Anti-falling gait risk research and early warning system for old people

Through artificial intelligence technology, combined with human body detection, skeleton line key point regression and gait recognition, a research and early warning system for anti-fall gait gait for the elderly has been formed, which has solved the problems of falling risk monitoring and early warning for the elderly, and improved the quality of life and medical efficiency.

CN118948261APending Publication Date: 2024-11-15BEIJING VOCATIONAL COLLEGE OF SOCIAL MANAGEMENT
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411320358.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-20
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The risk of falling in the elderly has increased in the context of population aging, and the existing technology is difficult to effectively monitor and early warning, resulting in low quality of life and health management and low efficiency of the medical system.

Method used

Using artificial intelligence technology, through human body detection, skeleton line key point regression, gait recognition and gait variation risk prediction, a set of anti-fall gait risk research and early warning system for the elderly is formed to realize real-time monitoring and early warning to reduce the risk of falling.

Benefits of technology

It has improved the quality of life and health management of the elderly, achieved early detection, early warning and early delivery of hospitals, improved the efficiency of the medical system, reduced the risk of falling, and enhanced the sense of peace of mind for family members.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118948261A_ABST
    Figure CN118948261A_ABST
Patent Text Reader

Abstract

The invention provides an anti-falling gait risk research and early warning system for old people, and relates to the technical field of gait research and falling early warning, the anti-falling gait risk research and early warning system for old people adopts four modes to carry out anti-falling gait risk research and early warning on old people, the first mode is human body detection and face detection functions, and the second mode is human body detection and face detection functions; the first mode is skeleton line key point regression prediction, the second mode is skeleton line key point regression prediction, the third mode is a gait recognition method, and the fourth mode is gait variation risk prediction intelligent analysis. According to the method, a rule and content matching-based early warning mechanism for key events such as falling of the old people is researched, a complete gait variation risk prediction intelligent analysis framework for the old people is formed, the limitation that health monitoring of the old people only depends on manual monitoring of nursing workers, family members and the like can be overcome, sudden health events are prevented, and the safety of the old people is improved. The life quality and the health management level of the old people are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of gait research and fall warning, and in particular to a gait risk research and warning system for preventing falls of the elderly. Background Art

[0002] As the trend of population aging becomes increasingly obvious, the elderly are prone to various problems such as vision loss and reduced nerve sensitivity due to the decline in physical functions, which increases the risk of falling. The elderly are a high-risk group for fall injuries. The demand for services and care for the elderly is more prominent, and care institutions are facing a shortage of personnel. The superposition of these two risks makes care services synonymous with high risks. The elderly are a special group. Due to physiological and pathological changes in elderly patients, accidents are more likely to occur. Once a fall occurs, if they are not rescued and treated in time, the best treatment time will be missed, which will not only affect the quality of life of the elderly, but also cause serious harm to the life and health of the elderly, and even be an important factor in the death of the elderly.

[0003] Fall prevention and early warning are important guarantees for protecting the life, health and quality of life of the elderly. Therefore, the present invention proposes a gait risk research and early warning system for preventing falls for the elderly. In home and nursing home scenarios, the system can perform artificial intelligence analysis on the elderly's behavior through real-time recognition of the elderly's movement trajectory and gait, which has very important practical significance in preventing falls among the elderly and timely detecting falls among the elderly. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the shortcomings of the existing technology, the present invention provides a gait risk research and early warning system for preventing falls in the elderly, and a real-time health monitoring and analysis technology for the elderly at home, which can realize target detection of the elderly's face and position, skeleton line key point regression detection, and evaluation and analysis of the elderly's fall alarm, prevent sudden health events, improve the elderly's quality of life and health management level, achieve the goal of early detection, early warning, early medical treatment, and improve the efficiency of the medical system, and reduce the risk of falls through intelligent technology, so that family members can feel at ease.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a gait risk research and early warning system for the elderly to prevent falls. The gait risk research and early warning system for the elderly to prevent falls adopts four methods to conduct gait risk research and early warning for the elderly to prevent falls. The first method is the human body detection and face detection function. Starting from the perspective of human body integrity, the environmental impact and partial occlusion on the design target detection network are analyzed to realize the general rules of position detection and face detection for the elderly. The second method is the skeleton line key point regression prediction. The relationship between the spatial position of the imaging device and the human body and the general logic in the process of human body posture conversion are analyzed to study the skeleton line key point regression method. The third method is the gait recognition method. Starting from the spatiotemporal sequence of the trajectory formed by the key points, the human body posture, action and behavior modeling method is studied. The fourth method is the intelligent analysis of gait abnormality risk prediction. The early warning mechanism of key events such as falls of the elderly based on rules and content matching is studied to form a complete set of intelligent analysis framework for gait abnormality risk prediction for the elderly.

[0008] Preferably, the human body detection and face detection functions of the elderly anti-fall gait risk research and early warning system are based on artificial intelligence deep learning technology, and use the RetinaFace network and ST-YOLO network to perform accurate human target detection on the elderly. It can monitor the activity status of the elderly in real time, determine whether they are in a normal activity state, and identify abnormal situations in a timely manner.

[0009] Preferably, the skeleton line key point regression prediction of the elderly anti-fall gait risk research and early warning system is based on the posture recognition technology of skeleton line key point regression detection, which monitors the activity posture of the elderly in real time and identifies potential fall risks. By analyzing the position and posture changes of the elderly's skeletal key points, it can determine whether they are in a state of fall risk and issue an alarm in time.

[0010] Preferably, the posture recognition technology of the skeleton key point regression detection adds a key point regression task on the basis of the human body detection ST-YOLO detection model, and associates all key points of a person with the Anchor. Each person has 17 key points, a total of 51 elements. For each Anchor, the Keypoint Head predicts 51 elements, and the Boxhead predicts 6 elements. For an Anchor with n key points, the overall prediction vector is defined as:

[0011]

[0012] The key point confidence is trained based on the visibility landmarks of the key points.

[0013] Preferably, the gait recognition method of the elderly anti-fall gait risk research and early warning system starts from the spatiotemporal sequence of trajectories formed by key points, studies the modeling methods of human posture, motion and behavior, realizes real-time monitoring and alarm of elderly fall events, and forms the results of key point regression into a sequence and regards it as a graph node. In the feature embedding module, the input information is converted into model embedding, including each key point encoding, edge encoding, explicit bond angle information and space encoding. In this layer, the size of all embeddings is set to atom-number×512. This feature encoding method can present the key features in the data to the model in a more intuitive way. The backbone network is intended to adopt a dual-branch network composed of MPNN modules and GraphTransformer modules.

[0014] Preferably, the gait recognition method has two modules: a feature embedding module and an MPNN-Transformer parallel message passing module. The feature embedding module is a network module that encodes the input in the AGT framework, converts the coordinates, relative relationship information and angle information in the key point sequence into coordinate encoding, edge encoding, centrality encoding and angle encoding, and finally linearly accumulates all embeddings into the network model. The backbone network of the AGT of the MPNN-Transformer parallel message passing module is composed of a message passing module MPNN and a self-attention module of a Graph Transformer.

[0015] Preferably, the gait variation risk prediction intelligent analysis of the elderly anti-fall gait risk research and early warning system is based on a key event early warning mechanism for elderly falls based on rule and content matching, and integrates technologies such as face detection, posture recognition and fall detection to achieve comprehensive monitoring and analysis of the elderly's daily living status, making it convenient for family members and medical staff to understand the elderly's health status in a timely manner and take corresponding measures.

[0016] Preferably, the critical event warning mechanism realizes rapid identification and warning of abnormal behavior by analyzing the behavior patterns in the video surveillance in combination with the preset rule base, and utilizes computer vision technology, especially gait recognition algorithm, to perform real-time analysis on the monitoring video stream and extract key behavior features. Once the content matching the fall-related behavior patterns predefined in the rule base is detected, the system will automatically trigger the warning mechanism and promptly notify the caregiver or emergency service, thereby possibly avoiding the occurrence of serious injury.

[0017] (III) Beneficial effects

[0018] The present invention provides a gait risk research and early warning system for preventing falls in the elderly. It has the following beneficial effects:

[0019] The present invention provides a research and early warning system for the risk of gait falling for the elderly. The system has a good theoretical foundation in target detection and segmentation, deep learning, and behavior recognition. Applying it to this project is an extension and deepening of its theoretical research. It is an intelligent analysis system for the gait behavior of the elderly based on deep learning, and forms a complete intelligent analysis framework for the risk prediction of gait abnormality for the elderly. Aiming at the problem of fall risk for the elderly, the method of modeling human posture, movement and behavior is studied from the time-space sequence of trajectories formed by key points. The health monitoring of the elderly can overcome the limitations of manual monitoring by caregivers, family members, etc., prevent sudden health events, improve the quality of life and health management level of the elderly, and achieve the goal of early detection, early warning, early medical treatment, and improving the efficiency of the medical system. Therefore, the research results are in great demand and have broad prospects for industrialization. It is of great significance to protecting people's livelihood and improving people's sense of happiness, and has huge social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 The overall technical roadmap of the elderly fall prevention gait risk research and early warning system of the present invention;

[0021] Figure 2 The ST-YOLO overall network structure diagram of the elderly anti-fall gait risk research and early warning system of the present invention;

[0022] Figure 3 It is a structural diagram of the SPPF module of the elderly anti-fall gait risk research and early warning system of the present invention;

[0023] Figure 4 It is a CST module structure diagram of the elderly anti-fall gait risk research and early warning system of the present invention;

[0024] Figure 5 This is a RetinaFace network architecture diagram of the elderly anti-fall gait risk research and early warning system of the present invention;

[0025] Figure 6 This is a ResNet152 network structure diagram of the elderly anti-fall gait risk research and early warning system of the present invention;

[0026] Figure 7 This is a key point regression task flow chart based on ST-YOLO for the elderly anti-fall gait risk research and early warning system of the present invention;

[0027] Figure 8 This is the AGT overall model diagram of the elderly anti-fall gait risk research and early warning system of the present invention. DETAILED DESCRIPTION

[0028] 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 in 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. In the description of this application, it should be noted that the terms used here are only for describing specific implementations, and are not intended to limit the exemplary implementations according to the present application. For ease of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship. The techniques, methods and devices known to ordinary technicians in the relevant fields may not be discussed in detail, but where appropriate, the techniques, methods and devices should be regarded as part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as merely exemplary, not as a limitation. Therefore, other examples of the exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0029] Embodiment 1:

[0030] like Figure 1-8 As shown, an embodiment of the present invention provides a gait risk research and early warning system for preventing falls for the elderly. The gait risk research and early warning system for preventing falls for the elderly adopts four methods to conduct gait risk research and early warning for preventing falls for the elderly. The first method is human body detection and face detection functions. Starting from the perspective of human body integrity, the environmental impact and partial occlusion on the design target detection network are analyzed to realize the general rules of elderly position detection and face detection. The second method is skeleton line key point regression prediction. The spatial position relationship between the imaging device and the human body and the general logic in the process of human body posture conversion are analyzed to study the skeleton line key point regression method. The third method is gait recognition method. Starting from the trajectory spatiotemporal sequence formed by key points, the human body posture, action and behavior modeling method are studied. The fourth method is gait abnormal risk prediction intelligent analysis. The warning mechanism of key events such as elderly falls based on rules and content matching is studied to form a complete set of gait abnormal risk prediction intelligent analysis framework for the elderly. The key problems that need to be solved by the gait risk research and early warning system for preventing falls for the elderly are:

[0031] S1: Adapt to diverse home environments. The system needs to be highly adaptable to cope with different room layouts, furniture placement, and daily activity patterns. Develop automatic analysis and compensation technology for "visual blind spots" and occlusion phenomena to ensure the continuity and accuracy of monitoring. At the same time, study intelligent perception and scene modeling methods for different environmental characteristics to improve the robustness and adaptability of the system in diverse home environments.

[0032] S2: Camera perspective optimization, the impact of different camera perspective changes on behavior recognition and gait analysis. Reduce the detection error caused by angle deviation by optimizing camera layout and adjusting algorithm parameters. Study multi-camera collaborative fusion technology and perspective-independent behavior recognition algorithms to ensure the accuracy and stability of behavior recognition and gait analysis under different perspectives. Further, explore the possibility of flexible monitoring with mobile camera equipment to improve the system's ability to cope with dynamic environments and temporary scenes.

[0033] S3: The impact of image quality fluctuations. Image quality fluctuations caused by changes in lighting and environmental conditions affect detection accuracy. Develop image processing algorithms that can maintain high detection accuracy under various lighting conditions, including low-light enhancement, backlight compensation, and mapping technology from black and white images to color images, to achieve unified processing and efficient recognition of different image qualities. In addition, it is necessary to study image preprocessing and enhancement technologies under dynamic lighting and complex environments to ensure the stability of the system in practical applications.

[0034] S4: Efficient real-time early warning mechanism, research on early warning mechanism for key events such as falls of the elderly based on rules and content matching. Combine multi-source sensor data (such as cameras, wearable devices, etc.) to establish a multimodal data fusion algorithm to improve the accuracy and response speed of early warning. Develop a convenient emergency response system to achieve real-time alarm and rapid processing of emergencies to ensure that the elderly can receive timely assistance in emergency situations. Furthermore, build an adaptive risk prediction model to improve early warning capabilities, thereby reducing the incidence of sudden health events.

[0035] Embodiment 2:

[0036] like Figure 1-8 As shown, the embodiment of the present invention provides a gait risk research and early warning system for preventing falls in the elderly. The working principle of the system is as follows Figure 1 As shown in the figure, the RetinaFace network and the ST-YOLO network are used to accurately detect human targets for the elderly, so that the position of people can be located in the home environment, and a model with excellent structure recognition ability is trained using a self-supervised method. The input image can obtain consistent and interpretable key points and skeleton edge maps to achieve skeleton line key point regression detection for the elderly. The AGT model is used for gait recognition. The input is the coordinates of the key points of the connection, and the output is the result of gait recognition.

[0037] The model mainly includes the feature embedding module FeatureEmbedding and the MPNNTransformer parallel message passing module. Through this model, important feature information can be extracted from the structural data of the human body, and this information can be used to accurately predict the gait attributes of the elderly.

[0038] Target detection: In order to analyze the activity status and gait of the elderly, the first step is to detect the elderly. Since the size and color of the human body in the indoor scene are different, and there may be partial occlusion, these add a certain degree of difficulty to the detection task of the human body. To meet the needs of the indoor scene detection task, the ST-YOLO detection model is designed based on YOLOv8. The model has excellent global feature modeling capabilities and is designed to achieve accurate recognition and two-dimensional positioning of the human body in indoor scenes. The overall network framework is as follows: Figure 2 shown.

[0039] The overall structure of ST-YOLO is mainly composed of three parts: feature extraction network, feature fusion network and prediction network. The backbone network is planned to adopt CSPDarkNet-53, which is designed based on the improved and optimized Darknet-53 network structure. The introduced CSP module can fuse features at different levels, enhance the expressiveness and discrimination of features, and the CSP module can better utilize feature information and improve the performance of the model. In the feature reuse stage, CSPDarkNet-53 uses residual connection and jump connection technology. These technologies can better utilize the information of the middle layer of the network, solve the problem of loss of carton texture details and edge information after multi-layer convolution and pooling of feature maps, and prevent gradient disappearance and gradient explosion. The image in the ST-YOLO network is input into the feature extraction network composed of ordinary convolution layer CST module and SPPF module. The ordinary convolution layer compresses the image resolution through downsampling operation. The CST module inputs the feature map into the convolution layer and Transformer In the branch composed of blocks, the other branch is directly passed as a bypass. The feature maps obtained by the two branches are fused and then output to the next module. Usually, multiple CST modules are stacked after the downsampling convolution layer. After multiple downsampling operations, a preliminary backbone network can be formed. The structural design of the SPPF module is as follows: Figure 3 As shown in the figure, its role is to effectively capture contextual information of different scales and expand the receptive field of the backbone network output features through spatial pyramid pooling and feature fusion without changing the size of the feature map.

[0040] Taking into account the changes in the height and posture of the elderly, it is necessary to further improve the global feature extraction capability of the network. CNN extracts global features through the pooling layer. The pooling layer captures the global features of the input data by downsampling the input data. When extracting global features, the pooling layer of CNN will cause some subtle features of the input data to be lost, and it is unable to fully capture the complex correlation between the local area and the whole. In addition, the convolution kernel in the CNN convolution layer is locally aware, and the parameter sharing mechanism enables the same convolution kernel to slide on the entire input image, resulting in the localization of features. In order to solve the problem of CNN feature locality, this system intends to design a CST module based on a hybrid CNN-Transformer, using a multi-head self-attention mechanism and a convolution fusion module to extract features. The overall structural design of the CST module is as follows: Figure 4 As shown in the figure, it consists of N SwinBottleneck stacks and residual connections. The SwinBottleneck module is a convolutional basic module designed based on the Transformer Block and convolutional modules in the Swin Transformer. The Transformer Block is a module that can perform global context modeling on input features. Through the self-attention mechanism, it can capture the long-range dependencies of input features, thereby enhancing the expressiveness of features and effectively capturing global correlation.

[0041] Considering that in indoor environments, it is very likely that the elderly’s body parts are blocked but their faces are not, we plan to use face detection and human body detection simultaneously. As long as one of the targets is detected, the activities of the elderly indoors can be located and analyzed. For the face detection part, the RetinaFace overall network architecture is used as follows: Figure 5 As shown in the figure, after facial features are extracted by ResNet152, the FPN feature pyramid is used to enhance the features, and then Multi-task Loss is input to determine the final coordinates of the key points. Specifically, the RetinaFace network first sets some priori boxes on the image, gives the corresponding confidence to these priori boxes through the prediction results of the network, deletes the priori boxes with low confidence that do not contain the face, and for the priori boxes with high confidence, translates their centers to obtain the coordinates of the five key points of the face.

[0042] There are two options for RetinaFace feature extraction network: ResNet152 and MobilenetV1-0.25. RetinaFace is used to detect facial key points in images. The ResNet152 network structure is as follows: Figure 6 As shown in the figure, its core structure is a residual structure, which uses skip connections to alleviate the gradient explosion phenomenon caused by stacking too many network layers.

[0043] Based on the ST-YOLO detection model for human detection, the process of adding key point regression tasks is as follows Figure 7 As shown in the figure, if a key point is visible or occluded, the Ground Truth confidence is set to 1, otherwise if the key point is outside the field of view, the confidence is set to 0. During the reasoning process, the confidence of the key point should be kept greater than 0.5. All other predicted key points are blocked and the predicted key point confidence is not used for evaluation. Since the network predicts all 17 key points for each detection, the key points outside the field of view need to be filtered out, otherwise there will be low confidence key points that lead to deformed skeletons.

[0044] In the process of gait recognition, the results of key point regression form a sequence and are regarded as graph nodes. The input information is converted into model embedding in the feature embedding module, including each key point encoding, edge encoding, explicit bond angle information and space encoding. In this layer, the size of all embeddings is set to atom-number×512. This novel feature encoding method can present the key features in the data to the model in a more intuitive way. The backbone network of this study is proposed to adopt a dual-branch network composed of MPNN modules and Graph Transformer modules. In order to balance the complexity and computational efficiency of the model, the MPNN network module is designed as 3 layers and the GT network module is designed as 4 layers to enhance the model's modeling ability for complex graph structures. In the dual-branch network, the feature representation is input into the global attention module and the local message passing module respectively. The message passing mechanism captures the local relationship and interaction between the node and its directly adjacent nodes by aggregating the local neighbor information of the node and updating the node representation. The AGT model combines MPNN and GT, comprehensively considers local information and global information, and realizes the learning of graph-level representation.

[0045] The feature embedding module is a network module that encodes the input in the AGT framework, converting the coordinates, relative relationship information and angle information in the key point sequence into coordinate encoding, edge encoding, centrality encoding and angle encoding, and finally linearly accumulating all embeddings into the network model, while the MPNN-Transformer parallel message passing module, such as Figure 8 As shown in the figure, the backbone network of AGT is composed of the message passing module MPNN and the self-attention module of Graph Transformer. By combining the two models, AGT fully integrates their advantages in capturing material topological structure information and modeling dependencies between nodes, making it show stronger characterization capabilities in crystal structure representation and property prediction.

[0046] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A gait risk research and early warning system for preventing falls in the elderly, characterized by: The elderly anti-fall gait risk research and early warning system adopts four methods to conduct anti-fall gait risk research and early warning for the elderly. The first method is the human body detection and face detection function. Starting from the perspective of human body integrity, it analyzes the impact of the environment and partial occlusion on the design target detection network, and realizes the general rules of elderly position detection and face detection. The second method is the skeleton line key point regression prediction, which analyzes the spatial position relationship between the imaging device and the human body and the general logic of the human body posture conversion process, and studies the skeleton line key point regression method. The third method is the gait recognition method, starting from the trajectory spatiotemporal sequence formed by the key points, and studying the human body posture, action and behavior modeling method. The fourth method is the intelligent analysis of gait abnormality risk prediction, which studies the early warning mechanism of key events such as elderly falls based on rules and content matching, and forms a complete set of gait abnormality risk prediction intelligent analysis framework for the elderly.

2. The elderly fall prevention gait risk research and early warning system according to claim 1 is characterized by: The human body detection and face detection functions of the elderly anti-fall gait risk research and early warning system are based on artificial intelligence deep learning technology. They use the RetinaFace network and the ST-YOLO network to perform accurate human target detection on the elderly. They can monitor the activity status of the elderly in real time, determine whether they are in a normal activity state, and identify abnormal situations in a timely manner.

3. The elderly fall prevention gait risk research and early warning system according to claim 1 is characterized by: The skeleton line key point regression prediction of the elderly anti-fall gait risk research and early warning system is based on the posture recognition technology of skeleton line key point regression detection. It monitors the activity posture of the elderly in real time and identifies potential fall risks. By analyzing the position and posture changes of the elderly's skeletal key points, it can determine whether they are in a state of fall risk and issue an alarm in time.

4. The elderly fall prevention gait risk research and early warning system according to claim 3 is characterized by: The posture recognition technology of skeleton line key point regression detection is based on the human body detection ST-YOLO detection model, and adds a key point regression task to associate all key points of a person with Anchor. Each person has 17 key points, a total of 51 elements. For each Anchor, Keypoint Head predicts 51 elements and Boxhead predicts 6 elements. For an Anchor with n key points, the overall prediction vector is defined as: The key point confidence is trained based on the visibility landmarks of the key points.

5. The elderly fall prevention gait risk research and early warning system according to claim 1 is characterized by: The gait recognition method of the elderly anti-fall gait risk research and early warning system starts from the spatiotemporal sequence of trajectories formed by key points, studies the modeling methods of human posture, movement and behavior, realizes real-time monitoring and alarm of elderly fall events, and organizes the results of key point regression into sequences and regards them as graph nodes. In the feature embedding module, the input information is converted into model embedding, including each key point encoding, edge encoding, explicit bond angle information and space encoding. In this layer, the size of all embeddings is set to atom-number×512. This feature encoding method can present the key features in the data to the model in a more intuitive way. The backbone network is planned to adopt a dual-branch network composed of MPNN module and Graph Transformer module.

6. The elderly fall prevention gait risk research and early warning system according to claim 5 is characterized by: The gait recognition method has two modules: a feature embedding module and an MPNN-Transformer parallel message passing module. The feature embedding module is a network module that encodes the input in the AGT framework, converts the coordinates, relative relationship information and angle information in the key point sequence into coordinate coding, edge coding, centrality coding and angle coding, and finally linearly accumulates all embeddings into the network model. The AGT backbone network of the MPNN-Transformer parallel message passing module is composed of a message passing module MPNN and a self-attention module of a Graph Transformer.

7. The elderly fall prevention gait risk research and early warning system according to claim 1 is characterized by: The gait risk research and early warning system for preventing falls in the elderly uses intelligent analysis to predict gait abnormality risks. It is based on a key event early warning mechanism for falls in the elderly based on rule and content matching, and integrates technologies such as face detection, posture recognition, and fall detection to achieve comprehensive monitoring and analysis of the daily living status of the elderly, making it convenient for family members and medical staff to understand the health status of the elderly in a timely manner and take corresponding measures.

8. The elderly fall prevention gait risk research and early warning system according to claim 7 is characterized by: The critical event warning mechanism analyzes behavioral patterns in video surveillance and combines them with a preset rule library to quickly identify and warn of abnormal behaviors. It uses computer vision technology, especially gait recognition algorithms, to perform real-time analysis of surveillance video streams and extract key behavioral features. Once content that matches the fall-related behavioral patterns predefined in the rule library is detected, the system will automatically trigger the warning mechanism and promptly notify caregivers or emergency services, thereby potentially avoiding serious injuries.

Citation Information

Cited By

  • Ward monitoring method and system based on multi-modal large model and edge calculation, terminal and storage medium

    CN121053587A