Home-based care interaction scene system based on AIoT technology

Abstract

The invention discloses a home-based care interaction scene system based on an AIoT technology, and belongs to the technical field of intelligent home-based care. The problem that an existing home-based care system cannot carry out real-time prediction and intelligent response on demands according to behavior data of target personnel is solved. A data acquisition unit acquires bioelectricity information and videos of a target person in different life scenes and stores the bioelectricity information and videos into a database; the data processing unit adopts a computer vision technology, OpenPose and DeepSORT are combined, and action postures and facial features of a target person are extracted from a video in real time; the behavior prediction model predicts the behavior of the target person at the next moment by using the historical bioelectricity information, the action posture and the facial features of the target person in the database, and the data processing unit further predicts the behavior of the target person according to the behavior of the target person predicted by the behavior prediction model. And the smart home equipment control unit controls the corresponding smart home equipment according to the received control signal. The method is suitable for home-based care interaction scenes.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent home-based elderly care. Background Art

[0002] At present, home-based elderly care is the most common way for the elderly in China. Spending their old age in a familiar environment can maximize the sense of belonging of the elderly. However, the software and hardware support for home-based elderly care still cannot fully meet the elderly care needs of the elderly.

[0003] AIoT (Artificial Intelligence of Things) technology is the product of the integration of artificial intelligence and Internet of Things technology. Artificial intelligence helps the Internet of Things to intelligently process massive amounts of data and improve the human-computer interaction experience. The Internet of Things, through the interconnection of all things, provides a large number of analyzable data objects for artificial intelligence through sensors and terminal devices, making artificial intelligence decision-making more intelligent.

[0004] However, the existing smart home systems are not suitable for the needs of home-based elderly care and cannot predict the needs of target persons based on real-time behavior data. Summary of the Invention

[0005] The present invention is to solve the problem that the existing home-based elderly care system cannot make real-time predictions and intelligent responses to the needs based on the behavior data of target persons, and now provides a home-based elderly care interaction scenario system based on AIoT technology.

[0006] The home-based elderly care interaction scenario system based on AIoT technology of the present invention includes: a data acquisition unit, a wireless local area network, a data processing unit, a database, a smart home device control unit, and a behavior prediction model;

[0007] The data acquisition unit is used to collect bioelectrical information and videos of target persons in different living scenarios through video acquisition devices, human infrared sensors, and bioelectric sensors; and send the collected bioelectrical information and videos to the data processing unit and the database through the wireless local area network;

[0008] The data processing unit combines computer vision technology, OpenPose (open source pose estimation library), and DeepSORT (Deep Simple Online and Realtime Tracking) to extract the action postures and facial features of target persons from the video in real time; and transmits the extracted information to the database and the behavior prediction model;

[0009] The behavior prediction model uses the bioelectrical information, action postures, and facial features of the target person in the previous moment in the database to predict the behavior of the target person in the next moment;

[0010] The behavior prediction model is pre-trained through the historical behavior information of the target person in different life scenarios and the corresponding video and bioelectricity information;

[0011] The data processing unit also sends a control signal to the smart home device control unit according to the behavior of the target person predicted by the behavior prediction model, and the smart home device control unit controls the corresponding smart home device according to the received control signal.

[0012] Further, in the present invention, the data acquisition unit includes a camera, a panoramic infrared camera, a wind speed sensor, and a light sensor, and is used for the elderly to wear smart bracelets and / or health monitoring clothes.

[0013] Further, in the present invention, the bioelectric sensor is used to collect heart rate, heart beat, blood pressure and body temperature information.

[0014] Further, in the present invention, the behavior prediction model is implemented by using a GAN model.

[0015] Further, in the present invention, the real-time extraction of the action posture and facial features of the target person from the video is implemented by using a CNN network.

[0016] Further, in the present invention, the behavior prediction model is trained by using historical behavior data and historical scenario data extracted from video streams, infrared images, and bioelectric signals;

[0017] The historical behavior data includes: action data and expression data;

[0018] The action data includes: walking, running, jumping, waving, nodding, shaking the head, shaking hands, hugging, and waving actions in different scenarios;

[0019] The expression data includes: smiling, frowning, blinking, staring, pursing the lips, opening the mouth, closing the eyes, and biting the lips in different scenarios;

[0020] The scenario data includes: location data and interaction item data;

[0021] The location data includes: bedroom, kitchen, living room, bathroom, study, dining room, and balcony;

[0022] The interaction item data includes: sofas, TVs, bedding, cooking utensils, eating utensils, kitchen utensils, stationery, office supplies, cleaning utensils, fitness utensils, and entertainment tools.

[0023] Further, in the present invention, the behavior prediction model uses the backpropagation algorithm and an optimizer to minimize the loss function of the model parameters.

[0024] Further, in the present invention, the control of the corresponding smart home device by the smart home device control unit according to the received control signal includes intelligent lighting adjustment, intelligent glass color change adjustment, intelligent ceiling switch adjustment, air speed adjustment and temperature adjustment of the intelligent air conditioner, and alarm triggering of the emergency warning device.

[0025] The present invention collects the video and bioelectrical signals of the person to be detected in the interaction scene in real time, extracts the actions and scenes of the person through the video stream, and inputs the bioelectrical signals and the actions and scenes of the person into the prediction model to predict the next action of the corresponding person. Based on the analysis of historical behavior patterns and real-time data, this system can not only improve the accuracy of behavior prediction, but also actively regulate smart home devices based on the prediction results, realizing the linkage response of devices such as intelligent lighting, air conditioners, and glass, thereby optimizing the comfort and safety of the home environment. The present invention significantly improves the adaptability of smart homes to the daily behaviors and personalized needs of the elderly, enhances the intelligent and user-friendly experience of home care, and effectively improves the matching degree between smart homes and the needs of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a system block diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0028] DETAILED DESCRIPTION OF THE EMBODIMENT 1: Refer to Figure 1 Specifically describe this embodiment. The home care interaction scene system based on AIoT technology described in this embodiment includes: a data acquisition unit 1, a wireless local area network 2, a data processing unit 3, a database 4, a smart home device control unit 5, and a behavior prediction model 6;

[0029] The data acquisition unit 1 is used to collect bioelectrical information and video of the target person in different living scenes through video acquisition devices, human infrared sensors, and bioelectrical sensors; and send the collected bioelectrical information and video to the data processing unit 3 and the database 4 through the wireless local area network 2.

[0030] The data processing unit 3 combines computer vision technology, OpenPose, and DeepSORT to extract the action postures and facial features of the target person from the video in real time; and transmits the extracted information to the database 4 and the behavior prediction model 6;

[0031] The behavior prediction model 6 predicts the behavior of the target person at the next moment by using the bioelectric information, action postures, and facial features of the target person in the database 4 at the previous moment.

[0032] The behavior prediction model 6 is pre-trained through the historical behavior information of the target person in different living scenarios and the corresponding video and bioelectric information.

[0033] The data processing unit 3 also sends a control signal to the smart home device control unit 5 according to the behavior of the target person predicted by the behavior prediction model 6, and the smart home device control unit 5 controls the corresponding smart home device according to the received control signal.

[0034] In this embodiment, the database stores the actions, expressions, bioelectricity, and corresponding interaction item information of different target persons at different historical moments in different scenarios.

[0035] Further, in the present invention, the data acquisition unit 1 includes a camera, a panoramic infrared camera, an air speed sensor, and a light sensor, and is used for the elderly to wear a smart bracelet and / or a health monitoring garment.

[0036] Further, in the present invention, the bioelectric sensor is used to collect heart rate, heart beat, blood pressure, and body temperature information.

[0037] Further, in the present invention, the behavior prediction model 6 is implemented by using a GAN model.

[0038] Further, in the present invention, the real-time extraction of the action postures and facial features of the target person from the video is implemented by using a CNN network.

[0039] Further, in the present invention, the behavior prediction model 6 is trained by using historical behavior data and historical scenario data extracted from video streams, infrared images, and bioelectric signals;

[0040] The historical behavior data includes: action data and expression data;

[0041] The action data includes: walking, running, jumping, waving, nodding, shaking the head, shaking hands, hugging, and waving actions in different scenarios;

[0042] The expression data includes: smiling, frowning, blinking, staring, pouting, opening the mouth, closing the eyes, and biting the lip expressions in different scenarios;

[0043] The scenario data includes: location data and interactive item data;

[0044] The location data includes: bedroom, kitchen, living room, bathroom, study, dining room, and balcony;

[0045] The interactive item data includes: sofa, TV, bedding, cooking utensils, eating utensils, kitchen utensils, stationery, office supplies, cleaning utensils, fitness equipment, and entertainment tools.

[0046] Further, in the present invention, the behavior prediction model 6 uses the backpropagation algorithm and an optimizer to minimize the loss function of the model parameters.

[0047] Further, in the present invention, the control of the corresponding smart home device by the smart home device control unit 5 according to the received control signal includes intelligent lighting adjustment, intelligent glass color change adjustment, intelligent ceiling switch adjustment, air speed adjustment and temperature adjustment of the intelligent air conditioner, and alarm triggering of the emergency warning device.

[0048] A home care interaction scenario system based on artificial intelligence Internet of Things provided by the present invention collects daily behavior information of the elderly at home through sensors such as cameras, infrared sensors, and bioelectric sensors.

[0049] The camera uses a binocular battery-powered camera with a 165-degree field of view and 1TB of internal storage space, and can provide high-quality video data for behavior analysis and tracking.

[0050] The camera also uses a panoramic infrared camera that can work at night or in low-light conditions to obtain high-quality video image signals for identifying the daily activities and abnormal behaviors of the elderly.

[0051] Wearable devices such as smart bracelets or health monitoring vests are introduced to continuously monitor the physiological parameters and activity status of the elderly.

[0052] A wireless network is used for intelligent service response to the daily life behaviors of the elderly. The information of the elderly is transmitted through a smart local area network. Advanced Internet of Things communication protocols such as NB-IoT and LoRaWAN are adopted to ensure the stability and low power consumption of data transmission. Edge computing is implemented for preliminary data processing and analysis, reducing the dependence on the cloud platform and reducing latency.

[0053] Process the information of the elderly through cloud platform and artificial intelligence technologies, and at the same time, intelligently process the data of the daily life behaviors of the elderly. Utilize image classification technology and object tracking technology of machine learning algorithms to perform pattern recognition and behavior prediction on the collected data, so as to more accurately judge the needs and health status of the elderly. Develop personalized services and provide customized intelligent scenarios according to the historical data and preferences of the elderly.

[0054] Introduce a multi-modal multi-granularity graph convolutional network. By combining data of multiple modalities (such as videos, audio, sensor data) and features of different granularities (such as human postures, actions, expressions), improve the recognition accuracy of the daily behaviors of the elderly.

[0055] Introduce a convolutional neural network (CNN). By learning the features in video frames, it can effectively identify and classify the daily activities of the elderly, such as walking, sitting down, standing up, etc.

[0056] Introduce a generative adversarial neural network (GAN) to learn the daily activity behaviors of the elderly. For the living behavior habits and scenarios of the elderly, combined with a time series prediction model, improve the accuracy of behavior prediction for the elderly.

[0057] Introduce intelligent AI behavior monitoring technology. By real-time tracking the targets in the video and analyzing their behavior patterns and intentions, identify the daily activities and abnormal behaviors of the elderly.

[0058] Adopt smart home for intelligent conversion of the interactive scenarios in the home care space. Through changing the positions and forms of spatial interfaces such as the ceiling, floor, and wall for spatial interaction, scene conversion, intelligent lighting, glass, lifting devices, etc., interact with the elderly and provide a comfortable living environment.

[0059] The present invention effectively improves the physical sensation, audio-visual comfort level of the elderly, improves the psychological emotions of the elderly, and activates the attention, autonomy and cognitive ability of the elderly.

[0060] The technical solution of the present invention is realized as follows:

[0061] Step 1: Collect the daily activity behavior data of the elderly at home through devices such as cameras, human infrared sensors, and bioelectric sensors, including video streams, infrared images, bioelectric signals, etc.;

[0062] Step 2: Conduct real-time monitoring and tracking of behaviors through cameras and video analysis components, capture the actions of the elderly, identify gait, and measure bioelectricity; specifically as follows:

[0063] Step 2.1: Use computer vision technology, combining OpenPose and DeepSORT for real-time monitoring and tracking. OpenPose can detect and estimate the key points of the human body in real time from the video stream, including the key points of the body, hands, and feet, as well as facial features. DeepSORT combines a deep learning object detector and a tracking algorithm, enabling the tracking of objects between consecutive frames;

[0064] Step 2.2: Preprocess the collected data, including image segmentation, normalization, enhancement, etc., to improve the quality of the data and the training effect of the model;

[0065] Step 3: For the collected and preprocessed data, apply a smart local area network to establish a daily activity behavior database for the elderly at home and a database of physiological and psychological needs information of the elderly;

[0066] Step 4: Apply a convolutional neural network to learn the physiological and psychological data of the elderly to judge user characteristics and preferences and identify and learn the activity scenarios of the elderly; specifically as follows:

[0067] Step 4.1: Apply a convolutional neural network (CNN) of deep learning for image classification to identify the daily activities of the elderly, such as walking, sitting, standing, etc. Conduct real-time monitoring and tracking of behaviors through a video analysis component, and capture the movements, gait recognition, and bioelectric measurement of the elderly;

[0068] Step 4.2: Select ResNet for the CNN model. By introducing a residual learning framework, the degradation problem in the training of deep networks is solved, improving the accuracy and generalization ability of the model. The Inception module increases the width of the model by using convolutional kernels of different sizes in parallel, improving the feature extraction ability;

[0069] Step 4.3: Further, for the personalized interaction of the elderly's behavior activities - scenarios - items, a new multi-modal multi-granularity graph convolutional network (MM-GCN) is adopted. By applying the graph convolutional network in four modalities (skeleton, bone, frame, nomination segment), the semantics of the elderly's activities are captured;

[0070] Step 4.4: Use the preprocessed data as the training set, including video frames or images of different activity types of the elderly. Train the model through the backpropagation algorithm and optimizers (such as Adam, SGD, etc.), and adjust the model parameters to minimize the loss function;

[0071] Step 5: The CNN model extracts features from the input video frames or images, and these features can characterize the behaviors of the elderly. Use a Softmax classifier or a support vector machine (SVM), etc., to classify the behaviors of the elderly according to the extracted features;

[0072] Step 6: Deploy the trained CNN model into the real-time monitoring system to classify the behaviors and scenarios of the elderly in real time, including eating in the restaurant, cooking in the kitchen, entertaining in the living room, exercising in the living room, resting in the bedroom, etc.

[0073] Step 7: Apply the generative adversarial neural network to learn the daily activity behaviors of the elderly, predict the behaviors of the elderly according to their living behavior habits and scenarios, and establish various combination possibilities of the behaviors and activity scenarios of the elderly.

[0074] Step 7.1: Use historical behavior data to train the GAN model. The models selected are the deep convolutional generative adversarial network DCGA for image data generation and the cyclic generative adversarial network CycleGAN for cross-domain data generation.

[0075] Step 7.2: Alternately train the generator and the discriminator to generate realistic daily activity data of the elderly.

[0076] Step 7.3: Use clustering algorithms such as k-means and DBSCAN to identify behavior patterns. Use decision tree algorithms to analyze the living habits of the elderly.

[0077] Step 7.4: Further, for the prediction accuracy, use ARIMA and LSTM to establish a time series prediction model, and fuse the GAN model and the time series model. Combine the prediction results of multiple models and perform weighted averaging or voting mechanism.

[0078] Step 8: Based on the prediction results, use the smart home system to perform spatial interaction and scene conversion by changing the positions and forms of spatial interfaces such as the ceiling, the floor, and the wall.

[0079] Step 9: Control the ceiling interface through a liftable device to adjust the space height according to the behavior needs of the elderly. Through the intelligent control system, realize the automatic lifting of the ceiling interface.

[0080] Step 10: Adjust the wall interface using environmental sensors such as wind speed and light to monitor the environmental changes in real time. Apply intelligent color-changing glass and rotating glass to automatically adjust the transparency and angle according to the environmental changes; and combine with intelligent lights to automatically adjust the wall color to relieve the physical and visual comfort of the elderly.

[0081] Step 11: Ensure the safety of the floor interface by using a gait analyzer and algorithm based on 2D-LiDAR (Two-Dimensional Light Detection and Ranging) to monitor the walking patterns and paths of the elderly. Combine fall alarms and airbags to provide fall prevention and protection measures to meet the walking safety needs of the elderly.

[0082] Step 12: Integrate artificial intelligence algorithms with Internet of Things devices to achieve real-time collection, processing, and response of data. Through an intelligent decision-making system, achieve autonomous linkage and response in different scenarios.

[0083] Specific application scenario: An elderly person is sitting on the sofa in the living room watching TV

[0084] Device installation: Install a binocular battery-powered camera, a panoramic infrared camera, environmental sensors such as wind speed and light, and have the elderly wear smart bracelets or health monitoring vests at home.

[0085] Data collection: Cameras, human infrared sensors, smart bracelets, or health monitoring vests collect daily activity behavior data of the elderly at home. The data forms are video streams, infrared images, and bioelectrical signals.

[0086] Real-time data monitoring and analysis: Use the combination of OpenPose (Open Pose Estimation open-source pose estimation) and DeepSORT (Deep Simple Online and Realtime Tracker deep simple online real-time tracker) to perform real-time monitoring, data analysis, and predictive estimation on the video streams and infrared images of the elderly.

[0087] OpenPose can monitor in real-time and use a deep neural network to estimate human poses. It can accurately detect and estimate the key points and pose information of the human body in images or videos, including the key points of the body, hands, face, and feet.

[0088] DeepSORT combines a deep learning object detector and a tracking algorithm, capable of tracking multiple targets between consecutive frames. It can be used to track the activities of the elderly and maintain the consistency of target identities.

[0089] The combination of OpenPose and DeepSORT can achieve real-time multi-human pose estimation and multi-target tracking in various scenarios, providing accurate and real-time data analysis and processing.

[0090] Feature extraction, training, and recognition: Use a Convolutional Neural Network (CNN) to extract features, learn and train, and recognize historical behavior data such as video streams, infrared images, and bioelectrical signals collected, including scenes (bedroom, kitchen, living room, bathroom, study, dining room, balcony), human postures (standing, sitting, lying, kneeling, squatting, crawling, etc.); actions (walking, running, jumping, waving, nodding, shaking the head, shaking hands, hugging, waving, etc.); expressions (smiling, frowning, blinking, staring, pouting, opening the mouth, closing the eyes, biting the lip); interactive items (sofa, TV, bedding, cooking utensils, eating utensils, kitchen utensils, stationery, office supplies, cleaning utensils, fitness equipment, entertainment tools, etc.).

[0091] Model feature extraction: The CNN model selects the ResNet (Residual Network) residual network and the Inception (Inception Module heuristic module) module; for the personalized interaction of the behavior - scene - item of the elderly, a new multi - modal multi - granularity graph convolutional network (MM - GCN) is adopted to perform feature extraction and semantic capture from four modalities: human bone points, skeletons, frames, and proposals respectively.

[0092] Train and optimize the algorithm: Use the backpropagation algorithm and optimizers (such as Adam, SGD, etc.) to train the model and adjust the model parameters to minimize the loss function.

[0093] Behavior classification: Deploy the trained CNN model to a real - time monitoring system, and use a Softmax classifier or a Support Vector Machine (SVM), etc., to classify the behaviors of the elderly according to the extracted features; combine scenes, human postures, actions, expressions, and interactive items respectively, including eating in the dining room, cooking in the kitchen, entertaining in the living room, exercising in the living room, resting in the bedroom, reading in the study, etc.

[0094] Behavior prediction: Use the GAN model obtained by training with historical behavior data to predict the behaviors of the elderly according to their living behavior habits and scenes.

[0095] The model selects the Deep Convolutional Generative Adversarial Network DCGA for image data generation and the Cycle Generative Adversarial Network CycleGAN for cross - domain data generation, and the generator and discriminator are alternately trained.

[0096] By using the K-Means clustering algorithm to divide the behavior data of the elderly into several clusters, different behavior patterns can be identified, and the characteristics of these patterns can be analyzed. The DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm can better understand the behavior chain of the elderly by discovering behavior sequences with different densities.

[0097] To improve the accuracy of prediction, an ARIMA (Autoregressive Integrated Moving Average) and an LSTM (Long Short-Term Memory) are used to establish a time series prediction model. The GAN model and the time series model are fused, and the prediction results of multiple models are combined. A weighted average or voting mechanism is used to improve the accuracy of behavior prediction.

[0098] Scene interaction: Case application: The elderly are sitting on the sofa in the living room watching TV

[0099] The panoramic infrared camera monitors that the elderly appear in the living room scene;

[0100] OpenPose uses a deep neural network to estimate human poses and can accurately detect and estimate the key points and pose information of the human body in images or videos, including the key points of the body, hands, face, and feet. By analyzing these key points and combining with DeepSORT, the sitting posture of the elderly can be detected.

[0101] The human infrared sensor can detect the interaction between the elderly and objects, such as the TV or the remote control. Devices such as smart bracelets or health monitoring vests can monitor the bioelectrical signals of the elderly to further confirm the interaction between the elderly and the TV or the remote control.

[0102] The multi-modal multi-granularity graph convolutional network (MM-GCN) recognizes videos and images and corresponds to the trained behavior pattern classification system. The scene (living room), human posture (sitting posture), and interacting objects (TV, remote control) are for the behavior pattern of living room entertainment.

[0103] By analyzing the behavior pattern of the elderly in the living room and the interaction with the TV, the GAN models (DCGAN and CycleGAN) trained with historical behavior data and the time series prediction model combining ARIMA and LSTM can predict the behavior that the elderly are about to watch TV.

[0104] The system detects and predicts the behavior that the elderly are sitting on the sofa in the living room watching TV, and activates the wall interface adjustment function of the smart furniture system.

[0105] According to the recognition of the elderly's facial expressions, combined with intelligent lighting, automatically adjust the brightness, color, and color temperature of the wall surface, including cold and warm colors, to improve the elderly's visual experience and immersion.

[0106] Use environmental sensors such as wind speed and light to monitor environmental changes in real time. The intelligent electrochromic glass can automatically adjust its transparency according to environmental changes. If the outdoor light is strong, the intelligent electrochromic glass will automatically darken to reduce glare, protect the elderly's eyes, and improve the elderly's visual comfort.

[0107] Although the present invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed, as long as they do not depart from the spirit and scope of the present invention as defined by the appended claims. It should be understood that different dependent claims and the features described herein can be combined in a manner different from that described in the original claims. It should also be understood that the features described in connection with a single embodiment can be used in other described embodiments.

Claims

1. An in-home elderly care interaction scenario system based on AIoT technology, characterized in that, Including: A data acquisition unit (1), a wireless local area network (2), a data processing unit (3), a database (4), a smart home device control unit (5), and a behavior prediction model (6); The data acquisition unit (1) is used to collect bioelectric information and videos of a target person in different living scenarios through a video acquisition device, a human infrared sensor, and a bioelectric sensor; and send the collected bioelectric information and videos to the data processing unit (3) and the database (4) via the wireless local area network (2); The data processing unit (3) combines computer vision technology, OpenPose, and DeepSORT to extract the action postures and facial features of the target person from the video in real time; and transmits the extracted information to the database (4) and the behavior prediction model (6); The behavior prediction model (6) predicts the behavior of the target person at the next moment by using the bioelectric information, action postures, and facial features of the target person at the previous moment in the database (4); The behavior prediction model (6) is pre-trained through the historical behavior information, corresponding videos, and bioelectric information of the target person in different living scenarios; The data processing unit (3) also sends a control signal to the smart home device control unit (5) according to the behavior of the target person predicted by the behavior prediction model (6), and the smart home device control unit (5) controls the corresponding smart home device according to the received control signal.

2. The home-based elderly care interaction scenario system based on AIoT technology according to claim 1, wherein The data acquisition unit (1) includes a camera, a panoramic infrared camera, a wind speed sensor, a light sensor, a smart bracelet for the elderly to wear, and / or a health monitoring garment.

3. The home-based elderly care interaction scenario system based on AIoT technology according to claim 2, characterized in that, The bioelectric sensor is used to collect heart rate, heart beat, blood pressure, and body temperature information.

4. The home-based elderly care interaction scenario system based on AIoT technology according to claim 1, characterized in that The behavior prediction model (6) is implemented by using a GAN model.

5. The home-based elderly care interaction scenario system based on AIoT technology according to claim 1, characterized in that, The real-time extraction of the action postures and facial features of the target person from the video is implemented by using a CNN network.

6. The home care interaction scenario system based on AIoT technology according to claim 1, wherein The behavior prediction model (6) is trained by using historical behavior data and historical scenario data extracted from video streams, infrared images, and bioelectric signals; The historical behavior data includes: action data and expression data; The action data includes: walking, running, jumping, waving, nodding, shaking the head, shaking hands, hugging, and waving actions in different scenarios; The expression data includes: smiling, frowning, blinking, staring, pouting, opening the mouth, closing the eyes, and biting the lip expressions in different scenarios; The scenario data includes: location data and interaction item data; The location data includes: bedroom, kitchen, living room, bathroom, study, dining room, and balcony; The interaction item data includes: sofa, TV, bedding, cooking utensils, eating utensils, kitchen utensils, books and stationery, office supplies, cleaning utensils, fitness utensils, and entertainment tools.

7. The home-based elderly care interaction scenario system based on AIoT technology according to claim 1, wherein The behavior prediction model (6) uses the backpropagation algorithm and an optimizer to minimize the loss function of the model parameters.

8. The home-based elderly care interaction scenario system based on AIoT technology according to claim 1, characterized in that The control of the corresponding smart home device by the smart home device control unit (5) according to the received control signal includes intelligent light adjustment, intelligent glass color change adjustment, intelligent ceiling switch adjustment, air speed adjustment and temperature adjustment of the intelligent air conditioner, and alarm triggering of the emergency warning device.

Images