T1 and T2 represent the time intervals of the left and right feet contacting the ground continuously. Step frequency is defined as the number of times a single foot contacts the ground in a unit time, in steps per second, or equal to 1 / T1. On the other hand, the balance rate is defined as the ratio of the time each foot contacts the ground, represented by T1 / T2. The physiological function sensor is installed in a suitable position to capture the individual's motion data. Then, the sensor collects the individual's posture information during walking in real time, including the position of the center of gravity and the movement of various parts of the body. Next, the analog signal of the sensor is converted into a digital signal using a data acquisition system for further analysis. In the data processing stage, filtering technology is used to remove noise and ensure the accuracy of the signal. Subsequently, the signal is analyzed by an algorithm to calculate key parameters such as step frequency (number of steps per minute) and the balance ratio of the body (the proportion of time each foot contacts the ground). Existing algorithms can be applied here. Finally, the calculated parameters such as step frequency and balance ratio are comprehensively analyzed to evaluate the individual's gait characteristics and athletic ability. This data can not only be used for health monitoring, but also help identify potential movement problems and provide targeted interventions.

[0059] In one of the embodiments, it also includes the recognition of heart rate and respiratory rate based on the thin film piezoelectric sensor, specifically including: obtaining data from the thin film piezoelectric sensor installed on the bed, and identifying the heart rate and respiratory rate of the elderly living alone based on the data from the thin film piezoelectric sensor.

[0060] Specifically, the bed is equipped with a highly sensitive thin-film piezoelectric sensor to intelligently monitor physiological indicators during nighttime rest. The sensor is placed under the sheets and can accurately detect subtle pressure fluctuations caused by basic physiological activities, including heart beats, breathing, and body movements. Through meticulous digital signal processing technology, it unravels the complex waveforms of heartbeat, breathing, and body movement, and then extracts key parameters including heart rate, respiratory rate, and motion signals. By summarizing and statistically analyzing this information, it can serve as a valuable tool for evaluating sleep quality.

[0061] Furthermore, first, the thin film piezoelectric sensor needs to be installed under the bed sheet to ensure that the tiny pressure changes caused by body movement, heartbeat and breathing can be captured. On this basis, the analog signal output by the sensor is sampled using a data acquisition system (such as ADC). Secondly, the collected signal is usually weak, so it needs to be amplified by an amplification circuit (such as an operational amplifier) ​​to improve the signal-to-noise ratio. Subsequently, the amplified analog signal is converted into a digital signal through an analog-to-digital converter (ADC) for subsequent processing. In the digital signal processing stage, a low-pass filter is first applied to remove high-frequency noise and interference, and retain the effective frequency components of the heartbeat, breathing and body movement signals. Then, signal features are extracted through time domain and frequency domain analysis techniques, including identifying periodic fluctuations in the signal to calculate the heart rate, analyzing the signal changes caused by breathing to extract the breathing frequency and depth, and detecting the body activity state by observing irregular fluctuations in the signal. Finally, the mean and standard deviation of the heart rate, breathing rate and other physiological parameters are calculated through statistical analysis to evaluate the health status.

[0062] Since this application is equipped with multiple sensors, the data of each sensor can be used in conjunction with each other and corroborate each other. This is to avoid distortion of the results obtained by relying on only one sensor. For example, multiple indicators of the elderly are identified through multiple sensors, and then the data from multiple sensors will be integrated to obtain the activity level and activity pattern of the elderly, and then correlation analysis will be performed. In particular, the elderly can work with their families and caregivers to access data and related physiological parameters through a web interface, allowing them to keep abreast of the situation and take quick action when necessary. This improves the elderly's ability to live independently and their overall quality of life, while effectively ensuring their safety.

[0063] Another embodiment of the present application provides a health monitoring system for elderly people living alone for implementing a health monitoring method for elderly people living alone, including a data acquisition module and a data analysis module, wherein the data acquisition module is used to collect data through sensors, and the data analysis module is used to identify and analyze based on the collected data.

[0064] The present application also conducted the following experiments to prove the feasibility of the method of the present application.

[0065] Step 1: Data Collection

[0066] Data were collected over 30 days from 10 elderly participants aged 65 years or older. The group consisted of six women and four men. Throughout the data collection period, the subjects were allowed to freely engage in daily activities without any specific restrictions, thus creating a natural and unobtrusive observation environment. Due to privacy considerations, this dataset will not be made public.

[0067] Step 2: Sensor Accuracy Evaluation

[0068] Develop and evaluate multiple sensor datasets to enable non-invasive monitoring of daily activities and health status of the elderly. The audio dataset captures sound events in the daily environment by strategically placing sensors and uses data augmentation techniques to address data imbalance, which is ultimately used to train and evaluate multiple deep learning models. Infrared sensor data is processed by a clustering-based position estimation algorithm to analyze the subject's activity patterns and identify behaviors such as naps, nighttime toileting, and sleep patterns. Thin-film piezoelectric sensors used in beds are used to capture key physiological indicators such as heart rate and respiratory rate, demonstrating their application value in real-world environments. Cadence monitoring verifies the accuracy of the sensor through laboratory testing, indicating that the subject maintains a good level of activity. The overall study demonstrates the potential and practicality of multimodal sensor systems in health monitoring of the elderly.

[0069] Step 2: Comprehensive data analysis

[0070] 1) Activity level analysis

[0071] By analyzing the trigger points of infrared sensors, the indoor activity level of the elderly is extracted, providing an indicator for evaluating the mobility of the elderly. The activity level is determined by calculating the transitions between different positions. Every time the sensor triggers a transition from one state to another, the activity level increases by 1. The higher the activity level, the healthier the elderly are.

[0072] 2) Activity pattern analysis

[0073] Through the coordinated application of multi-sensor data, accurate monitoring of the elderly's activities throughout the day is achieved. Using the comprehensive data of audio, bed, infrared and physiological function sensors, the system can effectively filter out recognition errors and improve the accuracy of activity detection. For example, through the collaboration of audio sensors and thin-film piezoelectric sensors on the bed, snoring can be more accurately identified, while the combination of audio sensors and infrared sensors optimizes the positioning and analysis of sound events. In addition, the study also demonstrated the potential of multimodal sensors in activity recognition, especially in real-time monitoring of elderly activities and preventing accidents such as falls. The overall results show that the combination of multiple sensor technologies and big data analysis can provide comprehensive support for the health management of the elderly.

[0074] 3) Correlation analysis

[0075] The study analyzed the correlation between multiple physiological and activity parameters of 10 elderly subjects, including average heart rate, respiratory rate, heart rate to respiratory rate ratio, sleep time, indoor activity level, cadence, and recreation time. The results showed that there was a strong correlation between average heart rate and heart rate to respiratory rate ratio, while the correlation between respiratory rate and heart rate to respiratory rate ratio was weak. Gender correlation analysis showed that males and females showed significant differences in activity level and sleep time.

[0076] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for monitoring the health of elderly people living alone, characterized in that: include: Collecting data through sensors, Based on the collected data, the following recognition is performed: Activity recognition based on audio data.

2. The health monitoring method for elderly people living alone according to claim 1, characterized in that: The activity recognition based on audio data is specifically implemented by applying the following modules: an audio event segmentation module, a feature extraction module and a classifier module.

3. The health monitoring method for elderly people living alone according to claim 2, characterized in that: The audio event segmentation module is used to achieve the following objectives: segmenting audio data into specific behavior events, The feature extraction module is used to achieve the following objectives: generate a Log-Mel spectrogram based on audio data, The classifier module is used to achieve the following purpose: output classification results based on the Log-Mel spectrum.

4. The health monitoring method for elderly people living alone according to claim 3, characterized in that: Specifically, the model used by the classifier module includes four blocks, each block contains a convolution layer and a pooling layer. After 4 blocks, a 512-dimensional embedding vector is extracted, followed by average pooling and reshaping operations, and then a new 512-dimensional embedding vector is obtained. Then the new 512-dimensional embedding vector is input into the attention mechanism module. The attention mechanism module consists of a multi-head self-attention mechanism and a feedforward neural network with residual connections and layer normalization. The attention mechanism module is followed by a flattening processing layer, a fully connected layer, and a softmax classification layer.

5. The health monitoring method for elderly people living alone according to claim 1, characterized in that: The sensors are of various types, and the results are obtained by comprehensive analysis of data obtained by the various sensors.

6. The health monitoring method for elderly people living alone according to claim 1, characterized in that: It also includes location recognition based on infrared sensors, including: Get the infrared sensor data, Eliminate outliers due to infrared sensors being falsely triggered by sunlight, Fill in the missing data, The data obtained by the infrared sensor is processed using a clustering algorithm to identify the location of elderly people living alone.

7. The health monitoring method for elderly people living alone according to claim 1, characterized in that: It also includes gesture recognition based on physiological function sensors, including: Get data from physiological function sensors, Recognize the posture of elderly people living alone based on data from physiological function sensors.

8. The health monitoring method for elderly people living alone according to claim 1, characterized in that: It also includes gait recognition based on physiological function sensors, including: Get data from physiological function sensors, Identify the gait of elderly people living alone based on data from physiological function sensors.

9. The health monitoring method for elderly people living alone according to claim 1, characterized in that: It also includes the recognition of heart rate and breathing rate based on thin film piezoelectric sensors, including: Get data from the thin-film piezoelectric sensor mounted on the bed, The heart rate and breathing rate of elderly people living alone are identified based on data from thin film piezoelectric sensors.

10. A health monitoring system for elderly people living alone for implementing the health monitoring method for elderly people living alone as described in claims 1 to 9, characterized in that: It includes a data acquisition module and a data analysis module. The data acquisition module is used to collect data through sensors, and the data analysis module is used to identify and analyze based on the collected data.

Images