VR management system and method for preventing venous thromboembolism, electronic equipment and storage medium

Abstract

The invention provides a VR management system and method for preventing venous thromboembolism, electronic equipment and a storage medium, and relates to the technical field of medical information, the VR management system comprises VR glasses and a central control unit, and a training action configuration module in the central control unit matches training action information according to a prevention VTE basic prevention scheme in combination with user basic information; the virtual scene building module builds a virtual training scene according to the training action information, sends a training task to the VR glasses, and obtains a user feedback action; the user action comparison model compares the user feedback action with the training action information to obtain a deviation action; and the loop decision module is used for returning the deviation actions to the virtual scene building module to execute a loop when the number of the deviation actions is not equal to 0. The technical problem that in the prior art, training actions are relatively fixed, adaptive adjustment is difficult to conduct according to the actual physical condition of a user, and consequently the readability of the training actions is poor is solved, and the adaptability and readability of the training actions are improved.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to a VR management system, method, electronic device and storage medium for preventing venous thromboembolism. Background Art

[0002] Venous thromboembolism (VTE) is a common clinical disease caused by venous obstructive reflux disorder due to venous thrombosis. The traditional basic prevention method of VTE is to conduct health education through doctors or nurses and guide patients to do basic prevention exercises, which is labor-intensive and has high human resource costs. At the same time, the content of basic prevention education is often not interesting and patients are not proactive, resulting in unsatisfactory prevention results.

[0003] With the application of virtual reality (VR) technology in the field of rehabilitation medicine, basic VTE prevention is carried out through the immersive experience and interactive functions of virtual scenes, which increases the fun of training methods, improves patient participation, provides multi-sensory stimulation, and improves patients' cognitive functions. However, the existing VR technology still has significant shortcomings in VTE prevention. On the one hand, the existing VR technology is still based on traditional prevention and education content for action display, and the rehabilitation movements and scene designs are too single to meet the physical conditions and training needs of different patients. On the other hand, movement guidance and feedback are often one-way transmission, and real-time correction and dynamic optimization cannot be achieved, and the patient's movement quality and training effect are difficult to guarantee. These problems limit the widespread application and effectiveness of VR technology in VTE prevention. Summary of the invention

[0004] The present application provides a VR management system, method, electronic device and storage medium for preventing venous thromboembolism, which solves the technical problem in the prior art that the training movements are relatively fixed and difficult to adapt according to the user's actual physical condition, resulting in poor readability of the training movements, and achieves the technical effect of improving the adaptability and readability of the training movements, thereby improving the user experience.

[0005] In view of the above problems, the present application also provides a VR management system for preventing venous thromboembolism, the system comprising: VR glasses; a central control unit, comprising: a training action configuration module, used to match training action information according to the basic prevention plan for VTE in combination with user basic information; a virtual scene construction module, used to build a virtual training scene according to the training action information, and send training tasks to the VR glasses to obtain user feedback actions; a user action comparison model, used to compare user feedback actions with the training action information to obtain deviation actions; a loop decision module, used to return the deviation actions to the virtual scene construction module to execute a loop when the number of the deviation actions is not equal to 0.

[0006] On the other hand, the present application also provides a VR management method for preventing venous thromboembolism, the method comprising: matching training action information according to a basic VTE prevention plan in combination with user basic information; building a virtual training scene according to the training action information, and sending training tasks to VR glasses to obtain user feedback actions; comparing user feedback actions with the training action information to obtain deviation actions; when the number of deviation actions is not equal to 0, returning the deviation actions to the virtual scene building module to execute a loop.

[0007] In a third aspect, the present application further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the execution steps of the above-mentioned VR management system for preventing venous thromboembolism are implemented.

[0008] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the execution steps of the above-mentioned VR management system for preventing venous thromboembolism are implemented.

[0009] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0010] Through modular design, this application realizes complete closed-loop control from training action configuration to feedback optimization, and provides a systematic solution for the personalized needs in the prevention of venous thromboembolism. First, the training action configuration module selects appropriate rehabilitation actions based on the patient's basic information and prevention plan, provides accurate input for the entire system, and ensures the scientificity and pertinence of personalized training. Then, the virtual scene construction module constructs an immersive scene based on the matching action information, making the training process interesting and interactive, and stimulating user participation. Through the user action comparison model, user actions are captured in real time and compared with the target action, deviation actions are identified, and accurate training feedback is provided. Finally, the loop decision module dynamically adjusts the scene or action configuration based on the deviation action results, forming a closed-loop optimization and continuously improving the training effect.

[0011] In summary, this application has constructed a rehabilitation training system that integrates fun, personalization and efficiency through precise configuration, virtual scene construction, real-time feedback and loop optimization, which overcomes the shortcomings of traditional prevention methods such as low compliance and delayed feedback, and significantly improves the readability and adaptability of VR technology training content for preventing venous thromboembolism, thereby enhancing the user experience.

[0012] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A schematic diagram of the structure of a VR management system for preventing venous thromboembolism provided in an embodiment of the present application.

[0014] Figure 2 A flowchart of matching training action information in a VR management system for preventing venous thromboembolism provided in an embodiment of the present application.

[0015] Figure 3 A flowchart of obtaining deviation actions in a VR management system for preventing venous thromboembolism provided in an embodiment of the present application.

[0016] Figure 4 A flowchart of a VR management method for preventing venous thromboembolism provided in an embodiment of the present application.

[0017] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0018] Figure numerals: VR glasses 10, central control unit 20, training action configuration module 21, virtual scene building module 22, user action comparison model 23, loop decision module 24, bus 300, receiver 301, processor 302, transmitter 303, memory 304, bus interface 305. DETAILED DESCRIPTION

[0019] The overall idea of ​​the technical solution provided by this application is as follows:

[0020] An interactive interface is provided by the VR glasses 10, and the central control unit 20 provides functional support: the training action configuration module 21 matches the training action information according to the basic prevention plan for VTE and the basic information of the user; the virtual scene construction module 22 builds a virtual training scene according to the training action information, and sends the training task to the VR glasses 10 to obtain the user feedback action; the user action comparison model 23 compares the user feedback action with the training action information to obtain the deviation action; when the number of the deviation actions is not equal to 0, the loop decision module 24 returns the deviation action to the virtual scene construction module 22 to execute the loop.

[0021] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically described below in conjunction with the drawings in the specification.

[0022] like Figure 1 As shown, an embodiment of the present application provides a VR management system for preventing venous thromboembolism, and the system includes: VR glasses 10 and a central control unit 20.

[0023] Specifically, the VR glasses 10 are a window for interaction between the user and the central control unit 20, and usually include a pair of display screens (one for each eye) and various sensors such as accelerometers, gyroscopes, and position sensors. The VR glasses 10 present virtual scenes and training tasks to the user through the display screens, and receive user action feedback through various sensors.

[0024] The central control unit 20 is the core control unit of the entire system, including a training action configuration module 21 , a virtual scene construction module 22 , a user action comparison model 23 and a loop decision module 24 .

[0025] The training action configuration module 21 is used to match the training action information according to the basic prevention plan for VTE in combination with the basic information of the user.

[0026] Specifically, the basic prevention program for VTE prevention is a training program designed based on medical standards or best practices, which includes a series of exercises or exercise activities to improve blood circulation and reduce the risk of thrombosis. Different users may need different training actions due to different physical conditions. The training action configuration module 21 matches and formulates appropriate training actions based on the basic prevention program for VTE prevention and combined with the user's personal basic information, such as age, gender, health status, etc., to ensure the effectiveness and safety of the training.

[0027] The virtual scene building module 22 is used to build a virtual training scene according to the training action information, and send the training task to the VR glasses 10 to obtain user feedback actions.

[0028] Specifically, the virtual scene building module 22 constructs a virtual training scene based on the action information provided by the training action configuration module 21, and transmits the training task to the VR glasses 10, while receiving the user's feedback action. This module provides an immersive training environment for users through VR technology, making the training more vivid and effective to enhance the user's sense of participation; and provides data for subsequent action comparison by real-time monitoring of the user's actions.

[0029] The user action comparison model 23 is used to compare the user feedback action with the training action information to obtain the deviation action.

[0030] Specifically, the user action comparison model 23 receives the user feedback action in the virtual scene construction module 22 as input information, compares the user feedback action with the preset training action information, identifies the deviation between the two, and marks the deviation as a deviation action. Through action comparison, the model evaluates whether each action of the patient is consistent with the preset training action, provides targeted feedback for subsequent deviation adjustments, and helps users adjust and optimize training actions.

[0031] The loop decision module 24 is used to return the deviation action to the virtual scene building module 22 to execute a loop when the number of the deviation actions is not equal to 0.

[0032] Specifically, the loop decision module 24 makes decisions based on the feedback results of the user action comparison model 23. When a deviation action is detected, the loop decision module 24 triggers the feedback mechanism and transmits the deviation action back to the virtual scene construction module 22 to update the training task and perform loop optimization. This module ensures that the VR system can continuously adjust and improve the training action until the expected training effect is achieved.

[0033] Further, such as Figure 2 As shown, the basic user information includes user age, user gender, user weight and user height. The training action configuration module 21 in the embodiment of the present application is also used to perform the following steps:

[0034] Step P11: Obtain the first exercise video of the basic prevention plan for preventing VTE.

[0035] Step P12: Perform semantic segmentation on the first motion video to obtain a first action sequence.

[0036] Step P13: input the first action sequence, the user age, the user gender, the user weight and the user height into an action matching model to obtain the training action information.

[0037] Specifically, user basic information is used to describe the user's basic physiological data, including basic data such as age, gender, weight and height. These basic information are the basis for adjusting the basic prevention plan for VTE.

[0038] An exercise video containing standard VTE prevention movements was obtained from the video library of the VTE Prevention Basic Prevention Program and recorded as the first exercise video. This video showed standardized training movements, such as a series of leg extensions or ankle rotations.

[0039] The first motion video is analyzed and processed through semantic segmentation technology, each frame in the video is segmented and labeled, and the entire video is decomposed into continuous actions to form a first action sequence. For example, leg bending, stretching or other movements in the video will be marked and extracted into an action sequence. Semantic segmentation not only recognizes actions, but also extracts key time points of actions, such as the start and end of the action.

[0040] The action matching model is a pre-built model that receives action sequences and basic user information such as age, gender, weight, and height. After internal algorithm and logic processing, it outputs training action information suitable for the user. The first action sequence in the first motion video and the basic information of the user, such as age, gender, weight, and height, are input into the action matching model. The model generates training action information that meets the actual situation of the user by analyzing these input data. For example, based on the weight and height of an older user, low-intensity leg lifts and slow gait exercises are recommended, while for younger users, slightly challenging walking or leg stretching movements may be recommended.

[0041] Through the above series of steps, the training action configuration module 21 can extract action information from standardized exercise videos and perform personalized training action configuration in combination with the patient's basic information, thereby ensuring that each user can obtain a VTE prevention training plan that matches their own physical condition and improving the feasibility of the training plan.

[0042] Furthermore, step P12 further includes:

[0043] Step P121: Collect motion video recording data and joint position identification data.

[0044] Step P122: Using the joint position identification data as supervision and the motion video recording data as input to train a convolutional neural network to generate a semantic segmentation model.

[0045] Step P123: Perform semantic segmentation on the first motion video according to the semantic segmentation model to obtain the first action sequence.

[0046] Specifically, the premise of semantic segmentation is to train a semantic segmentation model suitable for motion video segmentation. First, collect motion video recording data and corresponding joint position identification data. Motion video recording data is the original video data used to capture standard motion. These videos record the specific details of the human body when performing certain actions, such as the movement trajectory of the joints, the start and end time of the action, etc. The joint position identification data is the positioning information of the key parts of the human body (such as elbows, knees, ankles and other joints) in the motion video recording data, which is used to accurately mark the spatial position of the joints in human body movements.

[0047] Data collection can be done using cameras or video recording equipment and motion capture equipment. For example, in a laboratory environment, wear a motion capture suit with markers to perform various types of movements of various parts of the body, record videos with multiple cameras and capture the positions of the markers, thereby obtaining motion video recording data and joint position identification data.

[0048] Construct a model framework based on convolutional neural network, such as U-Net, FCN (fully convolutional network), etc. Use a deep learning framework (such as PyTorch or TensorFlow), use motion video recording data as input, and joint position identification data as supervision signals to train the convolutional neural network model framework. During the training process, the convolutional neural network performs multi-layer convolution operations (extracting local features), pooling operations (reducing dimensions to maintain key features), and full connection operations (completing classification or segmentation) on the input motion video data, outputs the recognition and classification results, compares the joint position identification data corresponding to the input motion video data and the output recognition and classification results, and continuously optimizes the parameters in the convolutional neural network based on the comparison results, so that the output of the convolutional neural network model gradually approaches the correct result (i.e., joint position identification data), and outputs the trained convolutional neural network model as a semantic segmentation model.

[0049] The first motion video is input into the semantic segmentation model for semantic segmentation. The model processes each frame of the video and classifies each pixel into a specific body part or action category. For example, the legs are marked as "bend" or "stretch"; the ankle movements are marked as "rotation" or "stationary". According to the time sequence of the video, these classification results are summarized into the first action sequence to clarify each action in the first motion video and its execution order.

[0050] The above steps use convolutional neural network training to generate a semantic segmentation model, which can accurately extract the time, location and execution method of the action from the standard training video, providing high-precision input for subsequent training action matching.

[0051] Furthermore, step P13 also includes:

[0052] Step P131: Configure the initial action sequence.

[0053] Step P132: Obtain a first user motion video of the initial action sequence, and obtain a first identified action sequence by performing semantic segmentation on the first user motion video.

[0054] Step P133: until the Mth user motion video of the initial action sequence is obtained, the Mth identified action sequence is obtained by performing semantic segmentation on the Mth user motion video.

[0055] Step P134: Add the first user basic information to the Mth user basic information to the user basic information record data set, and add the first identification action sequence to the Mth identification action sequence to the matching identification action data set.

[0056] Step P135: Using the initial action sequence and the user basic information record data set as input and the matching identification action data set as supervision, a convolutional neural network is trained to generate the action matching model.

[0057] Specifically, semantic segmentation is performed on the standard video in the basic prevention program for VTE prevention to generate a time series of action decomposition, namely the initial action sequence. A user (first user) is selected to perform movements according to this initial action sequence, and a video recording device (such as an ordinary camera or a professional motion capture camera) is used to record the motion video to obtain the first user motion video of the first user. The first user motion video is analyzed using semantic segmentation technology to identify each action in the video and convert it into a first identified action sequence. This sequence reflects the action of the first user when he actually performs the initial action sequence.

[0058] Continue to collect motion videos of different users exercising according to the initial motion sequence, and also use semantic segmentation technology to analyze the user's motion videos to obtain the corresponding identification motion sequence. M represents the number of users who collected the video (M is a positive integer), and a total of M user motion videos and their corresponding M identification motion sequences are obtained.

[0059] Each time a video is captured, the basic information of the corresponding user is collected, including the user's age, gender, weight, and height. The collected M sets of user basic information are added to the user basic information record dataset, and each user's identification action sequence is added to the matching identification action dataset. These two datasets provide training data for the next step of training the action matching model.

[0060] Construct a convolutional neural network framework, take the initial action sequence and the user basic information record data set as input data, and use the matching identification action data set as the supervision signal to train the convolutional neural network framework. For each user's basic information and initial action sequence combination, the convolutional neural network will extract and process features and output a predicted action sequence. By comparing the difference between the output of the convolutional neural network and the supervision data (i.e., the corresponding identification action sequence in the matching identification action data set), the back propagation algorithm is used to continuously adjust the weight parameters of the network to reduce this difference. Continue to iterate training until the convolutional neural network reaches convergence. At this point, the training is completed, and the trained convolutional neural network is output as an action matching model, which can accurately output the matching action sequence based on the initial action sequence and the user's basic information.

[0061] By collecting a large amount of user action data to train the action matching model, it is possible to adjust the intensity and complexity of the action according to the user's specific situation to ensure the feasibility and safety of the training action.

[0062] Further, such as Figure 3 As shown, the user action comparison model 23 in the embodiment of the present application is also used to perform the following steps:

[0063] Step P31: Obtain an action comparison identification model, wherein the action comparison identification model has a first feature extraction channel, a second feature extraction channel and a deviation action identification layer, and the first feature extraction channel and the second feature extraction channel are twin networks.

[0064] Step P32: Input the user feedback action into the first feature extraction channel to obtain a first action feature vector.

[0065] Step P33: Input the training action information into the second feature extraction channel to obtain a second action feature vector.

[0066] Step P34: Compare the first action feature vector and the second action feature vector through the deviation action identification layer to obtain the deviation action.

[0067] Specifically, the action comparison identification model is a trained deep learning model that identifies the deviation of the user when performing the training action by extracting and comparing the features of the user feedback action with the preset training action. The model includes a first feature extraction channel, a second feature extraction channel and a deviation action identification layer. The first feature extraction channel and the second feature extraction channel are twin networks with the same structure and parameters, which are used to extract the features of the user feedback action and the training action information respectively.

[0068] The actions performed by the user according to the training requirements (i.e., user feedback actions) are input into the first feature extraction channel. This channel extracts key feature information from the user feedback actions, such as angle changes and position changes of joints, and converts these feature information into the first action feature vector and outputs it to the deviation action identification layer.

[0069] The training action information (i.e., the standard action sequence) is input into the second feature extraction channel. This channel extracts the features of the preset training actions and converts them into the second action feature vector and outputs them to the deviation action identification layer.

[0070] The deviation action identification layer compares the first action feature vector and the second action feature vector, identifies the difference between the two feature vectors, and thus identifies the deviation action between the user feedback action and the training action information.

[0071] Through the above steps, the action comparison identification model can accurately compare the differences between the actions performed by the user and the preset training actions. The twin network architecture ensures the symmetry of feature extraction between the user and the preset training actions, making the action comparison process more reliable. The action feature vector extracted by the feature extraction channel can accurately capture the details of the action, such as changes in joint angles and positions. Finally, through the deviation action identification layer, deviations in user actions can be identified and marked, including inconsistent joint positions, insufficient movement amplitude, and inconsistent action timing. Such real-time feedback can continuously optimize the training plan according to the user's action performance, and improve the adaptability and readability of the training plan.

[0072] Furthermore, step P34 also includes:

[0073] Step P341: Extract the first action feature vector at the first moment based on the first action feature vector.

[0074] Step P342: Extract the second action feature vector at the first moment based on the second action feature vector.

[0075] Step P343: Calculate the Euclidean distance between the first action feature vector at the first moment and the second action feature vector at the first moment to obtain a first action deviation parameter.

[0076] Step P344: When the first action deviation parameter is greater than or equal to the deviation parameter threshold, the first moment feedback action and the first moment training action are added to the deviation action.

[0077] Specifically, the first action feature vector is a vector obtained by extracting the features of the user feedback action through the first feature extraction channel, which contains a vector sequence of multiple time steps and represents the features of the user feedback action at different times. In order to analyze the deviation at a specific time, it is necessary to extract the sub-vector at the corresponding time from this vector sequence, that is, the first action feature vector at the first time.

[0078] Similarly, the second action feature vector is also a vector sequence, representing the features of the training action at different moments. The sub-vector corresponding to the first moment is extracted from it, that is, the second action feature vector at the first moment.

[0079] By calculating the Euclidean distance between the first action feature vector at the first moment and the second action feature vector at the first moment, a value, namely the first action deviation parameter, can be obtained, which quantifies the difference between the two vectors at the first moment, and further reflects the deviation between the user feedback action and the preset training action. A larger calculated Euclidean distance means that the deviation between the user feedback action and the preset training action is larger.

[0080] The calculated first action deviation parameter is compared with the preset deviation parameter threshold. The deviation parameter threshold is a critical value set according to the specific application scenario and requirements. If the deviation parameter is greater than or equal to the threshold, it means that there is a significant difference between the user's action and the standard action. The first moment feedback action and the first moment training action are marked as deviation actions and added to the deviation action list. These deviation actions remind users that they need to adjust their actions to meet the expected requirements. This mechanism can provide accurate feedback to help users gradually correct their actions and achieve standardized training goals.

[0081] Furthermore, the execution step of the loop decision module 24 in the embodiment of the present application also includes: when the number of the deviation actions is equal to 0, returning to the training action configuration module 21 to update the training action information to execute the loop.

[0082] Specifically, when the number of deviation actions is equal to 0, it means that all actions performed by the user meet the standard requirements, that is, the accuracy of the user's actions reaches the expected goal. The result is returned to the training action configuration module 21, and the training action configuration module 21 will adjust the next training action according to the user's performance and progress, and match the new training action information, such as proceeding to the next training stage, increasing the complexity or intensity of the training, etc. The subsequent training process is carried out according to the new training action information, thus forming a cycle. This loop mechanism can continuously optimize the training process while maintaining the user's participation and training effect.

[0083] In summary, the VR management system for preventing venous thromboembolism provided by the embodiment of the present application has the following technical effects:

[0084] The VR glasses 10 provide users with an immersive training environment, which enhances the interactivity and participation of training. The training action configuration module 21 adjusts the training content according to the basic information of the user and tailors the training action for each user. After configuring the training action according to the basic information of the user, the virtual scene construction module 22 will generate a virtual training scene based on this information and transmit the corresponding training task to the VR glasses 10; at the same time, the module obtains the user's feedback action in real time to ensure the authenticity and interactivity of the training and provide data support for subsequent action comparison. The user action comparison model 23 is responsible for comparing the difference between the feedback action performed by the user in the virtual scene and the standard training action. By extracting action features and calculating the Euclidean distance, the action deviation is quantified, feedback is provided, and the user is helped to adjust the action. The combination of the virtual scene construction module 22 and the action comparison model can monitor the user's action in real time, and timely identify the action deviation through feature comparison, provide accurate feedback, help the user gradually correct the action, and improve the accuracy and effectiveness of the training. When the user's action deviates from the standard action, the loop decision module 24 will feed back the deviation action to the virtual scene to remind the user to make adjustments. If the number of deviations is zero, return and update the training task. This mechanism ensures the accuracy of the user's movements, while constantly adjusting the training content to avoid the monotony of training.

[0085] Overall, the embodiments of the present application construct a rehabilitation training system that combines fun, personalization, and efficiency through precise configuration, virtual scene construction, real-time feedback, and loop optimization. It overcomes the shortcomings of traditional prevention methods such as low compliance and delayed feedback, significantly improves the readability and adaptability of training content for preventing venous thromboembolism with VR technology, and enhances the user experience.

[0086] Embodiment 2, as Figure 4 As shown, the embodiment of the present application provides a VR management method for preventing venous thromboembolism, the method comprising:

[0087] Step S1: According to the basic prevention plan for VTE, the training action information is matched in combination with the basic information of the user.

[0088] Step S2: Building a virtual training scene according to the training action information, and sending the training task to the VR glasses to obtain user feedback actions.

[0089] Step S3: Compare the user feedback action with the training action information to obtain the deviation action.

[0090] Step S4: when the number of the deviation actions is not equal to 0, the deviation actions are returned to the virtual scene construction module to execute a loop.

[0091] Furthermore, the basic user information includes user age, user gender, user weight and user height, and step S1 further includes:

[0092] Obtain a first motion video of the basic VTE prevention program; perform semantic segmentation on the first motion video to obtain a first action sequence; input the first action sequence, the user age, the user gender, the user weight and the user height into an action matching model to obtain the training action information.

[0093] Furthermore, semantic segmentation is performed on the first motion video to obtain a first action sequence, including:

[0094] Collect motion video recording data and joint position identification data; use the joint position identification data as supervision and the motion video recording data as input to train a convolutional neural network to generate a semantic segmentation model; perform semantic segmentation on the first motion video according to the semantic segmentation model to obtain the first action sequence.

[0095] Furthermore, the first action sequence, the user age, the user gender, the user weight and the user height are input into an action matching model to obtain the training action information, including:

[0096] Configure an initial action sequence; obtain a first user motion video of the initial action sequence, and obtain a first identification action sequence by performing semantic segmentation on the first user motion video; until the Mth user motion video of the initial action sequence is obtained, and obtain the Mth identification action sequence by performing semantic segmentation on the Mth user motion video; add the first user basic information to the Mth user basic information to the user basic information record data set, and add the first identification action sequence to the Mth identification action sequence to the matching identification action data set; use the initial action sequence and the user basic information record data set as input, and the matching identification action data set as supervision, to train a convolutional neural network to generate the action matching model.

[0097] Further, step S3 includes:

[0098] An action comparison identification model is obtained, wherein the action comparison identification model has a first feature extraction channel, a second feature extraction channel and a deviation action identification layer, and the first feature extraction channel and the second feature extraction channel are twin networks; the user feedback action is input into the first feature extraction channel to obtain a first action feature vector; the training action information is input into the second feature extraction channel to obtain a second action feature vector; the first action feature vector and the second action feature vector are compared through the deviation action identification layer to obtain the deviation action.

[0099] Further, comparing the first action feature vector and the second action feature vector through the deviation action identification layer to obtain the deviation action includes:

[0100] According to the first action feature vector, extract the first action feature vector at the first moment; according to the second action feature vector, extract the second action feature vector at the first moment; calculate the Euclidean distance between the first action feature vector at the first moment and the second action feature vector at the first moment to obtain a first action deviation parameter; when the first action deviation parameter is greater than or equal to the deviation parameter threshold, add the first moment feedback action and the first moment training action into the deviation action.

[0101] Further, when the number of the deviation actions is equal to 0, the process returns to the training action configuration module to update the training action information and execute the loop.

[0102] Embodiment three, based on the same inventive concept of a VR management system for preventing venous thromboembolism in the aforementioned embodiment one, the present application also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the execution steps of the system described in any one of the aforementioned embodiments one.

[0103] like Figure 5 As shown, the bus architecture is represented by bus 300, which may include any number of interconnected buses and bridges, and bus 300 connects various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also connect various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. Bus interface 305 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same element, namely a transceiver, providing a unit for communicating with various other devices on a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 may be used to store data used by processor 302 when performing operations.

[0104] Embodiment 4, based on the same inventive concept of a VR management system for preventing venous thromboembolism in the aforementioned embodiment 1, the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the execution steps of the aforementioned VR management system for preventing venous thromboembolism and can achieve the same technical effects. To avoid repetition, it will not be described here.

[0105] Although the present application has been described in conjunction with specific features and embodiments thereof, it is obvious that various modifications and combinations may be made thereto without departing from the spirit and scope of the present application. Accordingly, this specification and the accompanying drawings are merely exemplary illustrations of the present application as defined therein, and are deemed to have covered any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A VR management system for preventing venous thromboembolism, characterized in that: include: VR glasses; Central control unit, including: A training action configuration module is used to match training action information according to the basic prevention plan for VTE prevention and combined with basic user information; A virtual scene building module, used to build a virtual training scene according to the training action information, and send the training task to the VR glasses to obtain user feedback actions; A user action comparison model, used to compare the user feedback action with the training action information to obtain a deviation action; The loop decision module is used to return the deviation action to the virtual scene construction module to execute the loop when the number of the deviation actions is not equal to 0.

2. The system according to claim 1, characterized in that The training action configuration module is used to match the training action information according to the basic prevention scheme for preventing VTE in combination with the basic information of the user, including: The basic user information includes user age, user gender, user weight and user height; Obtaining the first exercise video of the basic prevention program for preventing VTE; Performing semantic segmentation on the first motion video to obtain a first action sequence; The first action sequence, the user age, the user gender, the user weight and the user height are input into an action matching model to obtain the training action information.

3. The system according to claim 2, characterized in that Performing semantic segmentation on the first motion video to obtain a first action sequence includes: Collect motion video recording data and joint position identification data; Using the joint position identification data as supervision and the motion video recording data as input to train a convolutional neural network to generate a semantic segmentation model; The first motion video is semantically segmented according to the semantic segmentation model to obtain the first action sequence.

4. The system according to claim 2, characterized in that Inputting the first action sequence, the user age, the user gender, the user weight and the user height into an action matching model to obtain the training action information includes: Configure the initial action sequence; Obtaining a first user motion video of the initial motion sequence, and obtaining a first identified motion sequence by performing semantic segmentation on the first user motion video; Until an Mth user motion video of the initial action sequence is obtained, an Mth identified action sequence is obtained by performing semantic segmentation on the Mth user motion video; Add the first user basic information to the Mth user basic information into the user basic information record data set, and add the first identification action sequence to the Mth identification action sequence into the matching identification action data set; The initial action sequence and the user basic information record data set are used as input, and the matching identification action data set is used as supervision to train a convolutional neural network to generate the action matching model.

5. The system according to claim 1, wherein: The user action comparison model is used to compare the user feedback action with the training action information to obtain the deviation action, including: Obtaining an action comparison identification model, wherein the action comparison identification model has a first feature extraction channel, a second feature extraction channel, and a deviation action identification layer, and the first feature extraction channel and the second feature extraction channel are twin networks; Inputting the user feedback action into the first feature extraction channel to obtain a first action feature vector; Inputting the training action information into the second feature extraction channel to obtain a second action feature vector; The first action feature vector and the second action feature vector are compared through the deviation action identification layer to obtain the deviation action.

6. The system according to claim 5, characterized in that Comparing the first action feature vector and the second action feature vector through the deviation action identification layer to obtain the deviation action includes: Extracting a first action feature vector at a first moment according to the first action feature vector; extracting a second action feature vector at a first moment according to the second action feature vector; Calculating the Euclidean distance between the first action feature vector at the first moment and the second action feature vector at the first moment to obtain a first action deviation parameter; When the first action deviation parameter is greater than or equal to a deviation parameter threshold, the first moment feedback action and the first moment training action are added to the deviation action.

7. The system according to claim 1, characterized in that The loop decision module execution step also includes: when the number of the deviation actions is equal to 0, returning to the training action configuration module to update the training action information to execute the loop.

8. A VR management method for preventing venous thromboembolism, characterized in that: The method is performed by a VR management system for preventing venous thromboembolism according to any one of claims 1 to 7, comprising: According to the basic prevention plan for VTE, the training action information is matched with the user's basic information; Building a virtual training scene according to the training action information, and sending the training task to the VR glasses to obtain user feedback actions; Comparing the user feedback action with the training action information to obtain the deviation action; When the number of the deviation actions is not equal to 0, the deviation actions are returned to the virtual scene construction module to execute a loop.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the execution steps of a VR management system for preventing venous thromboembolism according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the execution steps of a VR management system for preventing venous thromboembolism as described in any one of claims 1 to 7 are implemented.

Images