A rehabilitation exercise evaluation method and device, an electronic device and a storage medium
By constructing a motion assessment model based on an adjacency matrix and attention mechanism of knowledge experts, the problem of accuracy in assessing rehabilitation movements for patients without expert guidance is solved, providing effective rehabilitation training guidance and ensuring rehabilitation effectiveness and safety.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BUFFALO ROBOT TECH
- Filing Date
- 2023-02-23
- Publication Date
- 2026-07-14
AI Technical Summary
When patients undergo rehabilitation training without expert guidance, they cannot accurately assess the correctness of the rehabilitation movements, leading to poor training results or potential injury.
A motion assessment model based on an adjacency matrix and attention mechanism constructed by knowledge experts is adopted. The model evaluates rehabilitation motions by generating time series of coordinates of the skeletal points to be tested, and combines classification training to generate guidance feedback information.
It enables accurate assessment of rehabilitation exercises for different patients, provides effective guidance and feedback, and ensures the correctness and safety of rehabilitation training.
Smart Images
Figure CN116071827B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent learning technology, and more specifically, to a rehabilitation exercise assessment method, device, electronic device, and storage medium. Background Technology
[0002] Rehabilitation training refers to activities that patients engage in to improve or restore their physical function. Professional therapists will prescribe exercise prescriptions for patients, instructing them to perform specific exercises to recover from different conditions (such as stroke, Parkinson's disease, back pain, etc.).
[0003] When patients undergo rehabilitation training without the assistance of experts (such as doctors / therapists), the correctness of their rehabilitation movements cannot be assessed, thus hindering their ability to receive adequate guidance and evaluation of prescribed exercises. Furthermore, incorrect or non-standard rehabilitation exercises often result in lower therapeutic outcomes and may even lead to more serious injuries. Therefore, there is an urgent need for automated rehabilitation movement assessment methods. Summary of the Invention
[0004] The purpose of this invention is to provide a rehabilitation exercise assessment method, device, electronic device, and storage medium that can adapt to the differences between different patients and accurately assess rehabilitation exercises.
[0005] To achieve the above objectives, the technical solutions adopted in the embodiments of the present invention are as follows:
[0006] In a first aspect, embodiments of the present invention provide a method for assessing rehabilitation exercises, characterized in that the method includes:
[0007] Generate a time series of skeletal point coordinates based on the video of the rehabilitation movements to be tested;
[0008] The time series of the coordinates of the bone points to be tested is input into a pre-trained motion assessment model to obtain the patient's rehabilitation motion score; wherein, the motion assessment model is generated based on the adjacency matrix constructed by knowledge experts, attention mechanism, classification training and assessment training;
[0009] Guidance feedback information is generated based on the scores of the rehabilitation movements.
[0010] Furthermore, the action evaluation model, based on an adjacency matrix constructed by knowledge experts and an attention mechanism, generates the following:
[0011] The adjacency matrix constructed by the knowledge expert is combined with the adjacency matrix of the initial action evaluation model to obtain the first intermediate action evaluation model;
[0012] The first intermediate action evaluation model is fused with the attention mechanism to obtain the second intermediate action evaluation model;
[0013] The second intermediate action evaluation model is trained by classification to obtain the third intermediate action evaluation model;
[0014] The third intermediate movement assessment model is evaluated, trained, and validated using a rehabilitation exercise assessment dataset to obtain the movement assessment model; wherein, the rehabilitation exercise assessment dataset consists of scores given by knowledge experts to multiple training rehabilitation exercise videos.
[0015] Furthermore, at time t, the first intermediate action evaluation model... The layer update formula is as follows:
[0016] ;
[0017] in, This represents the number of adjacency matrices constructed by the knowledge experts. It is a non-linear activation function; This represents the normalization of the adjacency matrix constructed by the knowledge expert and the adjacency matrix of the initial action evaluation model; A represents the adjacency matrix of the initial action evaluation model; k The adjacency matrix represents the knowledge experts' constructs. This represents time t. Layer update formula; for The trainable parameters of the layer.
[0018] Furthermore, the formula for the second intermediate action evaluation model is as follows:
[0019] ;
[0020] in, a represents the number of adjacency matrices constructed by the knowledge experts; k Represents weight; This represents the output characteristics of the first intermediate action evaluation model.
[0021] Furthermore, the step of classifying and training the second intermediate action evaluation model to obtain the third intermediate action evaluation model includes:
[0022] The parameters of the second intermediate action evaluation model are updated based on the first loss function to obtain the third intermediate action evaluation model; wherein, the formula for the first loss function is as follows:
[0023] ;
[0024] in, This is a key focus for experts; a represents the number of adjacency matrices constructed by the knowledge experts;k Represents weight.
[0025] Furthermore, the method also includes:
[0026] The time series of the bone point coordinates to be tested is filtered to obtain the filtered time series of the bone point coordinates to be tested.
[0027] Furthermore, the method also includes:
[0028] The filtered time series of the coordinates of the bone points to be tested is sampled at equal intervals.
[0029] Secondly, embodiments of the present invention also provide a rehabilitation exercise assessment device, the device comprising:
[0030] The time series generation module is used to generate a time series of skeletal point coordinates based on the video of the rehabilitation movement to be tested.
[0031] The scoring module is used to input the time series of the coordinates of the bone points to be tested into a pre-trained motion assessment model to obtain the patient's rehabilitation motion score; wherein, the motion assessment model is generated based on the adjacency matrix constructed by knowledge experts, attention mechanism, classification training and assessment training;
[0032] The guidance feedback generation module is used to generate guidance feedback information based on the rehabilitation action scores.
[0033] Thirdly, embodiments of the present invention also provide an electronic device, including a memory and one or more processors, wherein the memory is used to store one or more programs; when the one or more programs are executed by the one or more processors, they implement the method as described in the first aspect.
[0034] Fourthly, embodiments of the present invention also provide a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method as described in the first aspect.
[0035] The present invention provides a rehabilitation exercise assessment method, device, electronic device, and storage medium. The method includes: generating a time series of skeletal point coordinates based on a video of the rehabilitation exercise to be tested; inputting the time series of skeletal point coordinates into a pre-trained exercise assessment model to obtain the patient's rehabilitation exercise score; wherein the exercise assessment model is generated based on an adjacency matrix constructed by knowledge experts, an attention mechanism, classification training, and assessment training; and generating guidance feedback information based on the rehabilitation exercise score. By artificially setting the skeletal point adjacency matrix according to different rehabilitation exercise requirements through expert knowledge, the model can better extract the spatiotemporal features of rehabilitation exercises; the introduction of an attention mechanism can efficiently integrate spatiotemporal features; this method can adapt to the differences between different patients, accurately assess rehabilitation exercises, and solves the problem that patients cannot obtain sufficient guidance and assessment for prescribed exercises when practicing without expert assistance.
[0036] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description
[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0038] Figure 1 A schematic diagram of the structure of an electronic device provided in an embodiment of the present invention is shown;
[0039] Figure 2 A flowchart illustrating a rehabilitation exercise assessment method provided by an embodiment of the present invention is shown;
[0040] Figure 3 A schematic diagram illustrating the training of an action evaluation model provided in an embodiment of the present invention is shown;
[0041] Figure 4 A flowchart illustrating another rehabilitation exercise assessment method provided by an embodiment of the present invention is shown;
[0042] Figure 5 A structural block diagram of a rehabilitation exercise assessment device provided in an embodiment of the present invention is shown.
[0043] Reference numerals: 100-Electronic device; 101-Memory; 102-Processor; 103-Communication interface; 300-Rehabilitation exercise assessment device; 301-Time series generation module; 302-Scoring module; 303-Guidance feedback generation module; 304-Model training module. Detailed Implementation
[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
[0045] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0046] Please refer to Figure 1 This is a schematic diagram of the structure of an electronic device 100 provided in an embodiment of this application. The electronic device 100 includes a memory 101, a processor 102, and a communication interface 103. The memory 101, processor 102, and communication interface 103 are electrically connected to each other directly or indirectly to realize data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines. The memory 101 can be used to store software programs and modules, such as the program instructions / modules corresponding to the rehabilitation exercise assessment method provided in the embodiment of this application. The processor 102 executes various functional applications and data processing by executing the software programs and modules stored in the memory 101. The communication interface 103 can be used to communicate with the node device 300 and the client 200 for signaling or data. In this application, the electronic device 100 may have multiple communication interfaces 103.
[0047] The memory 101 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.
[0048] Processor 102 can be an integrated circuit chip with signal processing capabilities. This processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
[0049] In this embodiment, a computer storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the rehabilitation exercise assessment method.
[0050] Below Figure 1 Based on the illustrated electronic device 100, this application provides a rehabilitation exercise assessment method. Please refer to [link to relevant documentation]. Figure 2 , Figure 2 This is a flowchart illustrating a rehabilitation exercise assessment method provided in an embodiment of this application. The rehabilitation exercise assessment method may include the following steps:
[0051] S201, Generate a time series of skeletal point coordinates based on the video of the rehabilitation movement to be tested.
[0052] The principle behind acquiring rehabilitation exercise videos is as follows: The software development kit (Azeure Kinect for Windows SDK) related to the image acquisition device is configured on electronic device 100, including the driver and interface program for the image acquisition device. The hardware needs to be connected to and fixed in place with the Azure Kinect (image acquisition device), ensuring the user is 0.5-3 meters away from the device; videos of the user performing rehabilitation exercises are then acquired via Azure Kinect.
[0053] The rehabilitation movement video to be tested can be a real-time acquired RGB-D video. Azure Kinect DK is used to convert the rehabilitation movement video into a time series of skeletal point coordinates.
[0054] S202, input the time series of the coordinates of the bone points to be tested into the pre-trained motion assessment model to obtain the patient's rehabilitation motion score.
[0055] The action evaluation model is generated based on the adjacency matrix constructed by knowledge experts, the attention mechanism, classification training, and evaluation training.
[0056] like Figure 3 As shown, the training principle of the motion assessment model is as follows: the adjacency matrix constructed by the knowledge expert is combined with the adjacency matrix of the initial motion assessment model to obtain the first intermediate motion assessment model; the first intermediate motion assessment model is fused with the attention mechanism to obtain the second intermediate motion assessment model; the second intermediate motion assessment model is trained by classification to obtain the third intermediate motion assessment model; the third intermediate motion assessment model is evaluated, trained and validated using the rehabilitation exercise assessment dataset to obtain the motion assessment model; wherein, the rehabilitation exercise assessment dataset consists of the scores given by the knowledge expert to multiple training rehabilitation exercise videos.
[0057] In this embodiment, during the training of the motion assessment model, multiple training rehabilitation motion videos are acquired using an image acquisition device, and the training rehabilitation motion videos are used to generate a time series of training skeletal point coordinates. This time series of training skeletal point coordinates is then input into the initial motion assessment model.
[0058] The initial action evaluation model can be an ST-GCN (Spatiotemporal Graph Convolutional Network) model. In the initial action evaluation model, the time series of training skeleton point coordinates is the input X, and the adjacency matrix constructed by the knowledge expert is A. Then, in time... time, The layer update formula is as follows:
[0059] ;
[0060] in, It is a non-linear activation function; This represents the normalization of the adjacency matrix constructed by knowledge experts; The adjacency matrix represents the initial action evaluation model; =A+I, representing the sum of the adjacency matrix and its own links in the initial action evaluation model; This represents time t. Layer update formula; for The trainable parameters of the layer.
[0061] To better extract the spatiotemporal characteristics of rehabilitation exercises, a bone point adjacency matrix A was artificially set based on expert knowledge and the requirements of different rehabilitation movements. k The number of k is determined by the expert knowledge and the rehabilitation exercise assessment dataset itself. For example, if it is necessary to obtain the skeletal point features corresponding to the upper limbs, trunk, and lower limbs, then the adjacency matrix of the upper limb skeletal points, the adjacency matrix of the trunk skeletal points, and the adjacency matrix of the lower limb skeletal points are constructed using expert knowledge, in which case the number of k is 3. Combining the adjacency matrix of the initial movement assessment model with the adjacency matrices of the upper limb skeletal points, the trunk skeletal points, and the lower limb skeletal points, the first intermediate movement assessment model can be obtained. The three output features of the first intermediate movement assessment model correspond to the skeletal point features of the upper limbs, trunk, and lower limbs, respectively; that is, the k output features of the first intermediate movement assessment model correspond to the skeletal point features of different rehabilitation movements.
[0062] The first intermediate action evaluation model at time t The layer update formula is as follows:
[0063] ;
[0064] in, The number of adjacency matrices constructed by knowledge experts; It is a non-linear activation function; This represents the normalization of the adjacency matrix constructed by the knowledge experts and the adjacency matrix of the initial action evaluation model; The adjacency matrix representing the initial action evaluation model; A k An adjacency matrix representing knowledge experts; This represents time t. Layer update formula; for The trainable parameters of the layer.
[0065] Since the output of the first intermediate movement assessment model consists of k output features, and rehabilitation exercise assessment is performed item by item, feature fusion is necessary. In this embodiment, the first intermediate movement assessment model is fused with the attention mechanism to obtain a second intermediate movement assessment model, the output of which is a single output feature. The second intermediate movement assessment model can be the ST-AGCN model.
[0066] In this embodiment, the attention mechanism can employ a gated attention mechanism, fusing the k output features from the first intermediate action evaluation model. The formula for the second intermediate action evaluation model is as follows:
[0067] ;
[0068] in, The number of adjacency matrices constructed by knowledge experts; a k This represents the weighting, i.e., the exercise details that patients should pay attention to during rehabilitation exercises; This represents the output feature of the first intermediate action evaluation model.
[0069] a k It is calculated using the following formula:
[0070] ;
[0071] in, , , These are the model parameters that need to be trained. , These are nonlinear activation functions.
[0072] To emphasize vectors The representational features, directly using vectors The rehabilitation exercises are categorized by focus, and the parameters of the second intermediate movement assessment model are updated based on the first loss function to obtain the third intermediate movement assessment model. The third intermediate movement assessment model can be a P-STAGGN model.
[0073] In other words, when a patient is performing rehabilitation exercises, the model does not know what movements the patient is making. Therefore, classification learning is used to enable the model to know what movements the patient is making, and the weights corresponding to the patient's movements are adjusted to the maximum. The classification results of the second intermediate model are compared with the focus of experts on the movement, and the parameters of the second intermediate model are updated to improve the convergence speed of the model.
[0074] The first loss function can be the cross-entropy loss function, whose formula is as follows:
[0075] ;
[0076] in, The first loss function; This is a key focus for experts; a represents the number of adjacency matrices constructed by the knowledge experts; k Represents weight.
[0077] After classification training, the third intermediate movement assessment model is evaluated, trained and validated using a rehabilitation exercise assessment dataset to obtain the movement assessment model.
[0078] In this embodiment, the third intermediate motion evaluation model is similar to the motion evaluation model, but a multilayer perceptron is added on the basis of the third intermediate motion evaluation model to form the final motion evaluation model.
[0079] The final motion evaluation model formula is as follows:
[0080] ;
[0081] in, Representing the The trainable parameters of the layer, Represents input in matrix form. represent The layer bias, where σ represents the ReLU activation function. This represents the output in vector form.
[0082] The final action evaluation model can also use the cross-entropy loss function, defined as Loss, as shown in the following formula:
[0083] .
[0084] During the training process, the rehabilitation exercise assessment dataset is divided into a training set and a validation set. The training set is used to train the third intermediate movement assessment model, and the validation set is used to validate the third intermediate movement assessment model. During the validation process, the model with the smallest validation loss function value is selected as the final movement assessment model.
[0085] The principle of acquiring the rehabilitation exercise assessment dataset is as follows: knowledge experts watch and analyze the training rehabilitation exercise videos, and score the training rehabilitation exercise videos on a scale of 0 to 5, where 0 points represent completely non-standard movements and 5 points represent very standard movements; based on the scores of the training rehabilitation exercise videos, the corresponding training skeletal point coordinate time series are labeled, thereby constructing the rehabilitation exercise assessment dataset.
[0086] In this embodiment, the output of the motion assessment model can be not only a score for rehabilitation motions, but also a rating level, which can be set to a rating level of 0-5.
[0087] S203, Generate guidance feedback information based on the rehabilitation action score.
[0088] like Figure 4 As shown, in order to reduce noise and jitter generated when the image acquisition device acquires rehabilitation movement videos, the rehabilitation movement assessment method in this embodiment further includes the following steps before inputting the time series coordinates of the bone points to be tested into the movement assessment model:
[0089] S204, filter the time series of the skeletal point coordinates to be tested to obtain the filtered time series of the skeletal point coordinates to be tested.
[0090] In this embodiment, Kalman filtering can be used to filter the time series of the coordinates of the skeleton points under test.
[0091] The working principle of Kalman filtering is as follows: For the key points that need to be processed, define the system state vector X and the observation vector Z of the system state of the Kalman filter at time t:
[0092] ;
[0093] ;
[0094] in, and The three-dimensional coordinates of the joints to be processed, and The velocity is the center velocity of the joint.
[0095] The system equations and observation equations for Kalman filtering are as follows:
[0096] ;
[0097] Where A is the state matrix, C is the driving matrix, and H represents the transformation matrix from state transition to measurement. For process noise, To observe noise.
[0098] By substituting the system's state value at time t into the equation model, the state values at times t+1 and t-1 can be predicted. Averaging the state values at times t, t+1, and t-1 improves the accuracy of the joint coordinates required for measurement.
[0099] To reduce model complexity and computational cost, please refer to the following before inputting the time series coordinates of the skeletal points to the motion evaluation model: Figure 4 The rehabilitation exercise assessment method in this embodiment also includes the following steps:
[0100] S205, perform equal-interval sampling processing on the filtered time series of the coordinates of the bone points to be tested.
[0101] In this embodiment, the equal interval length can be set to 3 seconds. The time length of the equal interval varies for different rehabilitation actions and can be set according to the type and requirements of the rehabilitation action.
[0102] To implement the rehabilitation exercise assessment method corresponding to S201-S205 and its possible sub-steps, this application provides a rehabilitation exercise assessment device. Please refer to [link to relevant documentation]. Figure 5 , Figure 5 This is a block diagram of a rehabilitation exercise assessment device provided in an embodiment of this application. The rehabilitation exercise assessment device 300 includes: a time series generation module 301, a scoring module 302, a guidance feedback generation module 303, and a model training module 304.
[0103] The time series generation module 301 is used to generate a time series of skeletal point coordinates based on the video of the rehabilitation action to be tested.
[0104] The scoring module 302 is used to input the time series of the coordinates of the bone points to be tested into a pre-trained motion assessment model to obtain the patient's rehabilitation motion score.
[0105] The guidance feedback generation module 303 is used to generate guidance feedback information based on the rehabilitation movement score.
[0106] Model training module 304 is used to generate action evaluation models based on adjacency matrices constructed by knowledge experts, attention mechanisms, classification training, and evaluation training.
[0107] It should be understood that the time series generation module 301, the scoring module 302, the guidance feedback generation module 303, and the model training module 304 can work together to implement the above S201 to S205 and their possible sub-steps.
[0108] In summary, this application provides a rehabilitation exercise assessment method, device, electronic device, and storage medium. The method includes: generating a time series of skeletal point coordinates based on a video of the rehabilitation exercise to be tested; inputting the time series of skeletal point coordinates into a pre-trained exercise assessment model to obtain the patient's rehabilitation exercise score; wherein the exercise assessment model is generated based on an adjacency matrix constructed by a knowledge expert, an attention mechanism, classification training, and assessment training; and generating guidance feedback information based on the rehabilitation exercise score. By manually setting the skeletal point adjacency matrix according to different rehabilitation exercise requirements through expert knowledge, the model can better extract the spatiotemporal features of rehabilitation exercises; the introduction of an attention mechanism can efficiently integrate spatiotemporal features; this method can adapt to the differences between different patients, accurately assess rehabilitation exercises, and solve the problem that patients cannot obtain sufficient guidance and assessment of prescribed exercises when practicing without expert assistance.
[0109] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0110] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of protection of the invention. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.
Claims
1. A method for assessing rehabilitation exercises, characterized in that, The method includes: Generate a time series of skeletal point coordinates based on the video of the rehabilitation movements to be tested; The time series of the coordinates of the bone points to be tested is input into a pre-trained motion assessment model to obtain the patient's rehabilitation motion score; Guidance feedback information is generated based on the rehabilitation exercise scores; The training steps of the action evaluation model include: The adjacency matrix constructed by knowledge experts is combined with the adjacency matrix of the initial action evaluation model to obtain the first intermediate action evaluation model; The first intermediate action evaluation model is fused with the attention mechanism to obtain the second intermediate action evaluation model; The second intermediate action evaluation model is trained by classification to obtain the third intermediate action evaluation model; The third intermediate movement assessment model is evaluated, trained, and validated using a rehabilitation exercise assessment dataset to obtain the movement assessment model; wherein, the rehabilitation exercise assessment dataset consists of scores given by knowledge experts to multiple training rehabilitation exercise videos.
2. The rehabilitation exercise assessment method as described in claim 1, characterized in that, The first intermediate action evaluation model at time t, The layer update formula is as follows: ； in, This represents the number of adjacency matrices constructed by the knowledge experts. It is a non-linear activation function; This represents the normalization of the adjacency matrix constructed by the knowledge expert and the adjacency matrix of the initial action evaluation model; A represents the adjacency matrix of the initial action evaluation model; k The adjacency matrix represents the knowledge experts' constructs. This represents time t. Layer update formula; for The trainable parameters of the layer.
3. The rehabilitation exercise assessment method as described in claim 1, characterized in that, The formula for the second intermediate action evaluation model is as follows: ； in, a represents the number of adjacency matrices constructed by the knowledge experts; k Represents weight; This represents the output characteristics of the first intermediate action evaluation model.
4. The rehabilitation exercise assessment method as described in claim 1, characterized in that, The process of classifying and training the second intermediate action evaluation model to obtain the third intermediate action evaluation model includes: The parameters of the second intermediate action evaluation model are updated based on the first loss function to obtain the third intermediate action evaluation model; wherein, the formula for the first loss function is as follows: ； in, This is a key focus for experts; a represents the number of adjacency matrices constructed by the knowledge experts; k Represents weight.
5. The rehabilitation exercise assessment method as described in claim 1, characterized in that, The method further includes: The time series of the bone point coordinates to be tested is filtered to obtain the filtered time series of the bone point coordinates to be tested.
6. The rehabilitation exercise assessment method as described in claim 5, characterized in that, The method further includes: The filtered time series of the coordinates of the bone points to be tested is sampled at equal intervals.
7. A rehabilitation exercise assessment device, characterized in that, The device includes: The time series generation module is used to generate a time series of skeletal point coordinates based on the video of the rehabilitation movement to be tested. The scoring module is used to input the time series of the coordinates of the bone points to be tested into a pre-trained motion assessment model to obtain the patient's rehabilitation motion score. The training content of the motion assessment model includes: combining the adjacency matrix constructed by knowledge experts with the adjacency matrix of the initial motion assessment model to obtain a first intermediate motion assessment model; fusing the first intermediate motion assessment model with an attention mechanism to obtain a second intermediate motion assessment model; performing classification training on the second intermediate motion assessment model to obtain a third intermediate motion assessment model; and evaluating, training, and validating the third intermediate motion assessment model using a rehabilitation motion assessment dataset to obtain the final motion assessment model. The rehabilitation motion assessment dataset consists of scores from knowledge experts on multiple training rehabilitation motion videos. The guidance feedback generation module is used to generate guidance feedback information based on the rehabilitation action scores.
8. An electronic device, characterized in that, The method includes a memory and one or more processors, the memory being used to store one or more programs; when the one or more programs are executed by the one or more processors, they implement the method as described in any one of claims 1 to 6.
9. A computer storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 6.