Practical training evaluation method and device, computer equipment and storage medium

Abstract

The invention relates to a practical training evaluation method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring an attitude sequence of an operator in a current practical training process; wherein the attitude sequence comprises a preset number of three-dimensional attitude models of continuous time frames; determining a shape difference sequence of the attitude sequence according to the attitude sequence and the reference attitude of the attitude sequence; wherein the shape difference sequence comprises a shape difference operator corresponding to each time frame, and the shape difference operator corresponding to each time frame represents the shape difference between the three-dimensional attitude model of the time frame and the reference attitude; determining a target action category of the attitude sequence according to the shape difference sequence; according to the target action category and the action category mapping relation, evaluating the practical training process of the operator; wherein the action category mapping relation comprises action descriptions associated with different action categories. By adopting the method, real-time and objective practical training examination and evaluation can be realized, and the training quality and the operation safety are improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a training and evaluation method, apparatus, computer equipment, and storage medium. Background Technology

[0002] With the deep integration of information technology and industrial manufacturing, high-end equipment fields such as electric locomotives, diesel locomotives, aerospace, and precision manufacturing place extremely high demands on the skill levels and operational standards of frontline operators. Traditional training methods, such as apprenticeship or physical equipment disassembly and assembly, suffer from high costs, high risks, severe equipment damage, long training cycles, and difficulty in quantifying assessment standards. Especially in high-risk operations such as locomotive maintenance, even minor operational errors can lead to serious safety accidents or economic losses.

[0003] To address this, virtual simulation training systems based on virtual reality and mixed reality have emerged, providing zero-risk, low-cost, immersive, and repeatable training by constructing realistic virtual environments. However, existing systems mostly focus on "visual" presentation, making it difficult to effectively assess the standardization of trainees' movements. Summary of the Invention

[0004] Therefore, it is necessary to provide a training assessment method, device, computer equipment, and storage medium to address the aforementioned technical problems, enabling real-time and objective training assessment and evaluation, thereby improving training quality and operational safety.

[0005] Firstly, this application provides a practical training evaluation method, including:

[0006] Obtain the posture sequence of the operator during the current training process; wherein, the posture sequence includes a three-dimensional posture model of a preset number of consecutive time frames;

[0007] Based on the attitude sequence and the reference attitude of the attitude sequence, a shape difference sequence of the attitude sequence is determined; wherein, the shape difference sequence includes a shape difference operator corresponding to each time frame, and the shape difference operator corresponding to each time frame represents the shape difference between the three-dimensional attitude model of the time frame and the reference attitude;

[0008] Based on the shape difference sequence, determine the target action category of the posture sequence;

[0009] The training process of the operator is evaluated based on the target action category and the action category mapping relationship; wherein, the action category mapping relationship includes action descriptions associated with different action categories.

[0010] In one embodiment, the three-dimensional pose model of each time frame includes multiple nodes, connecting edges between nodes, and target position information of each node; the nodes represent acquisition points located on the operator, and the connecting edges represent the connection relationship between the acquisition points;

[0011] The nodes and connecting edges between nodes in the 3D pose model of different time frames are the same, but the target position information of each node is different.

[0012] The reference pose includes multiple nodes, the connecting edges between nodes, and the reference position information of each node.

[0013] In one embodiment, determining the shape difference sequence of the attitude sequence based on the attitude sequence and a reference attitude of the attitude sequence includes:

[0014] Determine the reference pose of the pose sequence;

[0015] For each time frame, the shape difference operator corresponding to the time frame is determined based on the target position information and reference position information of each node;

[0016] The shape difference sequence of the attitude sequence is constructed based on the shape difference operator corresponding to each time frame.

[0017] In one embodiment, determining the reference pose of the pose sequence includes:

[0018] The three-dimensional pose model corresponding to the first time frame in the pose sequence is used as the reference pose of the pose sequence.

[0019] In one embodiment, determining the shape difference operator corresponding to the time frame based on the target position information and reference position information of each node includes:

[0020] For each node, determine the moving distance and direction between the target location information and the reference location information; and,

[0021] Based on the moving distance and moving direction, determine the displacement vector of the node;

[0022] Based on the displacement vector of each node, the shape difference operator corresponding to the time frame is determined.

[0023] In one embodiment, determining the target action category of the pose sequence based on the shape difference sequence includes:

[0024] Spatial features are extracted from the shape difference sequence to obtain a spatial feature sequence; wherein, the spatial feature sequence includes a spatial feature vector corresponding to each time frame;

[0025] The temporal dependencies between the spatial feature vectors are analyzed to obtain the spatiotemporal features of the attitude sequence;

[0026] Based on the spatiotemporal characteristics, the target action category of the posture sequence is determined.

[0027] In one embodiment, evaluating the operator's training process based on the target action category and the action category mapping relationship includes:

[0028] From the action category mapping relationship, find the target action description associated with the target action category;

[0029] The training process of the operator is evaluated based on the target action description and the standard action description required by the training task.

[0030] Secondly, this application also provides a practical training and evaluation device, including:

[0031] The sequence acquisition module is used to acquire the posture sequence of the operator in the current training process; wherein, the posture sequence includes a three-dimensional posture model of a preset number of consecutive time frames.

[0032] The difference determination module is used to determine the shape difference sequence of the attitude sequence based on the attitude sequence and the reference attitude of the attitude sequence; wherein, the shape difference sequence includes a shape difference operator corresponding to each time frame, and the shape difference operator corresponding to each time frame represents the shape difference between the three-dimensional attitude model of the time frame and the reference attitude;

[0033] A category determination module is used to determine the target action category of the posture sequence based on the shape difference sequence;

[0034] The training evaluation module is used to evaluate the training process of the operator based on the target action category and the action category mapping relationship; wherein, the action category mapping relationship includes action descriptions associated with different action categories.

[0035] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0036] Obtain the posture sequence of the operator during the current training process; wherein, the posture sequence includes a three-dimensional posture model of a preset number of consecutive time frames;

[0037] Based on the attitude sequence and the reference attitude of the attitude sequence, a shape difference sequence of the attitude sequence is determined; wherein, the shape difference sequence includes a shape difference operator corresponding to each time frame, and the shape difference operator corresponding to each time frame represents the shape difference between the three-dimensional attitude model of the time frame and the reference attitude;

[0038] Based on the shape difference sequence, determine the target action category of the posture sequence;

[0039] The training process of the operator is evaluated based on the target action category and the action category mapping relationship; wherein, the action category mapping relationship includes action descriptions associated with different action categories.

[0040] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0041] Obtain the posture sequence of the operator during the current training process; wherein, the posture sequence includes a three-dimensional posture model of a preset number of consecutive time frames;

[0042] Based on the attitude sequence and the reference attitude of the attitude sequence, a shape difference sequence of the attitude sequence is determined; wherein, the shape difference sequence includes a shape difference operator corresponding to each time frame, and the shape difference operator corresponding to each time frame represents the shape difference between the three-dimensional attitude model of the time frame and the reference attitude;

[0043] Based on the shape difference sequence, determine the target action category of the posture sequence;

[0044] The training process of the operator is evaluated based on the target action category and the action category mapping relationship; wherein, the action category mapping relationship includes action descriptions associated with different action categories.

[0045] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0046] Obtain the posture sequence of the operator during the current training process; wherein, the posture sequence includes a three-dimensional posture model of a preset number of consecutive time frames;

[0047] Based on the attitude sequence and the reference attitude of the attitude sequence, a shape difference sequence of the attitude sequence is determined; wherein, the shape difference sequence includes a shape difference operator corresponding to each time frame, and the shape difference operator corresponding to each time frame represents the shape difference between the three-dimensional attitude model of the time frame and the reference attitude;

[0048] Based on the shape difference sequence, determine the target action category of the posture sequence;

[0049] The training process of the operator is evaluated based on the target action category and the action category mapping relationship; wherein, the action category mapping relationship includes action descriptions associated with different action categories.

[0050] The aforementioned training evaluation method, device, computer equipment, and storage medium acquire the operator's posture sequence during the current training process. The posture sequence includes a preset number of consecutive time frames representing a 3D posture model. Based on the posture sequence and its reference posture, a shape difference sequence is determined. This shape difference sequence includes a shape difference operator corresponding to each time frame, representing the shape difference between the 3D posture model of that time frame and the reference posture. Based on the shape difference sequence, the target action category of the posture sequence is determined. The operator's training process is evaluated based on the target action category and the action category mapping relationship. The action category mapping relationship includes action descriptions associated with different action categories. This scheme, based on 3D posture data and difference operators, transforms action differences into quantifiable indicators, eliminating the subjective randomness of manual evaluation and accurately locating the deviation of each frame's posture from the reference. Furthermore, based on the target action category and action category mapping relationship determined by the shape difference sequence, the standardization of the operator's actions can be accurately identified, thereby achieving real-time, objective training assessment and evaluation, improving training quality and operational safety. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart illustrating a training evaluation method in one embodiment;

[0053] Figure 2 This is a flowchart illustrating the process of determining the shape difference sequence of an attitude sequence in one embodiment;

[0054] Figure 3 This is a flowchart illustrating the process of determining the shape difference operator in one embodiment;

[0055] Figure 4 This is a schematic diagram of the three-dimensional vector field of the shape difference operator in one embodiment;

[0056] Figure 5This is a flowchart illustrating the process of determining the target action category of a pose sequence in one embodiment;

[0057] Figure 6 This is a temporal evolution diagram of the shape difference operator sequence in one embodiment;

[0058] Figure 7 This is a flowchart illustrating the evaluation of the operator's training process in one embodiment;

[0059] Figure 8 This is a structural block diagram of the training and evaluation device in one embodiment;

[0060] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0062] The training and evaluation method provided in this application can be applied to scenarios involving high-precision, standardized skills training, motion evaluation, and real-time assessment of specific operations (such as industrial maintenance and railway transportation) in virtual reality, mixed reality, or augmented reality environments. The training and evaluation method provided in this application is executed by computer equipment, which can be the backend server of a virtual simulation training system.

[0063] In one exemplary embodiment, such as Figure 1 As shown, a practical training evaluation method is provided. Taking the application of this method to a server as an example, the specific steps include:

[0064] S101, Obtain the posture sequence of the operator during the current training process.

[0065] The pose sequence includes a preset number of consecutive time frames of a 3D pose model. The 3D pose model can be a 3D mesh model of the human body.

[0066] Optionally, the operator's posture sequence during the current training process can be collected using a data acquisition device. In this embodiment, the data acquisition device can employ the mature passive optical motion capture system – the Qualisys system. This system uses reflective markers affixed to key joints and parts of the operator's body as acquisition points, and utilizes multiple high-speed infrared cameras to capture the three-dimensional spatial positions of these acquisition points from different angles. Finally, a high-fidelity three-dimensional human body mesh model is reconstructed in real time using software, serving as a three-dimensional posture model.

[0067] For example, a complete current training process is defined as a dynamic sequence of poses. The pose sequence consists of a preset number (e.g., It consists of ) consecutive time frames, and can be represented as: ,in, The total number of time frames depends on the action duration and sampling frequency. . (in, () represents a time frame in the attitude sequence The three-dimensional pose model.

[0068] In this embodiment, the 3D pose model for each time frame includes multiple nodes, connecting edges between nodes, and target position information for each node; nodes represent acquisition points located on the operator, and connecting edges represent the connection relationships between acquisition points; the nodes and connecting edges of the 3D pose models in different time frames are the same, but the target position information of each node is different. For example, the 3D pose model for each time frame may consist of a set of nodes. and a topology Common definition, that is Since the connections between the operator's acquisition points remain constant across all time frames, the nodes and their connections in the 3D pose model are identical across different time frames. However, as the operator's pose changes, the positions of the acquisition points shift, resulting in different target position information for each node in the 3D pose model across different time frames. (Node set) Include Each node can be represented as: .in, This indicates the number of data collection points, or the number of nodes. Each node... It is a three-dimensional vector, representing the first... The node at the th Cartesian coordinates at each time frame represent position information, i.e. The acquired raw pose sequence can be mathematically considered as a sequence with dimension 1. The tensor.

[0069] It should be noted that the raw pose data acquired may contain noise, global displacement and rotation, and scale inconsistencies due to individual body size differences, all of which can interfere with the accuracy of subsequent data processing. Therefore, in this embodiment, the acquired raw pose sequence needs to be preprocessed to generate a normalized pose sequence. Specifically, the preprocessing process may include noise removal, translation alignment, and scale normalization.

[0070] For example, to eliminate high-frequency jitter and measurement errors, a time-dimensional low-pass filter (Gaussian filter or moving average filter) can be applied to the coordinate trajectory of each vertex to eliminate noise in the original pose sequence. Taking a one-dimensional Gaussian filter as an example, the nodes... New coordinates after filtering calculate:

[0071]

[0072] in, It is the size of the filtering window; It is the Gaussian kernel function, defined as:

[0073]

[0074] in, It is the standard deviation of the Gaussian distribution.

[0075] This operation can smooth the motion trajectory of each node, resulting in a denoised sequence. .

[0076] To eliminate the impact of the human body's global displacement within a scene on motion recognition, the 3D pose model of each time frame needs to be aligned to a unified coordinate center, typically the origin. First, the pose model of each time frame is calculated... The geometric center (point mass) ,in:

[0077]

[0078] Subtract the centroid from all nodes in the given time frame to obtain the centered nodes. After this processing, the centroid of the 3D pose model of each video frame is located at the origin, thus eliminating the interference of global translation.

[0079] To eliminate the influence of different operator sizes, the 3D pose model of each video frame needs to be scaled to a uniform scale. In this embodiment, the average distance from all nodes to the centroid can be calculated as a scale factor and used for normalization. The 3D pose model corresponding to the first time frame is then used... Calculate the scaling factor based on the following:

[0080]

[0081] in, This represents the L2 norm.

[0082] Then, by dividing all centered nodes in each time frame of the sequence by the scale factor, we get:

[0083]

[0084] This operation ensures that the 3D pose models from different operators have similar scales before entering subsequent processing, enhancing the algorithm's generalization ability. After the above preprocessing, a normalized pose sequence is finally obtained. ,in, , .

[0085] S102, Based on the attitude sequence and the reference attitude of the attitude sequence, determine the shape difference sequence of the attitude sequence.

[0086] The shape difference sequence includes a shape difference operator for each time frame, which represents the shape difference between the 3D pose model and the reference pose for that time frame. The reference pose includes multiple nodes, the connecting edges between nodes, and the reference position information of each node. The reference pose serves as the reference for the 3D pose model.

[0087] Optionally, in order to quantify the change of attitude over time, it is first necessary to establish a reference baseline. In the embodiments of this application, the reference attitude can be a three-dimensional attitude model of any time frame of the attitude sequence, or it can be the mean model of the attitude sequence.

[0088] In one embodiment, the 3D pose model corresponding to the first time frame in the pose sequence can be used as the reference pose of the pose sequence. It should be noted that the 3D pose model of the first time frame typically represents the starting or static preparatory pose of an action. Using this as the reference pose, the deformation of all subsequent time frames unfolds relative to this initial state, conforming to the intuitive human understanding of action—a process from stillness to movement and then to completion. For example, a pose sequence is selected... The 3D pose model corresponding to the first time frame in the data. The baseline pose for this pose sequence is denoted as . The set of nodes for this reference attitude. That is, the set of nodes of the 3D pose model in the first time frame. ,Right now:

[0089]

[0090] Furthermore, the shape difference operator for each time frame can be described by calculating the displacement vector between the nodes of the 3D pose model corresponding to that time frame and the nodes of the reference pose.

[0091] S103, determine the target action category of the posture sequence based on the shape difference sequence.

[0092] Among them, the target action category refers to the category of operations performed by the operator during the training process.

[0093] Optionally, an action classification model can be pre-trained, and the shape difference sequence can be input into the action classification model so that the action classification model can analyze the shape difference sequence based on the model parameters and output the target action category of the pose sequence.

[0094] S104. Evaluate the operator's training process based on the target action category and the mapping relationship between action categories.

[0095] The action category mapping relationship includes action descriptions associated with different action categories.

[0096] Optionally, the action description associated with the target action category can be found from the action category mapping relationship. Furthermore, the training process of the operator can be evaluated based on the action description.

[0097] In the aforementioned training evaluation method, the operator's posture sequence during the current training process is obtained. This posture sequence includes a 3D posture model spanning a preset number of consecutive time frames. Based on the posture sequence and its reference posture, a shape difference sequence is determined. This shape difference sequence includes a shape difference operator corresponding to each time frame, representing the shape difference between the 3D posture model of that time frame and the reference posture. Based on the shape difference sequence, the target action category of the posture sequence is determined. The operator's training process is evaluated based on the target action category and the action category mapping relationship. The action category mapping relationship includes action descriptions associated with different action categories. This scheme, based on 3D posture data and difference operators, transforms action differences into quantifiable indicators, eliminating the subjective randomness of manual evaluation and accurately locating the deviation between each frame's posture and the reference. Furthermore, based on the target action category and action category mapping relationship determined by the shape difference sequence, the standardization of the operator's actions can be accurately identified, thereby achieving real-time, objective training assessment and evaluation, improving training quality and operational safety.

[0098] Optionally, in an exemplary embodiment, such as Figure 2 As shown, a method for determining the shape difference sequence of an attitude sequence is provided, which specifically includes the following steps:

[0099] S201, determine the reference attitude of the attitude sequence.

[0100] Optionally, the 3D pose model corresponding to the first time frame in the pose sequence can be used as the reference pose of the pose sequence.

[0101] S202, for each time frame, determine the shape difference operator corresponding to the time frame based on the target position information and reference position information of each node.

[0102] Optionally, a shape difference operator for each time frame can be constructed by calculating the shape difference between the 3D pose model and the reference model for each time frame in the pose sequence. For example, for each time frame, the relative displacement between the target position information and the reference position information of each node within that time frame can be calculated, and the shape difference operator for that time frame can be constructed based on the relative displacement of each node.

[0103] S203, construct the shape difference sequence of the attitude sequence based on the shape difference operator corresponding to each time frame.

[0104] Optionally, by performing the above calculations on each time frame of the pose sequence, the original pose sequence, consisting of absolute coordinates, is obtained. It is then successfully transformed into a new sequence of shape difference operators consisting of relative displacements.

[0105] In this embodiment, by determining the baseline pose, a unified reference is provided for the pose difference calculation of all time frames. The differential sequence constructed in the order of time frames fully records the trend of pose difference changes over time, and can depict the dynamic evolution of the action, rather than isolated single-frame differences.

[0106] Optionally, in an exemplary embodiment, such as Figure 3 As shown, a method for determining a shape difference operator is provided, which specifically includes the following steps:

[0107] S301, for each node, determine the moving distance and moving direction between the target position information and the reference position information, and determine the displacement vector of the node based on the moving distance and moving direction.

[0108] Optionally, for pose sequences 3D pose model corresponding to any time frame Its shape is similar to the reference template. The difference can be described by calculating the displacement vector between the corresponding nodes of the two. Defined at the... The first time frame The displacement vectors of the nodes are:

[0109]

[0110] in, Indicates the first The first time frame The target location information of each node can be represented in the form of coordinates; Indicating the first in the reference template The reference position information of each node can also be represented in the form of coordinates.

[0111] Displacement vector It captures the relative movement of a single node from its initial position to its current position, including direction and size information.

[0112] S302, determine the shape difference operator corresponding to the time frame based on the displacement vector of each node.

[0113] Optionally, the displacement vectors of all nodes in a time frame are collected to form the shape difference operator corresponding to that time frame. The shape difference operator can be represented as a... The matrix is:

[0114]

[0115] Among them, matrix A complete description of the time frame At that time, the operator's entire body undergoes global deformation relative to the initial posture.

[0116] It should be noted that for the first time frame of the attitude sequence, its shape difference operator... It is The zero matrix.

[0117] like Figure 4 The diagram shown is a three-dimensional vector field schematic of the shape difference operator provided in this application embodiment, used to intuitively demonstrate the vertex displacement of the three-dimensional pose model corresponding to the current time frame relative to the reference pose.

[0118] In this embodiment, through feature extraction, the pose sequence shifts the focus from the absolute position of the nodes to the "motion" and "deformation" that better reflect the essence of the action, effectively capturing the core features of the action; due to the use of a relative calculation method, this representation has stronger robustness to minor global translations or rotations that could not be completely eliminated in the preprocessing stage.

[0119] Optionally, in one embodiment, the process of determining the target action category of the pose sequence based on the shape difference sequence can be accomplished by an action classification model, which may include a convolutional network for extracting spatial features and a long short-term memory network for action recognition and classification. Based on this, as... Figure 5 As shown, a method for determining the target action category of a pose sequence is provided, which specifically includes the following steps:

[0120] S501, spatial feature extraction is performed on the shape difference sequence to obtain the spatial feature sequence.

[0121] The spatial feature sequence includes the spatial feature vector corresponding to each time frame.

[0122] Optionally, each shape difference operator in the shape difference sequence obtained in the above embodiments is a In this embodiment, the shape difference operator can be considered as a matrix with height of [missing information]. A two-dimensional pseudo-image with a width of 3 and a channel count of 1. Each pixel value in the image represents a specific node on the human body surface in a specific direction. Displacements on the surface. Convolutional networks can capture the local and global patterns of these displacements in the combined motion of a human model by learning a set of optimized convolutional kernels.

[0123] In this application embodiment, a typical convolutional network architecture involved in this application includes the following parts:

[0124] (1) Convolutional Layers. Convolutional layers are the core of convolutional networks, extracting local features through convolution operations. There are several convolutional layers. For the first convolutional layer, its input is the feature map output from the previous layer. (in, This layer contains Each convolutional kernel has 10 convolutional kernels. (in, The dimensions of ) are .

[0125] No. Layer Feature maps generated by each convolutional kernel an element on The calculation is as follows:

[0126]

[0127] in, It is the first Layer One input feature map; It is the first Layer Each convolutional kernel acts on the input feature map Part of it; It is the first Layer The bias terms corresponding to each convolution kernel; It is a non-linear activation function. By stacking multiple convolutional layers, the network can learn more complex shape and component features—that is, complex deformation patterns of the entire limb or torso—from simple edges and textures—that is, small-scale vertex co-locations.

[0128] To introduce non-linearity and enable the network to learn more complex functions, each convolutional layer is typically followed by an activation function. In the embodiments of this application, a modified linear unit is preferably used as the activation function, which can effectively alleviate the gradient vanishing problem.

[0129]

[0130] Pooling layers are typically interspersed between consecutive convolutional layers. Their function is to reduce the spatial dimensionality of the feature map, thereby reducing the number of network parameters, increasing the receptive field, and enhancing the robustness of features to small translations. In the embodiments of this application, max pooling is used to pool a feature map... The pooling window operates as follows:

[0131]

[0132] This operation extracts the maximum value from a local region of the feature map as the representative of that region, preserving the most salient features. After multiple convolutions and pooling, a set of high-order feature maps is obtained. These two-dimensional feature maps need to be "flattened" to stretch all elements of the feature maps into a one-dimensional vector, denoted as . The flattened vector is then input into one or more fully connected layers for final feature integration and mapping. The mathematical representation of a fully connected layer is:

[0133]

[0134] in, It is the weight matrix of the fully connected layer; It is the bias vector; This is the activation function of that layer. Ultimately, the output of the convolutional network model is a vector image with dimension 1. One-dimensional real vector This is the spatial feature vector.

[0135] Spatial feature vectors It is the first The spatial feature representation of the pose deformation of each time frame is encoded, which condenses the original... All key spatial information in the matrix.

[0136] For shape difference sequences The shape difference operator corresponding to each time frame in the sequence performs the above process to obtain a new sequence of feature vectors. .

[0137] In one embodiment, such as Figure 6 The diagram shown is a temporal evolution diagram of the shape difference operator sequence provided in this application, used to visualize the spatiotemporal data structure as input to a convolutional network.

[0138] S502 analyzes the temporal dependencies between spatial feature vectors to obtain the spatiotemporal characteristics of the attitude sequence.

[0139] S503, based on spatiotemporal characteristics, determines the target action category of the posture sequence.

[0140] Optionally, the extraction of spatiotemporal features and the identification of target action categories can be achieved through a Long Short-Term Memory (LSTM) network. In this embodiment, each spatial feature vector is input into the LTM network frame by frame. The core computation process of an LTM network unit at any given time frame is as follows: Its input includes the spatial feature vector of the current time frame. The hidden state of the previous time step and the cell state at the previous time step. The Long Short-Term Memory (LSTM) network unit is mainly composed of three gates that work together to control the flow of information:

[0141] (1) Forgetting gate. The function of the forgetting gate is to determine the cell state from the previous moment. Which information is discarded? This will be reviewed. and and for Each element in the code outputs a value within the range [0, 1]. 1 represents "completely keep", and 0 represents "completely discard". The output of the forget gate is within the range [0, 1]. 1 represents "completely keep", and 0 represents "completely discard". The calculation is as follows:

[0142]

[0143] in, It is an activation function. . and These are the weight matrix and bias vector of the forget gate, respectively, which are the parameters that the network needs to learn. This means concatenating the hidden state from the previous moment with the current input vector.

[0144] (2) Input Gate. The input gate determines which new information needs to be stored in the cell state. It consists of two steps: 1. A Sigmoid layer determines which values ​​need to be updated, and its output is... 2. A tanh layer creates a new candidate value vector. It may be added to the cell state: .

[0145] (3) Cell state update. Now, the old cell state is combined with the new candidate value to generate the cell state at the current moment. The update rule is as follows: .in, This represents the Hadamard product, which is the product of elements.

[0146] (4) Output gate. The output gate determines the hidden state at the current moment. This hidden state will be output as the current time step and passed to the next time step. A sigmoid layer determines which parts of the cell state will be output, and its output is... The cell state is passed through a tanh layer (its values ​​are scaled to the [-1, 1] interval) and compared with... Multiplying them together yields the final hidden state. .

[0147] Long Short-Term Memory Networks arrive Process the entire feature sequence sequentially After processing the last time step Then, the Long Short-Term Memory network outputs the final hidden state. In other words, spatiotemporal features can be considered a highly condensed representation of the entire dynamic pose. To obtain the final classification result, [the following steps are taken]. The input is fed into a fully connected layer, which uses the Softmax activation function to output a probability distribution. Assume there are a total of predefined standard action categories. There are [number] categories. .

[0148] First, a score for each category is calculated using a fully connected layer, denoted as a score vector:

[0149]

[0150] in, and These are the weights and biases of the classification layer. Then, the score vector is transformed into a probability distribution vector using the Softmax function.

[0151]

[0152] in, It is the input sequence Category The predicted probability is calculated by... Ultimately, the predicted target action category is the category with the highest probability.

[0153] In this embodiment, the abstract posture difference sequence is transformed into a concrete spatial feature vector through spatial feature extraction, which fully preserves the morphological differences of each frame of action. Temporal dependency analysis is performed on the spatial feature vector to characterize the dynamic change law of the action. The spatiotemporal features simultaneously cover spatial morphological differences and temporal change laws, which can fully characterize complex training actions and ensure the accuracy of the determined target action category.

[0154] It should be noted that the action classification model can include a convolutional network for extracting spatial features and a long short-term memory network for action recognition and classification. The action classification model can be trained using supervised learning. Once the model is trained, it can also generate action category mappings, which can be stored as an action library.

[0155] Optionally, in one embodiment, such as Figure 7 As shown, a method for evaluating the training process of operators is provided, which specifically includes the following steps:

[0156] S701, from the action category mapping relationship, find the target action description associated with the target action category.

[0157] Optionally, a unique target action description can be found in a preset mapping database based on the target action category.

[0158] S702, evaluate the operator's training process based on the target action description and the standard action description required by the training task.

[0159] Optionally, the description of the target action can be compared with the standard action description required by the training task. If the target action description matches the standard action description, it is rated as "correct operation". If the target action description belongs to a predefined category of incorrect action, it is rated as "incorrect operation".

[0160] Based on the comparison results, the system can provide trainees with immediate and clear feedback in the virtual simulation training interface. For the assessment of "correct operation", the system can display positive incentive information on the interface, such as a green success prompt, score increase, or voice praise, to reinforce the memory of correct operation.

[0161] For assessments of "operational errors," the system displays corrective information. This information includes indicating the error type (obtained by querying a mapping table, such as "Error: Detected posture is bent over instead of a standard squatting posture"), and can be visually marked on the virtual instructor or trainee's own virtual body using highlighting or animation to guide them on how to perform the action correctly. Furthermore, the system allows setting a confidence threshold. When the model's prediction confidence for all categories is low, the system can classify it as "incorrect movement" or "unrecognizable," and prompt the user to retry with a more standard posture.

[0162] In this embodiment, by describing and comparing the spatial, temporal, and logical dimensions of the precise positioning deviation, a general assessment is avoided. By combining the degree of deviation with preset rules, the subjective randomness of manual assessment is eliminated, and the assessment results of the same action can be reproduced.

[0163] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0164] Based on the same inventive concept, this application also provides a training evaluation device for implementing the training evaluation method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more training evaluation device embodiments provided below can be found in the limitations of the training evaluation method described above, and will not be repeated here.

[0165] In one exemplary embodiment, such as Figure 8 As shown, a training evaluation device 800 is provided, including: a sequence acquisition module 810, a difference determination module 820, a category determination module 830, and a training evaluation module 840, wherein:

[0166] The sequence acquisition module 810 is used to acquire the posture sequence of the operator in the current training process; wherein, the posture sequence includes a three-dimensional posture model of a preset number of consecutive time frames.

[0167] The difference determination module 820 is used to determine the shape difference sequence of the attitude sequence based on the attitude sequence and the reference attitude of the attitude sequence; wherein, the shape difference sequence includes the shape difference operator corresponding to each time frame, and the shape difference operator corresponding to each time frame represents the shape difference between the three-dimensional attitude model of the time frame and the reference attitude.

[0168] The category determination module 830 is used to determine the target action category of the posture sequence based on the shape difference sequence.

[0169] The training evaluation module 840 is used to evaluate the training process of the operator based on the target action category and the action category mapping relationship; wherein, the action category mapping relationship includes the action description associated with different action categories.

[0170] The aforementioned training evaluation device acquires the operator's posture sequence during the current training process. This posture sequence includes a preset number of consecutive time frames representing a 3D posture model. Based on the posture sequence and its reference posture, a shape difference sequence is determined. This shape difference sequence includes a shape difference operator corresponding to each time frame, representing the shape difference between the 3D posture model of that time frame and the reference posture. Based on the shape difference sequence, the target action category of the posture sequence is determined. The operator's training process is evaluated based on the target action category and the action category mapping relationship. The action category mapping relationship includes action descriptions associated with different action categories. This scheme, based on 3D posture data and difference operators, transforms action differences into quantifiable indicators, eliminating the subjective randomness of manual evaluation and accurately locating the deviation of each frame's posture from the reference. Furthermore, based on the target action category and action category mapping relationship determined by the shape difference sequence, the standardization of the operator's actions can be accurately identified, thereby achieving real-time, objective training assessment and evaluation, improving training quality and operational safety.

[0171] In one embodiment, the 3D pose model of each time frame includes multiple nodes, connecting edges between nodes, and target position information of each node; nodes represent acquisition points located on the operator, and connecting edges represent the connection relationship between acquisition points; the nodes and connecting edges of the 3D pose models of different time frames are the same, but the target position information of each node is different; the reference pose includes multiple nodes, connecting edges between nodes, and reference position information of each node.

[0172] In one embodiment, the difference determination module 820 includes:

[0173] The attitude determination unit is used to determine the reference attitude of the attitude sequence.

[0174] The operator determination unit is used to determine the shape difference operator corresponding to each time frame based on the target position information and reference position information of each node.

[0175] The sequence determination unit is used to construct the shape difference sequence of the attitude sequence based on the shape difference operator corresponding to each time frame.

[0176] In one embodiment, the attitude determination unit is specifically used for:

[0177] The 3D pose model corresponding to the first time frame in the pose sequence is used as the reference pose of the pose sequence.

[0178] In one embodiment, the operator determination unit is specifically used for:

[0179] For each node, determine the moving distance and direction between the target position information and the reference position information; and determine the displacement vector of the node based on the moving distance and direction; and determine the shape difference operator corresponding to the time frame based on the displacement vector of each node.

[0180] In one embodiment, the category determination module 830 is specifically used for:

[0181] Spatial features are extracted from the shape difference sequence to obtain a spatial feature sequence, which includes a spatial feature vector corresponding to each time frame. The temporal dependency between the spatial feature vectors is analyzed to obtain the spatiotemporal features of the pose sequence. Based on the spatiotemporal features, the target action category of the pose sequence is determined.

[0182] In one embodiment, the training evaluation module 840 is specifically used for:

[0183] From the action category mapping relationship, find the target action description associated with the target action category; based on the target action description and the standard action description required by the training task, evaluate the operator's training process.

[0184] Each module in the aforementioned training and evaluation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0185] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and databases. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media to run. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a practical evaluation method.

[0186] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0187] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0188] Obtain the posture sequence of the operator during the current training process; wherein, the posture sequence includes a three-dimensional posture model of a preset number of consecutive time frames;

[0189] Based on the attitude sequence and the reference attitude of the attitude sequence, the shape difference sequence of the attitude sequence is determined; wherein, the shape difference sequence includes the shape difference operator corresponding to each time frame, and the shape difference operator corresponding to each time frame represents the shape difference between the three-dimensional attitude model of the time frame and the reference attitude.

[0190] Based on the shape difference sequence, determine the target action category of the posture sequence;

[0191] The training process of the operator is evaluated based on the target action category and the action category mapping relationship; the action category mapping relationship includes action descriptions associated with different action categories.

[0192] In one embodiment, the 3D pose model of each time frame includes multiple nodes, connecting edges between nodes, and target position information of each node; nodes represent acquisition points located on the operator, and connecting edges represent the connection relationship between acquisition points; the nodes and connecting edges of the 3D pose models of different time frames are the same, but the target position information of each node is different; the reference pose includes multiple nodes, connecting edges between nodes, and reference position information of each node.

[0193] In one embodiment, when the processor executes a computer program to determine the shape difference sequence of the attitude sequence based on the attitude sequence and a reference attitude of the attitude sequence, it also performs the following steps:

[0194] Determine the reference pose of the pose sequence; for each time frame, determine the shape difference operator corresponding to the time frame based on the target position information and reference position information of each node; construct the shape difference sequence of the pose sequence based on the shape difference operator corresponding to each time frame.

[0195] In one embodiment, when the processor executes a computer program to determine a reference pose for the pose sequence, it also performs the following steps:

[0196] The 3D pose model corresponding to the first time frame in the pose sequence is used as the reference pose of the pose sequence.

[0197] In one embodiment, when the processor executes a computer program to determine the shape difference operator corresponding to a time frame based on the target position information and reference position information of each node, it also performs the following steps:

[0198] For each node, determine the moving distance and direction between the target position information and the reference position information; and determine the displacement vector of the node based on the moving distance and direction; and determine the shape difference operator corresponding to the time frame based on the displacement vector of each node.

[0199] In one embodiment, when the processor executes a computer program to determine the target action category of a pose sequence based on a shape difference sequence, it also performs the following steps:

[0200] Spatial features are extracted from the shape difference sequence to obtain a spatial feature sequence, which includes a spatial feature vector corresponding to each time frame. The temporal dependency between the spatial feature vectors is analyzed to obtain the spatiotemporal features of the pose sequence. Based on the spatiotemporal features, the target action category of the pose sequence is determined.

[0201] In one embodiment, when the processor executes a computer program to evaluate the operator's training process based on the target action category and the action category mapping relationship, it also performs the following steps:

[0202] From the action category mapping relationship, find the target action description associated with the target action category; based on the target action description and the standard action description required by the training task, evaluate the operator's training process.

[0203] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0204] Obtain the posture sequence of the operator during the current training process; wherein, the posture sequence includes a three-dimensional posture model of a preset number of consecutive time frames;

[0205] Based on the attitude sequence and the reference attitude of the attitude sequence, the shape difference sequence of the attitude sequence is determined; wherein, the shape difference sequence includes the shape difference operator corresponding to each time frame, and the shape difference operator corresponding to each time frame represents the shape difference between the three-dimensional attitude model of the time frame and the reference attitude.

[0206] Based on the shape difference sequence, determine the target action category of the posture sequence;

[0207] The training process of the operator is evaluated based on the target action category and the action category mapping relationship; the action category mapping relationship includes action descriptions associated with different action categories.

[0208] In one embodiment, the 3D pose model of each time frame includes multiple nodes, connecting edges between nodes, and target position information of each node; nodes represent acquisition points located on the operator, and connecting edges represent the connection relationship between acquisition points; the nodes and connecting edges of the 3D pose models of different time frames are the same, but the target position information of each node is different; the reference pose includes multiple nodes, connecting edges between nodes, and reference position information of each node.

[0209] In one embodiment, when the processor executes a computer program to determine the shape difference sequence of the attitude sequence based on the attitude sequence and a reference attitude of the attitude sequence, it also performs the following steps:

[0210] Determine the reference pose of the pose sequence; for each time frame, determine the shape difference operator corresponding to the time frame based on the target position information and reference position information of each node; construct the shape difference sequence of the pose sequence based on the shape difference operator corresponding to each time frame.

[0211] In one embodiment, when the processor executes a computer program to determine a reference pose for the pose sequence, it also performs the following steps:

[0212] The 3D pose model corresponding to the first time frame in the pose sequence is used as the reference pose of the pose sequence.

[0213] In one embodiment, when the processor executes a computer program to determine the shape difference operator corresponding to a time frame based on the target position information and reference position information of each node, it also performs the following steps:

[0214] For each node, determine the moving distance and direction between the target position information and the reference position information; and determine the displacement vector of the node based on the moving distance and direction; and determine the shape difference operator corresponding to the time frame based on the displacement vector of each node.

[0215] In one embodiment, when the processor executes a computer program to determine the target action category of a pose sequence based on a shape difference sequence, it also performs the following steps:

[0216] Spatial features are extracted from the shape difference sequence to obtain a spatial feature sequence, which includes a spatial feature vector corresponding to each time frame. The temporal dependency between the spatial feature vectors is analyzed to obtain the spatiotemporal features of the pose sequence. Based on the spatiotemporal features, the target action category of the pose sequence is determined.

[0217] In one embodiment, when the processor executes a computer program to evaluate the operator's training process based on the target action category and the action category mapping relationship, it also performs the following steps:

[0218] From the action category mapping relationship, find the target action description associated with the target action category; based on the target action description and the standard action description required by the training task, evaluate the operator's training process.

[0219] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0220] Obtain the posture sequence of the operator during the current training process; wherein, the posture sequence includes a three-dimensional posture model of a preset number of consecutive time frames;

[0221] Based on the attitude sequence and the reference attitude of the attitude sequence, the shape difference sequence of the attitude sequence is determined; wherein, the shape difference sequence includes the shape difference operator corresponding to each time frame, and the shape difference operator corresponding to each time frame represents the shape difference between the three-dimensional attitude model of the time frame and the reference attitude.

[0222] Based on the shape difference sequence, determine the target action category of the posture sequence;

[0223] The training process of the operator is evaluated based on the target action category and the action category mapping relationship; the action category mapping relationship includes action descriptions associated with different action categories.

[0224] In one embodiment, the 3D pose model of each time frame includes multiple nodes, connecting edges between nodes, and target position information of each node; nodes represent acquisition points located on the operator, and connecting edges represent the connection relationship between acquisition points; the nodes and connecting edges of the 3D pose models of different time frames are the same, but the target position information of each node is different; the reference pose includes multiple nodes, connecting edges between nodes, and reference position information of each node.

[0225] In one embodiment, when the processor executes a computer program to determine the shape difference sequence of the attitude sequence based on the attitude sequence and a reference attitude of the attitude sequence, it also performs the following steps:

[0226] Determine the reference pose of the pose sequence; for each time frame, determine the shape difference operator corresponding to the time frame based on the target position information and reference position information of each node; construct the shape difference sequence of the pose sequence based on the shape difference operator corresponding to each time frame.

[0227] In one embodiment, when the processor executes a computer program to determine a reference pose for the pose sequence, it also performs the following steps:

[0228] The 3D pose model corresponding to the first time frame in the pose sequence is used as the reference pose of the pose sequence.

[0229] In one embodiment, when the processor executes a computer program to determine the shape difference operator corresponding to a time frame based on the target position information and reference position information of each node, it also performs the following steps:

[0230] For each node, determine the moving distance and direction between the target position information and the reference position information; and determine the displacement vector of the node based on the moving distance and direction; and determine the shape difference operator corresponding to the time frame based on the displacement vector of each node.

[0231] In one embodiment, when the processor executes a computer program to determine the target action category of a pose sequence based on a shape difference sequence, it also performs the following steps:

[0232] Spatial features are extracted from the shape difference sequence to obtain a spatial feature sequence, which includes a spatial feature vector corresponding to each time frame. The temporal dependency between the spatial feature vectors is analyzed to obtain the spatiotemporal features of the pose sequence. Based on the spatiotemporal features, the target action category of the pose sequence is determined.

[0233] In one embodiment, when the processor executes a computer program to evaluate the operator's training process based on the target action category and the action category mapping relationship, it also performs the following steps:

[0234] From the action category mapping relationship, find the target action description associated with the target action category; based on the target action description and the standard action description required by the training task, evaluate the operator's training process.

[0235] It should be noted that the data involved in this application (including but not limited to data used for analysis, data stored, data displayed, etc.) are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0236] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0237] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0238] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A practical training assessment method, characterized in that, The method includes: Obtain the posture sequence of the operator during the current training process; wherein, the posture sequence includes a three-dimensional posture model of a preset number of consecutive time frames; Based on the attitude sequence and the reference attitude of the attitude sequence, a shape difference sequence of the attitude sequence is determined; wherein, the shape difference sequence includes a shape difference operator corresponding to each time frame, and the shape difference operator corresponding to each time frame represents the shape difference between the three-dimensional attitude model of the time frame and the reference attitude; Based on the shape difference sequence, determine the target action category of the posture sequence; The training process of the operator is evaluated based on the target action category and the action category mapping relationship; wherein, the action category mapping relationship includes action descriptions associated with different action categories.

2. The method according to claim 1, characterized in that, The 3D pose model of each time frame includes multiple nodes, connecting edges between nodes, and target position information of each node; the nodes represent acquisition points located on the operator, and the connecting edges represent the connection relationship between the acquisition points; The nodes and connecting edges between nodes in the 3D pose model of different time frames are the same, but the target position information of each node is different. The reference pose includes multiple nodes, the connecting edges between nodes, and the reference position information of each node.

3. The method according to claim 2, characterized in that, Determining the shape difference sequence of the attitude sequence based on the attitude sequence and the reference attitude of the attitude sequence includes: Determine the reference pose of the pose sequence; For each time frame, the shape difference operator corresponding to the time frame is determined based on the target position information and reference position information of each node; The shape difference sequence of the attitude sequence is constructed based on the shape difference operator corresponding to each time frame.

4. The method according to claim 3, characterized in that, Determining the reference pose of the pose sequence includes: The three-dimensional pose model corresponding to the first time frame in the pose sequence is used as the reference pose of the pose sequence.

5. The method according to claim 3, characterized in that, The step of determining the shape difference operator corresponding to the time frame based on the target position information and reference position information of each node includes: For each node, determine the moving distance and direction between the target location information and the reference location information; and, Based on the moving distance and moving direction, determine the displacement vector of the node; Based on the displacement vector of each node, the shape difference operator corresponding to the time frame is determined.

6. The method according to claim 1, characterized in that, Determining the target action category of the pose sequence based on the shape difference sequence includes: Spatial features are extracted from the shape difference sequence to obtain a spatial feature sequence; wherein, the spatial feature sequence includes a spatial feature vector corresponding to each time frame; The temporal dependencies between the spatial feature vectors are analyzed to obtain the spatiotemporal features of the attitude sequence; Based on the spatiotemporal characteristics, the target action category of the posture sequence is determined.

7. The method according to claim 1, characterized in that, The evaluation of the operator's training process based on the target action category and the action category mapping relationship includes: From the action category mapping relationship, find the target action description associated with the target action category; The training process of the operator is evaluated based on the target action description and the standard action description required by the training task.

8. A practical training assessment device, characterized in that, The device includes: The sequence acquisition module is used to acquire the posture sequence of the operator in the current training process; wherein, the posture sequence includes a three-dimensional posture model of a preset number of consecutive time frames. The difference determination module is used to determine the shape difference sequence of the attitude sequence based on the attitude sequence and the reference attitude of the attitude sequence; wherein, the shape difference sequence includes a shape difference operator corresponding to each time frame, and the shape difference operator corresponding to each time frame represents the shape difference between the three-dimensional attitude model of the time frame and the reference attitude; A category determination module is used to determine the target action category of the posture sequence based on the shape difference sequence; The training evaluation module is used to evaluate the training process of the operator based on the target action category and the action category mapping relationship; wherein, the action category mapping relationship includes action descriptions associated with different action categories.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images