Remote old people monitoring method based on artificial intelligence

Abstract

The invention provides a remote elderly monitoring method and system based on artificial intelligence, and relates to the technical field of elderly behavior recognition, and the method comprises the steps: firstly obtaining a current image and a physiological monitoring signal of an elderly in a target area; preprocessing the current image and the physiological data to obtain a preprocessed image and the physiological data; performing multi-modal fusion on the preprocessed image and the physiological data based on a fusion algorithm to form fusion data; inputting the fusion data into a trained old people behavior model to obtain a monitoring result; the model is constructed through a walking elderly sample set and a wheelchair elderly sample set. And giving an alarm according to a monitoring result. According to the method, the multi-modal information of the image and the physiological signal is combined, the visual information and the physiological state are comprehensively considered, and the monitoring accuracy and reliability are greatly improved. The complementarity of different source information is effectively utilized by multi-modal fusion, and the behavior recognition capability in a complex scene is enhanced. By considering individual difference information, false alarm and missing alarm can be reduced, and the monitoring effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of elderly behavior recognition, and in particular to a remote elderly monitoring method based on artificial intelligence. Background Art

[0002] As the global aging process accelerates, the number of elderly people continues to increase, and related care and safety issues are becoming increasingly prominent. The elderly often face problems such as decreased physical health and ability to take care of themselves, and are in urgent need of effective monitoring and care solutions.

[0003] In recent years, surveillance technology has developed rapidly, especially the progress of video surveillance, sensor technology and data analysis technology, which makes real-time monitoring and remote management possible. The application of smart cameras, motion sensors and environmental monitoring equipment in homes and public places is becoming more and more popular.

[0004] Breakthroughs in artificial intelligence (AI) technology in areas such as image recognition, behavior analysis, and data mining have provided strong technical support for elderly monitoring. Through deep learning and machine learning algorithms, the system can automatically identify and analyze the behavior patterns of the elderly and detect potential dangers in a timely manner.

[0005] The invention patent with application number 202310092656.X discloses a behavior safety monitoring system for the elderly, including: management module, monitoring module, central control module, display module, alarm module and terminal module; the monitoring module is used to monitor the behavior of the elderly and transmit the monitoring information to the central control module; the central control module is used to analyze the monitoring information and obtain the analysis result; when the analysis result is dangerous, the on-site video screen is projected and displayed through the display module; when the analysis result is dangerous, the alarm module issues an alarm; when the analysis result is dangerous, the terminal module is used to receive the warning remotely. This application can actively perform intelligent analysis of video information, automatically identify dangerous behaviors before they occur, and complete the warning; at the same time, it can perform 7×24 all-weather automatic intelligent analysis of video, without the need for on-duty personnel to monitor the monitor, and only complete the warning processing through sound alarms and voice prompts when a warning occurs.

[0006] However, changes in the monitoring environment, such as lighting, background noise, weather, etc., will affect the performance of the camera and the accuracy of the sensor. For example, in low-light conditions, the camera may not be able to clearly capture the movements of the elderly, resulting in misjudgment. In addition, the physical conditions and behavioral habits of different elderly people are different. If the algorithm fails to fully consider individual differences, the normal behavior of some elderly people may be misjudged as dangerous behavior. In addition, some people may use different methods to complete the same action. For example, the way people using walkers and wheelchairs move is very different from their behavior when walking. Summary of the invention

[0007] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a remote elderly monitoring method based on artificial intelligence to improve the monitoring effect.

[0008] To achieve the above object, the present invention provides the following solutions:

[0009] A remote elderly monitoring method based on artificial intelligence, comprising:

[0010] Acquire current images and physiological monitoring signals of the elderly in the target area;

[0011] Preprocessing the current image and the physiological monitoring signal respectively to obtain a preprocessed image and preprocessed physiological data;

[0012] Based on a fusion algorithm, multi-modal fusion is performed on the pre-processed image and the pre-processed physiological data to obtain fused data;

[0013] The fusion data is input into a trained elderly behavior model to obtain monitoring results; the elderly behavior model is constructed by using a walking elderly sample set and a wheelchair elderly sample set;

[0014] An alarm is issued according to the monitoring result.

[0015] Preferably, the preprocessing process of the current image includes:

[0016] Taking each pixel of the current image as the center, a square window is taken, and detecting the change amount of each pixel under the square window;

[0017] When the change amount of the pixel point is greater than a preset threshold, the current image in the corresponding square window is smoothed using the median value of the pixel points in the square window to obtain smoothed pixel points;

[0018] Continuously moving the square window until the smoothing process of the entire current image is completed to obtain a smoothed image;

[0019] constructing a pixel enhancement function based on the convolution value of the smoothed image;

[0020] The pixel enhancement function is used to perform image enhancement processing on the current image to obtain the preprocessed image.

[0021] Preferably, the calculation formula for the change amount of each pixel point under the square window is:

[0022]

[0023] Among them, f x(x i ,y j ) represents the gradient value of the current image in the x direction, f y (x i ,y j ) represents the gradient value of the current image in the y direction, f(x i ,y j ) indicates that the current image is in (x i ,y j ) position, r represents the gray value of the pixel f(x i ,y j )’s change.

[0024] Preferably, the output formula of the smoothed pixel point is:

[0025] I denoised (x,y)=median(I original (xh,yh),…,I original (x+h,y+h))

[0026] In the formula, I denoised (x, y) is the smoothed pixel point, h is the side length of the square window, I original (x,y) is the value of the pixel at the (x,y) position on the current image.

[0027] Preferably, constructing a pixel enhancement function based on the convolution value of the smoothed image comprises:

[0028] Performing convolution processing on the smoothed image to obtain a convolved image;

[0029] Determine an enhancement coefficient based on the pixel value of the convolved image;

[0030] A pixel enhancement function is constructed based on the enhancement coefficient; wherein the pixel enhancement function is: Wherein, S(x,y) represents the preprocessed image, E(x,y) represents the enhancement coefficient, G(x,y) represents the pixel value of the convolved image at (x,y), I(x,y) represents the smoothed image, and σ represents the mean square error between the smoothed image and the convolved image.

[0031] Preferably, the preprocessing process of the physiological monitoring signal includes:

[0032] Shifting the physiological monitoring signal to obtain a shifted signal;

[0033] Performing wavelet decomposition on the shifted signal to obtain a plurality of wavelet coefficients;

[0034] Determine the filtering threshold according to the decomposition scale and length of the signal;

[0035] Constructing a denoising function according to the filtering threshold;

[0036] The noise of the translated signal is removed by using the denoising function to obtain the preprocessed physiological data.

[0037] Preferably, based on a fusion algorithm, multimodal fusion is performed on the preprocessed image and the preprocessed physiological data to obtain fused data, including:

[0038] Extracting image features from the preprocessed image to obtain a spatial feature vector; the spatial feature vector includes: target posture, position and relative motion information;

[0039] Extracting signal features from the preprocessed physiological data to obtain a physiological feature vector; the physiological feature vector includes: average heart rate, average blood pressure and coefficient of variation;

[0040] Performing spatial alignment and data standardization on the spatial feature vector and the physiological feature vector according to a preset time window to obtain a standardized feature vector;

[0041] The standardized feature vectors are combined based on feature fusion technology to obtain a final feature vector.

[0042] Preferably, the calculation formula of the final feature vector is:

[0043] F=α·S p +β·S l +γ·S m ;

[0044] Among them, F is the final feature vector, S p is the spatial eigenvector, S l is the physiological feature vector, S m is the time feature vector, α, β and γ are the importance weights of the preset spatial feature vector, physiological feature vector and time vector respectively.

[0045] Preferably, the method for constructing the elderly behavior model includes:

[0046] Obtaining a preset sample set of walking elderly people and a sample set of wheelchair elderly people; the sample set of walking elderly people and the sample set of wheelchair elderly people include a plurality of final feature vectors of samples;

[0047] Build a CNN model for multi-target detection;

[0048] The CNN model is trained according to the walking elderly sample set and the wheelchair elderly sample set to obtain a trained first classifier and a trained second classifier;

[0049] Using the first classifier to identify the safety status in the walking elderly sample set to obtain a first confidence set, and using the second classifier to identify the safety status in the wheelchair elderly sample set to obtain a second confidence set;

[0050] Summing the confidence levels of each safety state in the first confidence set and the second confidence set to obtain the sum of safety state confidences of the first confidence set and the second confidence set respectively;

[0051] Selecting the data with the largest sum of confidences of the safe state as the training threshold data, and retraining the classifiers corresponding to the sum of confidences of the other objects, so that the classification confidence of each classifier is greater than or equal to the training threshold data, and finally cascading the trained classifiers to obtain a classification network;

[0052] The trained LSTM neural network is connected to the classification network to obtain the elderly behavior model.

[0053] A remote elderly monitoring system based on artificial intelligence, comprising:

[0054] A data acquisition module, used to acquire current images and physiological monitoring signals of the elderly in the target area;

[0055] A preprocessing module, used to preprocess the current image and the physiological monitoring signal respectively to obtain a preprocessed image and preprocessed physiological data;

[0056] A fusion module, used for performing multimodal fusion on the preprocessed image and the preprocessed physiological data based on a fusion algorithm to obtain fused data;

[0057] A monitoring module, used for inputting the fusion data into a trained elderly behavior model to obtain monitoring results; the elderly behavior model is constructed by using a walking elderly sample set and a wheelchair elderly sample set;

[0058] An alarm module is used to generate an alarm according to the monitoring result.

[0059] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0060] The present invention provides a remote elderly monitoring method and system based on artificial intelligence, the method comprising: obtaining the current image and physiological monitoring signal of the elderly in the target area; preprocessing the current image and the physiological monitoring signal respectively to obtain the preprocessed image and preprocessed physiological data; based on the fusion algorithm, multimodally fusing the preprocessed image and the preprocessed physiological data to obtain fused data; inputting the fused data into a trained elderly behavior model to obtain a monitoring result; the elderly behavior model is constructed by a walking elderly sample set and a wheelchair elderly sample set; and alarming according to the monitoring result. The present invention combines the multimodal information of image data and physiological monitoring signals, so that the system can comprehensively consider visual information and physiological state, greatly improving the accuracy and reliability of monitoring. Multimodal fusion can effectively utilize the complementarity of information from different sources, which is helpful to improve the ability to recognize behaviors in complex scenes. And when constructing the model, it includes a walking elderly sample set and a wheelchair elderly sample set, which can adapt to the behavior patterns of different elderly people. Considering the information of individual differences can effectively reduce false alarms and missed alarms, and improve the monitoring effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0062] Figure 1 A flow chart of a method provided by an embodiment of the present invention;

[0063] Figure 2 A schematic diagram of the system structure provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] The purpose of the present invention is to provide a remote elderly monitoring method and system based on artificial intelligence, which can improve the monitoring effect.

[0066] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0067] Figure 1 A flow chart of a method provided by an embodiment of the present invention, such as Figure 1 As shown, the present invention provides a remote elderly monitoring method based on artificial intelligence, comprising:

[0068] Step 100: Acquire current images and physiological monitoring signals of the elderly in the target area;

[0069] Step 200: Preprocessing the current image and the physiological monitoring signal respectively to obtain a preprocessed image and preprocessed physiological data;

[0070] Step 300: Based on a fusion algorithm, multimodal fusion is performed on the preprocessed image and the preprocessed physiological data to obtain fused data;

[0071] Step 400: input the fusion data into a trained elderly behavior model to obtain monitoring results; the elderly behavior model is constructed by using a walking elderly sample set and a wheelchair elderly sample set;

[0072] Step 500: Issue an alarm based on the monitoring result.

[0073] Specifically, the device selection and arrangement in step 100 of this embodiment is as follows:

[0074] 1. Arrange image acquisition equipment:

[0075] Install high-definition cameras in target areas where the elderly are located (such as homes, nursing institutions, etc.) to ensure coverage of the elderly's activities. These cameras can be fixed or pan-tilt cameras for better viewing angle adjustment.

[0076] Using a depth camera (such as Kinect or Intel RealSense) can obtain three-dimensional spatial data to help identify the posture and movement status of the elderly.

[0077] 2. Arrange physiological monitoring equipment:

[0078] Smart watches or health monitoring instruments (such as heart rate monitors and blood pressure monitors) are worn on the elderly to record physiological data in real time. These devices usually have Bluetooth or Wi-Fi functions, which can transmit data to the monitoring system in real time.

[0079] Installing fixed physiological monitoring sensors at home, such as heart rate sensors installed in the room or on the bed, can also continuously monitor the physiological signals of the elderly.

[0080] Optionally, the image data acquisition method of this embodiment includes:

[0081] The camera acquires a real-time video stream, extracts static images periodically (such as capturing a picture every few seconds), or performs image acquisition based on motion triggers.

[0082] Frame capture using computer vision ensures images are acquired only when the elderly are present, optimizing storage and processing resources.

[0083] Furthermore, the physiological signal collection method of this embodiment includes:

[0084] Physiological data (such as heart rate, blood pressure, pulse, etc.) are obtained through wearable devices, which upload the data to the monitoring system at fixed time intervals.

[0085] Read physiological data from fixed sensor devices in real time and transmit the data to the data processing center via the cloud or local server.

[0086] Furthermore, the data transmission method of this embodiment includes:

[0087] Network connectivity: Ensure that installed cameras and wearable devices transmit data via a stable network connection (such as Wi-Fi, Bluetooth, or other communication methods).

[0088] The data format uses standardized data protocols (such as HTTP, MQTT, etc.) to ensure that data can be transmitted to the central processing unit effectively and stably.

[0089] Exemplarily, this embodiment also performs a data synchronization step to add a timestamp to the acquired images and physiological signals to achieve subsequent data alignment and analysis. Ensure that the two types of data correspond to each other in the same time window so that they can be correctly matched during multimodal fusion. In addition, this embodiment stores the acquired data in a local cache (such as an edge computing device) first to prevent data loss when the network is unstable. The preprocessed data can be transferred to a cloud server or a local database for subsequent analysis and processing.

[0090] Through the above steps, the monitoring system can efficiently obtain the current images and physiological monitoring signals of the elderly, and ensure real-time and accuracy, providing basic data support for subsequent behavior detection and early warning analysis.

[0091] Preferably, the preprocessing process of the current image includes:

[0092] Taking each pixel of the current image as the center, a square window is taken, and detecting the change amount of each pixel under the square window;

[0093] When the change amount of the pixel point is greater than a preset threshold, the current image in the corresponding square window is smoothed using the median value of the pixel points in the square window to obtain smoothed pixel points;

[0094] Continuously moving the square window until the smoothing process of the entire current image is completed to obtain a smoothed image;

[0095] constructing a pixel enhancement function based on the convolution value of the smoothed image;

[0096] The pixel enhancement function is used to perform image enhancement processing on the current image to obtain the preprocessed image.

[0097] Preferably, the calculation formula for the change amount of each pixel point under the square window is:

[0098]

[0099] Among them, f x (x i ,y j ) represents the gradient value of the current image in the x direction, f y (x i ,y j ) represents the gradient value of the current image in the y direction, f(x i ,y j ) indicates that the current image is in (x i ,y j ) position, r represents the gray value of the pixel f(x i ,y j )’s change.

[0100] Preferably, the output formula of the smoothed pixel point is:

[0101] I denoised (x,y)=median(I original (xh,yh),…,I original (x+h,y+h))

[0102] In the formula, I denoised (x, y) is the smoothed pixel point, h is the side length of the square window, I original (x,y) is the value of the pixel at the (x,y) position on the current image.

[0103] Specifically, in practical applications, since noise is unavoidable during the image acquisition process, and the noise is generally significantly different from the surrounding pixels, this embodiment can detect the change in each pixel under the square window to find the noise on the infrared image, and use the median of the pixel sequence under the square window as the smoothed pixel point, which can better maintain the image detail information while removing the noise.

[0104] Preferably, constructing a pixel enhancement function based on the convolution value of the smoothed image comprises:

[0105] Performing convolution processing on the smoothed image to obtain a convolved image;

[0106] Determine an enhancement coefficient based on the pixel value of the convolved image;

[0107] A pixel enhancement function is constructed based on the enhancement coefficient; wherein the pixel enhancement function is: Wherein, S(x,y) represents the preprocessed image, E(x,y) represents the enhancement coefficient, G(x,y) represents the pixel value of the convolved image at (x,y), I(x,y) represents the smoothed image, and σ represents the mean square error between the smoothed image and the convolved image.

[0108] Specifically, this embodiment can quantify the degree of difference between the image and the image after convolution by calculating the mean square error between the two. When the mean square error is large, it means that the original image and the image after convolution are very different and need to be enhanced; when the mean square error is small, it means that the original image and the image after convolution are small and no enhancement can be considered. In image enhancement, whether the image needs to be enhanced can be determined based on the size of the mean square error, so that the image can achieve a better visual effect, thereby obtaining more accurate visual information.

[0109] Preferably, the preprocessing process of the physiological monitoring signal includes:

[0110] Shifting the physiological monitoring signal to obtain a shifted signal;

[0111] Performing wavelet decomposition on the shifted signal to obtain a plurality of wavelet coefficients;

[0112] Determine the filtering threshold according to the decomposition scale and length of the signal;

[0113] Constructing a denoising function according to the filtering threshold;

[0114] The noise of the translated signal is removed by using the denoising function to obtain the preprocessed physiological data.

[0115] Preferably, based on a fusion algorithm, multimodal fusion is performed on the preprocessed image and the preprocessed physiological data to obtain fused data, including:

[0116] Extracting image features from the preprocessed image to obtain a spatial feature vector; the spatial feature vector includes: target posture, position and relative motion information;

[0117] Extracting signal features from the preprocessed physiological data to obtain a physiological feature vector; the physiological feature vector includes: average heart rate, average blood pressure and coefficient of variation;

[0118] Performing spatial alignment and data standardization on the spatial feature vector and the physiological feature vector according to a preset time window to obtain a standardized feature vector;

[0119] The standardized feature vectors are combined based on feature fusion technology to obtain a final feature vector.

[0120] Furthermore, this embodiment performs image preprocessing again, performs clarity enhancement, denoising, illumination correction and target detection (such as identifying and marking the position of the elderly) on the acquired current image, so that the image is suitable for further analysis; and performs denoising and normalization on the collected physiological monitoring signals (such as heart rate, blood pressure, etc.) to eliminate outliers in order to standardize data from different sources.

[0121] Furthermore, the present embodiment can use a deep learning model (such as a convolutional neural network, CNN) to extract spatial features from the preprocessed image, including the target's posture, position, and relative motion information. And extract the features of physiological data, such as parameters such as average heart rate and coefficient of variation, through signal processing techniques (such as Fourier transform, spectrum analysis, etc.). Exemplarily, in physiological signal analysis, the coefficient of variation can be used to measure the stability of the physiological state of the elderly. For example, the coefficient of variation of heart rate can reveal the physiological fluctuations of the elderly in daily activities. A lower CV value may indicate a more stable heart rate, whereas a high CV value may indicate an unstable physical state or potential health problems. When monitoring the health status of the elderly, the coefficient of variation can provide additional information to help caregivers judge the health changes of the elderly and the physiological feedback of their daily activities. For example, by analyzing the coefficient of variation of the heart rate data of the elderly over a period of time, their emotional fluctuations, exercise status, or potential health crises can be discovered, providing a more personalized care plan.

[0122] Specifically, this embodiment aligns the image and physiological data in time at the same time to deal with potential time differences from different sources. For example, time windows or synchronization markers are used to enable features from different data sources to be compared in the same period. Data normalization techniques (such as Z-Score normalization or Min-Max normalization) are further used to unify the scale of the image and physiological features to eliminate the impact of different scales on the fusion results.

[0123] Optionally, this embodiment combines the standardized image features and physiological signal features into a feature vector, and uses feature-level fusion technology (such as splicing, weighted summation) to merge these features. In addition, this embodiment can also perform decision-level fusion: combining independent models to make preliminary judgments on image and physiological data (such as triggering an alarm only when both are abnormal, and fusion using methods such as voting or Bayesian inference).

[0124] Through this multimodal fusion, the present embodiment can fully utilize the complementary characteristics of visual information and physiological information to improve the level of accurate recognition and monitoring of the behavioral status of the elderly.

[0125] Preferably, the calculation formula of the final feature vector is:

[0126] F=α·S p +β·S l +γ·S m ;

[0127] Among them, F is the final feature vector, S p is the spatial eigenvector, S l is the physiological feature vector, S m is the time feature vector, α, β and γ are the importance weights of the preset spatial feature vector, physiological feature vector and time vector respectively.

[0128] Specifically, this embodiment can flexibly adjust the influence of different features in the fusion result by adjusting the values ​​of α, β, and γ. For example, in some cases, physiological features may better reflect the state of the elderly than spatial features, and the weight of β can be increased. The weighting coefficient (importance weight) can be automatically tuned by machine learning algorithms (such as random search and Bayesian optimization) to ensure the best performance in specific application scenarios.

[0129] Optionally, the time feature of this embodiment refers to dynamic information or change trends related to time, which are usually used to analyze the change pattern of behavior. It includes:

[0130] Time series data, such as the trend of physiological signals (heart rate, blood pressure, etc.) changing over time.

[0131] Behavior duration: the duration of a certain behavior (such as walking, sitting, etc.) of the elderly.

[0132] Behavior frequency refers to the number of times an elderly person's behavior occurs within a certain period of time.

[0133] Exemplarily, the time features of this embodiment can be obtained through the time series data of spatial features and physiological features. In the process of optimizing the weighted coefficients, the weight of the time features can help the model better understand the dynamic behavior of the changes. By adjusting the relative weights of the time features and the spatial features, this embodiment can more accurately identify the behavioral state of the elderly, especially in a rapidly changing environment.

[0134] In addition, this embodiment can also use historical monitoring data to evaluate the model performance of different configurations through cross-validation, and determine the best weighted combination on this basis.

[0135] Furthermore, the application scenarios of the weighting coefficient adjustment of this embodiment are as follows:

[0136] Application scenario 1: General monitoring (such as daily life):

[0137] α=0.4

[0138] β=0.5

[0139] γ=0.1

[0140] Note: In this scenario, physiological characteristics are more important to help monitor the health status of the elderly, while spatial characteristics assist in analyzing the direction of daily activities.

[0141] Application scenario 2: fall detection:

[0142] α=0.6

[0143] β=0.3

[0144] γ=0.1

[0145] Note: In fall monitoring, spatial features are particularly important because the posture and position changes of the elderly when they fall will significantly affect the monitoring results.

[0146] Application scenario 3: Emergency response (such as heart attack):

[0147] α=0.2

[0148] β=0.7

[0149] γ=0.1

[0150] Note: In medical emergencies such as heart disease, physiological characteristics become more critical because timely acquisition of physiological signals (such as heart rate changes) can help make medical decisions quickly.

[0151] Application scenario 4: Emotional and psychological state monitoring:

[0152] α=0.3

[0153] β=0.4

[0154] γ=0.3

[0155] Note: When monitoring the emotional changes of the elderly, temporal characteristics (such as activity frequency) are equally important and can effectively analyze the changes in the emotional and psychological states of the elderly in different time periods.

[0156] Furthermore, by reasonably setting and adjusting the weighting coefficients, the feature combination can be optimized for various specific application scenarios to achieve personalized, accurate and effective monitoring of the elderly. This method not only improves the flexibility and accuracy of monitoring, but also provides greater possibilities for future smart health management.

[0157] Preferably, the method for constructing the elderly behavior model includes:

[0158] Obtaining a preset sample set of walking elderly people and a sample set of wheelchair elderly people; the sample set of walking elderly people and the sample set of wheelchair elderly people include a plurality of final feature vectors of samples;

[0159] Build a CNN model for multi-target detection;

[0160] The CNN model is trained according to the walking elderly sample set and the wheelchair elderly sample set to obtain a trained first classifier and a trained second classifier;

[0161] Using the first classifier to identify the safety status in the walking elderly sample set to obtain a first confidence set, and using the second classifier to identify the safety status in the wheelchair elderly sample set to obtain a second confidence set;

[0162] Summing the confidence levels of each safety state in the first confidence set and the second confidence set to obtain the sum of safety state confidences of the first confidence set and the second confidence set respectively;

[0163] Selecting the data with the largest sum of confidences of the safe state as the training threshold data, and retraining the classifiers corresponding to the sum of confidences of the other objects, so that the classification confidence of each classifier is greater than or equal to the training threshold data, and finally cascading the trained classifiers to obtain a classification network;

[0164] The trained LSTM neural network is connected to the classification network to obtain the elderly behavior model.

[0165] Furthermore, the model building process of this embodiment includes:

[0166] 1. Data collection and preprocessing: collect sample sets of walking elderly people and wheelchair elderly people. Each sample contains a feature vector generated after feature extraction. Preprocess the data, including normalization, missing value processing and denoising, to ensure the data quality of the input model.

[0167] 2. Construct a CNN model for multi-target detection. Use the convolutional neural network (CNN) structure to build a model for capturing spatial features to adapt to the diversity of features of walking and wheelchair-bound elderly people. Design the hierarchical structure of the model, including convolutional layers, pooling layers, and fully connected layers, to extract higher-level features.

[0168] 3. Model training: Use the sample set of walking elderly people as training data for the first classifier, and the sample set of wheelchair elderly people as training data for the second classifier. Use the cross entropy loss function and optimization algorithm (such as Adam or SGD) to train the CNN model so that it can effectively identify the walking state and wheelchair state.

[0169] 4. Generation of confidence sets: Use the trained classifiers to identify the safety states of walking and wheelchair-bound elderly people, and generate the first confidence set (walking elderly people) and the second confidence set (wheelchair-bound elderly people). The confidence is the probability output of each safety state predicted by the model, reflecting the degree of confidence in each state.

[0170] 5. Calculate the total confidence of the safety status. Sum the confidence of each state in the first confidence set and the second confidence set to obtain the total value to determine the current safety status.

[0171] 6. Selection and adjustment of training threshold data: select the value with the largest total confidence as the training threshold data and optimize the model. Retrain the data according to the needs of the classifier to ensure that the confidence of the classifier output is greater than or equal to the threshold, so as to improve the accuracy and robustness of the classification.

[0172] 7. Cascade classification network construction, cascading multiple classifiers to form a comprehensive classification network. This network can simultaneously identify the status of walking and wheelchair-bound elderly people, producing more comprehensive prediction results.

[0173] 8. Connect the LSTM neural network. After the classification network, connect the long short-term memory network (LSTM), which is designed to process time series features, capture the temporal changes of behavioral states, and provide dynamic prediction capabilities for the elderly behavior model. LSTM can identify the persistence and transformation of behavioral patterns and help predict future behavioral states.

[0174] In this embodiment, a method combining CNN and LSTM in deep learning is used to extract spatial and temporal information and enhance the accuracy of identifying the behavior of the elderly. The classifier can be dynamically adjusted according to the confidence of the feedback to improve the flexibility and adaptability of the model. By adjusting the training threshold, the model can optimize its performance under different environmental conditions and user states. The introduction of LSTM improves the ability to model the temporal dynamics of behavioral patterns, allowing the model to analyze the changing trends of the elderly's behavior and support long-term monitoring and analysis. Through data analysis and model building, decision support can be provided to caregivers, the quality and efficiency of elderly care can be improved, and the sense of security of the elderly's life can be improved. The model framework can be easily expanded to adapt to other types of behavior monitoring tasks and has good industry applicability.

[0175] In summary, this embodiment provides a safe and efficient monitoring and management solution for the elderly through deep learning and multimodal data fusion, and makes full use of advanced machine learning algorithms to improve monitoring and early warning capabilities.

[0176] Optionally, the monitoring result of this embodiment includes the following contents:

[0177] 1. Behavior recognition results: The recognized behavior states of the elderly, such as walking, sitting, falling, using a wheelchair, etc. For different behavior states, there may be different confidence values, indicating the model's confidence in identifying the state.

[0178] 2. Safety status assessment: The first classifier and the second classifier are used to assess whether the elderly are in a safe state. If the recognition result indicates that the elderly are at potential risk (such as falling or staying abnormally), an alarm needs to be triggered.

[0179] 3. Physiological indicator data: monitor physiological signals, such as heart rate, blood pressure, etc. If the physiological data is found to be out of the normal range (such as heart rate is too fast or too slow), an alarm is also required.

[0180] 4. Time information: The timestamp of the behavior and the comparison with previous behaviors. If a behavior lasts too long (such as exceeding normal activity time), it may indicate a risk.

[0181] Furthermore, the alarm mechanism of this embodiment is as follows:

[0182] Set alarm conditions and set thresholds based on monitoring results. For example, if the confidence level of a fall is greater than a certain set value (such as 0.8), or if a physiological parameter exceeds the normal range (such as a heart rate higher than 120 bpm), an alarm is triggered.

[0183] Alarm triggering: once the monitoring results meet the alarm conditions, the system will automatically trigger the alarm mechanism. The alarm methods can be diversified, including:

[0184] Sound alarm: The device or mobile phone will sound an alarm to alert people nearby.

[0185] SMS or app notifications: Instantly send SMS or push notifications to caregivers or family members to inform them to handle the current abnormal situation.

[0186] Autonomous calling: The system can dial preset emergency contacts, such as family members or nursing agencies, via telephone.

[0187] The alarm information includes: the identity information of the elderly, the current behavior status of the identification, the confidence level and safety assessment (such as whether they have fallen), relevant physiological data (such as heart rate, blood pressure, etc.), time and location of occurrence (such as monitoring area number or room name).

[0188] Furthermore, the response mechanism of this embodiment is: after the alarm is triggered, the system should design a response mechanism to record the alarm event and guide the caregiver to take appropriate actions. For example, the system can provide a one-click call to emergency medical services.

[0189] In summary, the entire alarm process of this embodiment is real-time, efficient and flexible, ensuring a quick response when the elderly are in danger, which can effectively improve the life safety of the elderly, reduce the pressure on caregivers, and enhance the security of families and society.

[0190] like Figure 2 As shown, corresponding to the above method, this embodiment also provides a remote elderly monitoring system based on artificial intelligence, including:

[0191] A data acquisition module, used to acquire current images and physiological monitoring signals of the elderly in the target area;

[0192] A preprocessing module, used to preprocess the current image and the physiological monitoring signal respectively to obtain a preprocessed image and preprocessed physiological data;

[0193] A fusion module, used for performing multimodal fusion on the preprocessed image and the preprocessed physiological data based on a fusion algorithm to obtain fused data;

[0194] A monitoring module, used for inputting the fusion data into a trained elderly behavior model to obtain monitoring results; the elderly behavior model is constructed by using a walking elderly sample set and a wheelchair elderly sample set;

[0195] An alarm module is used to generate an alarm according to the monitoring result.

[0196] The beneficial effects of the present invention are as follows:

[0197] (1) The present invention combines multimodal information of image data and physiological monitoring signals, enabling the system to comprehensively consider visual information (such as the posture and movement of the elderly) and physiological status (such as heart rate and blood pressure), greatly improving the accuracy and reliability of monitoring. Multimodal fusion can effectively utilize the complementarity of information from different sources and improve the ability to recognize behaviors in complex scenarios.

[0198] (2) The present invention can obtain images and physiological data of the elderly in real time, providing support for timely identification of potential dangers. Through rapid processing and analysis, an alarm can be immediately issued when abnormal behavior occurs, ensuring an immediate response and promoting the safety of the elderly.

[0199] (3) The present invention performs intelligent preprocessing on the acquired data, reducing noise and uncertainty and improving data quality. The automated processing method of the present invention helps the system better adapt to environmental changes, such as changes in lighting or interference conditions, and avoids the need for manual supervision.

[0200] (4) The elderly behavior model of the present invention is constructed based on a machine learning algorithm and has the ability to self-learn and adapt. Through continuous training and data accumulation, the model can improve its recognition accuracy over time and gradually optimize its understanding and response to individual behaviors.

[0201] (5) The present invention includes a sample set of walking elderly people and a sample set of wheelchair elderly people when constructing the model, so that the system can adapt to the behavior patterns of different elderly people. Considering the information of individual differences can effectively reduce false positives and false negatives and improve the monitoring effect.

[0202] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0203] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A remote elderly monitoring method based on artificial intelligence, characterized in that: include: Acquire current images and physiological monitoring signals of the elderly in the target area; Preprocessing the current image and the physiological monitoring signal respectively to obtain a preprocessed image and preprocessed physiological data; Based on a fusion algorithm, multi-modal fusion is performed on the pre-processed image and the pre-processed physiological data to obtain fused data; The fusion data is input into a trained elderly behavior model to obtain monitoring results; the elderly behavior model is constructed by using a walking elderly sample set and a wheelchair elderly sample set; An alarm is issued according to the monitoring result.

2. The artificial intelligence-based remote elderly monitoring method according to claim 1 is characterized in that: The preprocessing process of the current image includes: Taking each pixel of the current image as the center, a square window is taken, and detecting the change amount of each pixel under the square window; When the change amount of the pixel point is greater than a preset threshold, the current image in the corresponding square window is smoothed using the median value of the pixel points in the square window to obtain smoothed pixel points; Continuously moving the square window until the smoothing process of the entire current image is completed to obtain a smoothed image; constructing a pixel enhancement function based on the convolution value of the smoothed image; The pixel enhancement function is used to perform image enhancement processing on the current image to obtain the preprocessed image.

3. The remote elderly monitoring method based on artificial intelligence according to claim 2 is characterized in that: The calculation formula for the change of each pixel under the square window is: Among them, f x (x i ,y j ) represents the gradient value of the current image in the x direction, f y (x i ,y j ) represents the gradient value of the current image in the y direction, f(x i ,y j ) indicates that the current image is in (x i ,y j ) position, r represents the gray value of the pixel f(x i ,y j )’s change.

4. The remote elderly monitoring method based on artificial intelligence according to claim 3 is characterized in that: The output formula of the smoothed pixel point is: I denoised (x,y)=median(I original (x-h,y-h),…,I original (x+h,y+h)) In the formula, I denoised (x, y) is the smoothed pixel point, h is the side length of the square window, I original (x,y) is the value of the pixel at the (x,y) position on the current image.

5. The artificial intelligence-based remote elderly monitoring method according to claim 4 is characterized in that: Constructing a pixel enhancement function based on the convolution value of the smoothed image, comprising: Performing convolution processing on the smoothed image to obtain a convolved image; Determine an enhancement coefficient based on the pixel value of the convolved image; A pixel enhancement function is constructed based on the enhancement coefficient; wherein the pixel enhancement function is: Wherein, S(x,y) represents the preprocessed image, E(x,y) represents the enhancement coefficient, G(x,y) represents the pixel value of the convolved image at (x,y), I(x,y) represents the smoothed image, and σ represents the mean square error between the smoothed image and the convolved image.

6. The remote elderly monitoring method based on artificial intelligence according to claim 1 is characterized in that: The preprocessing process of the physiological monitoring signal includes: Shifting the physiological monitoring signal to obtain a shifted signal; Performing wavelet decomposition on the shifted signal to obtain a plurality of wavelet coefficients; Determine the filtering threshold according to the decomposition scale and length of the signal; Constructing a denoising function according to the filtering threshold; The noise of the translated signal is removed by using the denoising function to obtain the preprocessed physiological data.

7. The remote elderly monitoring method based on artificial intelligence according to claim 1 is characterized in that: Based on a fusion algorithm, multimodal fusion is performed on the preprocessed image and the preprocessed physiological data to obtain fused data, including: Extracting image features from the preprocessed image to obtain a spatial feature vector; the spatial feature vector includes: target posture, position and relative motion information; Extracting signal features from the preprocessed physiological data to obtain a physiological feature vector; the physiological feature vector includes: average heart rate, average blood pressure and coefficient of variation; Performing spatial alignment and data standardization on the spatial feature vector and the physiological feature vector according to a preset time window to obtain a standardized feature vector; The standardized feature vectors are combined based on feature fusion technology to obtain a final feature vector.

8. The artificial intelligence-based remote elderly monitoring method according to claim 7 is characterized in that: The calculation formula of the final feature vector is: F=α·S p +β·S l +γ·S m ; Where F is the final feature vector, S p is the spatial eigenvector, S l is the physiological feature vector, S m is the time feature vector, α, β and γ are the importance weights of the preset spatial feature vector, physiological feature vector and time vector respectively.

9. The artificial intelligence-based remote elderly monitoring method according to claim 1 is characterized in that: The method for constructing the elderly behavior model includes: Obtaining a preset sample set of walking elderly people and a sample set of wheelchair elderly people; the sample set of walking elderly people and the sample set of wheelchair elderly people include a plurality of final feature vectors of samples; Build a CNN model for multi-target detection; The CNN model is trained according to the walking elderly sample set and the wheelchair elderly sample set to obtain a trained first classifier and a trained second classifier; Using the first classifier to identify the safety status in the walking elderly sample set to obtain a first confidence set, and using the second classifier to identify the safety status in the wheelchair elderly sample set to obtain a second confidence set; Summing the confidence levels of each safety state in the first confidence set and the second confidence set to obtain the sum of safety state confidences of the first confidence set and the second confidence set respectively; Selecting the data with the largest sum of confidences of the safe state as the training threshold data, and retraining the classifiers corresponding to the sum of confidences of the other objects, so that the classification confidence of each classifier is greater than or equal to the training threshold data, and finally cascading the trained classifiers to obtain a classification network; The trained LSTM neural network is connected to the classification network to obtain the elderly behavior model.

10. A remote elderly monitoring system based on artificial intelligence, characterized in that: include: A data acquisition module, used to acquire current images and physiological monitoring signals of the elderly in the target area; A preprocessing module, used to preprocess the current image and the physiological monitoring signal respectively to obtain a preprocessed image and preprocessed physiological data; A fusion module, used for performing multimodal fusion on the preprocessed image and the preprocessed physiological data based on a fusion algorithm to obtain fused data; A monitoring module, used for inputting the fusion data into a trained elderly behavior model to obtain monitoring results; the elderly behavior model is constructed by using a walking elderly sample set and a wheelchair elderly sample set; An alarm module is used to generate an alarm according to the monitoring result.

Images