Virtual table tennis motion control system based on computer
By integrating monitoring, data preprocessing, model creation and adaptive control algorithms into the virtual table tennis motion control system, the problem of poor capture of table tennis motion characteristics in the existing technology is solved, and more accurate table tennis motion simulation and player skill improvement are achieved.
Patent Information
- Application Number
- CN202510755677.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-07
- Publication Date
- 2025-09-05
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In existing technologies, neural network structures cannot effectively and accurately capture the spatiotemporal characteristics of table tennis movements, especially when considering the speed, rotation and trajectory of the ball, resulting in poor performance of the model when processing complex dynamic scenes.
A computer-based virtual table tennis motion control system is adopted, including a monitoring module, a data preprocessing module, a model creation module, an adaptive control algorithm module and a decision-making and execution module. High-precision cameras, 3D lidars, accelerometers and gyroscopes are used for data collection. Through PyTorch neural network training and adaptive control algorithms, the trajectory of the table tennis ball and the hitting position of the racket are predicted, and adaptive adjustments are made based on the player's historical data.
It improves the model's ability to capture the spatiotemporal characteristics of table tennis, enhances the model's prediction and classification capabilities, provides an immersive sports experience, and improves the fun of the game and the players' table tennis skills.
Smart Images
Figure CN120596893A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of motion control systems, in particular to a computer-based virtual table tennis motion control system. Background Art
[0002] The Virtual Table Tennis Control System is a computer-based table tennis simulation environment that allows users to play and train in a virtual space. Within the system, users can choose different opponents to play against or engage in single-player training mode, adjusting various parameters and settings to simulate different match scenarios and conditions.
[0003] The neural network structure used in the existing technology is still unable to effectively and accurately capture the spatiotemporal characteristics of table tennis. In particular, when the structure does not fully consider the characteristics of table tennis, such as the speed, rotation and trajectory of the ball, the model performs poorly when processing complex dynamic scenes. Therefore, we propose a computer-based virtual table tennis motion control system to solve the above problems. Summary of the Invention
[0004] The purpose of the present invention is to provide a computer-based virtual table tennis motion control system to solve the problems in the current market raised by the above background technology.
[0005] To achieve the above object, the present invention provides the following technical solutions:
[0006] A computer-based virtual table tennis control system includes a monitoring module, a serving and receiving device, a data preprocessing module, a model creation module, an adaptive control algorithm module, and a decision-making and execution module;
[0007] The data preprocessing module receives the raw data collected by the monitoring module and cleans, denoises, and standardizes the collected data. The model creation module predicts the trajectory, speed, and rotation of the table tennis ball, as well as the optimal hitting position and angle of the racket. The adaptive control algorithm module calculates the optimal control instructions based on the prediction results of the deep learning model and real-time sensor data.
[0008] In the creation model module, use PyTorch's nn module to define nn.Conv2d, nn.ReLU, nn.MaxPool2d and nn.Linear. In PyTorch, use nn.CrossEntropyLoss (cross entropy loss) in the nn module to select the corresponding loss function; in PyTorch, use the optimizer in the optim module to specify hyperparameters; in the training loop, use PyTorch's forward and backward methods to calculate losses and gradients, and use the optimizer's step method to update model parameters. At the same time, use the validation set to verify the model performance, load the test data, set the model to evaluation mode, close the layers used during training, and then predict the test data, calculate performance indicators, and use PyTorch's torch.save and torch.load functions to save and load model parameters.
[0009] In the adaptive control algorithm module, real-time video streams are obtained through the monitoring module, and visual technology is used to identify and track the position, speed and direction of the player and the ball. The established model is used to analyze and predict the movement trajectory of the table tennis ball. At the same time, the player's historical batting data is combined to predict his possible reaction. Based on the speed, rotation, landing point of the ball and the player's reaction time, a rule-based control strategy is used to formulate a control strategy, and the action instructions are executed through the controller to ensure that the racket on the serving and receiving device can contact the ball at the best time and in the best condition. By creating a model module to analyze the batting feedback data, identify successful batting patterns and reasons for failure, and adjust the parameters and structure of the control algorithm based on the information.
[0010] As a further optimization solution of the present invention, the monitoring module includes a high-precision camera, a 3D laser radar, an accelerometer and a gyroscope;
[0011] Among them, a high-precision camera is installed above or on the side of the table, responsible for capturing the movement trajectory of the ping-pong ball and racket; a 3D lidar is installed around or above the table to measure the distance and speed between the ball and the racket; an accelerometer and gyroscope are installed on the handle of the table tennis racket. The accelerometer can measure the acceleration and tilt angle of the racket, while the gyroscope can measure the angular velocity and rotation direction of the racket.
[0012] As a further optimization solution of the present invention, the data preprocessing module includes a data collection and integration unit, a data cleaning and denoising unit, a data standard and normalization unit, a feature extraction unit, a data enhancement and expansion unit, and a data division unit;
[0013] The data collection and integration unit is responsible for collecting raw data from the monitoring modules and integrating this data into a unified format and framework. The data cleaning and denoising unit is used to remove errors, outliers, duplicate data, and sensor noise from the raw data. The data standardization and normalization unit is used to convert sensor and measurement unit data to a unified scale. The feature extraction unit is used to extract the speed, acceleration, and rotation of table tennis from the preprocessed data. The data enhancement and expansion unit increases the diversity and size of the dataset through data enhancement techniques. The data partitioning unit is used to divide the preprocessed dataset into training, validation, and test sets.
[0014] Specifically, data interfaces and communication protocols are used to receive data streams from the monitoring module, and data formats are converted and synchronized. Kalman filtering and threshold setting are used to identify and correct abnormal data points and reduce the impact of noise. The minimum-maximum normalization method is used to map the data to a specific range and eliminate dimensional differences. Feature engineering technology and technical support from professionals are used to construct a table tennis feature set. The technology is enhanced through rotation, translation, scaling, and noise injection, and the original data is transformed to generate new training samples. The data set is divided into different subsets through stratified sampling, and the data is saved and managed through a database.
[0015] As a further optimization solution of the present invention, a model creation module includes a model definition unit, a loss function definition unit, an optimizer definition unit, a model training unit, a model evaluation and testing unit, and a model saving and loading unit;
[0016] Among them, the convolution layer, pooling layer, and fully connected layer in the architecture of the convolutional neural network are defined through the model definition unit, and the convolution kernel size, step size, and padding of each layer are specified; the loss function definition unit is used to define the loss function used to calculate the error between the predicted value and the true value during the model training process; the optimizer definition unit is used to define the stochastic gradient descent used to update the model parameters; the model training unit is used to iteratively train the model using training data, calculate the gradient and update the model parameters through the backpropagation algorithm; the model evaluation and testing unit is used to evaluate and test the trained model using test data; the model saving and loading unit is used to save the trained model parameters and load the saved model parameters for inference or continued training.
[0017] As a further optimization solution of the present invention, the adaptive control algorithm module includes a state perception unit, a prediction and decision unit, a prediction and decision unit, a real-time adjustment execution unit, and a feedback and optimization unit;
[0018] The state perception unit collects and analyzes information about the player's hitting movements, ball path changes, and opponent's movements. The prediction and decision-making unit predicts the ball's future trajectory and the player's possible hitting movements based on the current state and historical data. The prediction and decision-making unit formulates a control strategy for adjusting the racket's angle, speed, and power based on the prediction results and the current state. The real-time adjustment execution unit converts the control strategy into specific action instructions to adjust the racket's angle, speed, and power in real time.
[0019] As a further optimization solution of the present invention, the decision and execution module includes a decision unit, an instruction generation unit, a control interface unit, an execution unit, a feedback monitoring unit, and an adjustment and optimization unit;
[0020] Use the model and logical judgment in the model creation module to evaluate the effects of different strategies and select the best hitting plan. Generate angle adjustment, speed control and power adjustment signals based on the optimal strategy, establish a connection with the racket hardware through serial communication and wireless communication technology, execute control instructions by controlling the racket on the serving and receiving equipment, and monitor the angle, speed and power of the racket through the monitoring module.
[0021] Compared with the prior art, the present invention has the following beneficial effects:
[0022] The neural network structure designed in the present invention improves the ability to capture the spatiotemporal characteristics of table tennis motion, thereby improving the prediction and classification capabilities of the model. In the loss function, the cross-entropy loss and other regularization terms are combined to better deal with specific problems in virtual table tennis motion control, reduce model overfitting, and improve the generalization ability of the model.
[0023] The present invention can simulate the trajectory, speed and rotation of table tennis through a physical engine and motion algorithm, providing players with an immersive sports experience. It can be adaptively adjusted according to the player's level and performance, increasing the fun of the game and helping to improve the player's table tennis skills.
[0024] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 This is a block diagram of the computer-based virtual table tennis motion control system of the present invention. DETAILED DESCRIPTION
[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0027] Example 1
[0028] See also Figure 1 , a computer-based virtual table tennis control system, including a monitoring module, a ball serving and receiving device, a data preprocessing module, a model creation module, an adaptive control algorithm module, and a decision and execution module;
[0029] The data preprocessing module receives the raw data collected by the monitoring module and cleans, denoises, and standardizes the collected data. The model creation module predicts the trajectory, speed, and rotation of the table tennis ball, as well as the optimal hitting position and angle of the racket. The adaptive control algorithm module calculates the optimal control instructions based on the prediction results of the deep learning model and real-time sensor data.
[0030] Among them, the data preprocessing module includes data collection and integration unit, data cleaning and denoising unit, data standardization and normalization unit, feature extraction unit, data enhancement and expansion unit, and data division unit;
[0031] The data collection and integration unit is responsible for collecting raw data from the monitoring modules and integrating this data into a unified format and framework. The data cleaning and denoising unit is used to remove errors, outliers, duplicate data, and sensor noise from the raw data. The data standardization and normalization unit is used to convert sensor and measurement unit data to a unified scale. The feature extraction unit is used to extract the speed, acceleration, and rotation of table tennis from the preprocessed data. The data enhancement and expansion unit increases the diversity and size of the dataset through data enhancement techniques. The data partitioning unit is used to divide the preprocessed dataset into training, validation, and test sets.
[0032] Specifically, data interfaces and communication protocols are used to receive data streams from the monitoring module, and data formats are converted and synchronized. Kalman filtering and threshold setting are used to identify and correct abnormal data points and reduce the impact of noise. The minimum-maximum normalization method is used to map the data to a specific range and eliminate dimensional differences. Feature engineering technology and technical support from professionals are used to construct a table tennis feature set. The technology is enhanced through rotation, translation, scaling, and noise injection, and the original data is transformed to generate new training samples. The data set is divided into different subsets through stratified sampling, and the data is saved and managed through a database.
[0033] Specifically, when estimating the state of a dynamic system through Mann filter measurements, the prediction step is: (\hat{x}{k|k-1}=F{k|k-1}\hat{x}{k-1|k-1}+B{k|k-1}u_{k})
[0034] Update steps:
[0035] (\hat{x}{k|k}=\hat{x}{k|k-1}+K_{k}(z_{k}-H_{k}\hat{x}{k|k-1}))
[0036] Where (\hat{x}{k|k-1}) is the prior estimate, (\hat{x}{k|k}) is the posterior estimate, (F{k|k-1}) is the state transition matrix, (B_{k|k-1}) is the control matrix, (u_{k}) is the control vector, (K_{k}) is the Kalman gain, (z_{k}) is the observation vector, and (H_{k}) is the observation matrix.
[0037] The formula used to linearly transform the data to [0,1] through minimum-maximum normalization is:
[0038] (x_{\text{normalized}}=\frac{x-x_{\text{min}}}{x_{\text{max}}-x_{\text{min}}})
[0039] Where (x) is the original data, (x_{\text{min}}) and (x_{\text{max}}) are the minimum and maximum values in the dataset, respectively.
[0040] The model creation module includes the model definition unit, loss function definition unit, optimizer definition unit, model training unit, model evaluation and testing unit, and model saving and loading unit;
[0041] Among them, the convolution layer, pooling layer, and fully connected layer in the architecture of the convolutional neural network are defined through the model definition unit, and the convolution kernel size, step size, and padding of each layer are specified; the loss function definition unit is used to define the loss function used to calculate the error between the predicted value and the true value during the model training process; the optimizer definition unit is used to define the stochastic gradient descent used to update the model parameters; the model training unit is used to iteratively train the model using training data, calculate the gradient and update the model parameters through the backpropagation algorithm; the model evaluation and testing unit is used to evaluate and test the trained model using test data; the model saving and loading unit is used to save the trained model parameters and load the saved model parameters for inference or continued training.
[0042] Use PyTorch's nn module to define nn.Conv2d, nn.ReLU, nn.MaxPool2d, and nn.Linear. In PyTorch, use nn.CrossEntropyLoss (cross entropy loss) in the nn module to select the corresponding loss function; in PyTorch, use the optimizer in the optim module to specify hyperparameters; in the training loop, use PyTorch's forward and backward methods to calculate losses and gradients, and use the optimizer's step method to update model parameters. At the same time, use the validation set to verify model performance, load test data, set the model to evaluation mode, turn off the layers used during training, and then predict the test data to calculate performance indicators. Use PyTorch's torch.save and torch.load functions to save and load model parameters.
[0043] Specifically, in the model definition unit, the convolution layer is:
[0044] [\text{Output}(N_i,C_{\text{out} / \text{in}},H,W)=\text{Input}(N_i,C_{\text{in}},H_{\text{in}},W_{\text {in}})\ast\text{Weight}(C_{\text{out}},C_{\text{in}},\text{kernel_size},\text{stride},\text{padding})]
[0045] Activation function: [f(x)=\max(0,x)]
[0046] Pooling layer:
[0047] [\text{Output}(N_i,C,H_{\text{out}},W_{\text{out}})=\max_{x\in\text{kernel_size}}\text{Input}(N_i,C,H_{\text{in}},W_{\text{in}})]
[0048] Fully connected layer: [y = xW^T + b]
[0049] Use nn.CrossEntropyLoss to calculate the cross entropy loss:
[0050] [L=-\sum_{c=1}^{C}y_{c}\log(p_{c})]
[0051] Where (y_c) is the cth element of the true label (one-hot encoded), and (p_c) is the probability of the cth category predicted by the model.
[0052] In the optimizer definition unit, the formula used for updating by stochastic gradient descent is:
[0053] [\theta=\theta-\eta\nabla_{\theta}J(\theta;x,y)]
[0054] Where (\theta) is the model parameter, (\eta) is the learning rate, and (\nabla_{\theta}J(\theta;x,y)) is the gradient of the loss function with respect to the model parameters.
[0055] Adam's update:
[0056] [v_t=\beta_2v_{t-1}+(1-\beta_2)g_t^2]
[0057] [\hat{m}_t=\frac{m_t}{1-\beta_1^t}]
[0058] [\hat{v}t=\frac{v_t}{1-\beta_2^t}]
[0059] [\theta_t=\theta{t-1}-\frac{\eta}{\sqrt{\hat{v}_t+\epsilon}}\hat{m}_t]
[0060] where (m_t) and (v_t) are estimates of the first moment (mean) and second moment (uncentered variance) of the gradient, respectively, (\beta_1) and (\beta_2) are hyperparameters, (\g_t) is the current gradient, (\eta) is the learning rate, and (\epsilon) is a constant to prevent division by zero.
[0061] In the model training unit, the forward and backward methods are used to calculate the loss and gradient, and then the optimizer's step method is used to update the model parameters.
[0062] Among them, forward propagation:
[0063] [\text{Output}=\text{Model}(\text{Input})]
[0064] Loss calculation:
[0065] [\text{Loss}=\text{LossFunction}(\text{Output},\text{Target})]
[0066] Backward Propagation:
[0067] [\nabla_{\theta}\text{Loss}=\frac{\partial\text{Loss}}{\partia l\text{Output}}\frac{\partial\text{Output}}{\partial\theta}]
[0068] Update using optimizer parameters:
[0069] [\theta=\text{Optimizer}.\text{step}(\theta,\nabla_{\theta}\text{Loss})]
[0070] The adaptive control algorithm module includes a state perception unit, a prediction and decision-making unit, a real-time adjustment execution unit, and a feedback and optimization unit;
[0071] The state perception unit collects and analyzes information about the player's hitting movements, ball path changes, and opponent's movements. The prediction and decision-making unit predicts the ball's future trajectory and the player's possible hitting movements based on the current state and historical data. The prediction and decision-making unit formulates a control strategy for adjusting the racket's angle, speed, and power based on the prediction results and the current state. The real-time adjustment execution unit converts the control strategy into specific action instructions to adjust the racket's angle, speed, and power in real time.
[0072] The feedback and optimization unit collects feedback data from each shot, analyzes the shot effect, and optimizes and improves the control algorithm. It includes a feedback generation unit, a feedback display and output unit, and a user interaction and adjustment unit.
[0073] Among them, the feedback generation unit generates a visual feedback interface and audio prompts based on the processed data; the feedback display and output unit displays the generated feedback interface or audio prompts to the players; and the user interaction and adjustment unit enables the players and the feedback system to confirm the accuracy of the feedback information and adjust the feedback parameters.
[0074] Real-time video streams are acquired through the monitoring module, and visual technology is used to identify and track the position, speed, and direction of the player and the ball. The established model is used to analyze and predict the trajectory of the table tennis ball. At the same time, the player's historical batting data is combined to predict his or her possible reaction. Based on the ball's speed, rotation, landing point, and the player's reaction time, a rule-based control strategy is used to formulate a control strategy. The controller executes the action instructions to ensure that the racket on the serving and receiving device can contact the ball at the best time and in the best condition. By creating a model module to analyze the batting feedback data, the successful batting patterns and the reasons for failure are identified, and the parameters and structure of the control algorithm are adjusted according to the information.
[0075] Use graphical user interface technology to create graphics or animations indicating the impact position, force, and rotation of the ball at the impact point; use audio synthesis technology to generate pitch changes and sound intensity changes related to the force and rotation of the ball; present the feedback interface and audio prompts to the player through a display screen, headphones, and output devices; and adjust the display method of the feedback or the volume of the audio prompts through buttons, sliders, or menus.
[0076] The decision and execution module includes a decision unit, an instruction generation unit, a control interface unit, an execution unit, a feedback monitoring unit, and an adjustment and optimization unit;
[0077] Among them, in the decision-making unit, the optimal batting strategy is analyzed and determined based on the output of the adaptive control algorithm; the results of the decision-making unit are converted into specific racket control instructions through the instruction generation unit; the control interface unit is responsible for transmitting the control instructions generated by the instructions to the racket on the serving and receiving equipment; in the execution unit, the angle, speed and power of the racket are actually adjusted according to the received control instructions; the feedback monitoring unit compares the actual execution results with the expected goals by monitoring the execution process of the racket; in the adjustment and optimization unit, the control instructions are adjusted and optimized according to the results of the feedback monitoring unit.
[0078] Use the model and logical judgment in the model creation module to evaluate the effects of different strategies and select the best hitting plan. Generate angle adjustment, speed control and power adjustment signals based on the optimal strategy, establish a connection with the racket hardware through serial communication and wireless communication technology, execute control instructions by controlling the racket on the serving and receiving equipment, and monitor the angle, speed and power of the racket through the monitoring module.
[0079] It should be understood that each part of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system.
[0080] Those skilled in the art will understand that all or part of the steps in the method for implementing the above-mentioned embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.
[0081] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or two or more units may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.
[0082] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.
Claims
1. A computer-based virtual table tennis control system, characterized in that: It includes monitoring module, ball serving and receiving equipment, data pre-processing module, model creation module, adaptive control algorithm module, and decision-making and execution module; The data preprocessing module receives the raw data collected by the monitoring module and cleans, denoises, and standardizes the collected data. The model creation module predicts the trajectory, speed, and rotation of the table tennis ball, as well as the optimal hitting position and angle of the racket. The adaptive control algorithm module calculates the optimal control instructions based on the prediction results of the deep learning model and real-time sensor data. In the model creation module, use PyTorch's nn module to define nn.Conv2d, nn.ReLU, nn.MaxPool2d, and nn.Linear. In PyTorch, use the cross entropy loss in the nn module and select the corresponding loss function. In PyTorch, use the optimizer in the optim module to specify hyperparameters. In the training loop, use PyTorch's forward and backward methods to calculate losses and gradients, and use the optimizer's step method to update model parameters. At the same time, use the validation set to verify model performance, load test data, set the model to evaluation mode, close the layers used during training, and then predict the test data. Calculate performance indicators, and use PyTorch's torch.save and torch.load functions to save and load model parameters. In the adaptive control algorithm module, real-time video streams are obtained through the monitoring module, and visual technology is used to identify and track the position, speed and direction of the player and the ball. The established model is used to analyze and predict the movement trajectory of the table tennis ball. At the same time, the player's historical batting data is combined to predict his possible reaction. Based on the speed, rotation, landing point of the ball and the player's reaction time, a rule-based control strategy is used to formulate a control strategy, and the action instructions are executed through the controller to ensure that the racket on the serving and receiving device can contact the ball at the best time and in the best condition. By creating a model module to analyze the batting feedback data, identify successful batting patterns and reasons for failure, and adjust the parameters and structure of the control algorithm based on the information.
2. The computer-based virtual table tennis control system according to claim 1, characterized in that: The monitoring module includes a high-precision camera, 3D lidar, accelerometer and gyroscope; Among them, a high-precision camera is installed above or on the side of the table, responsible for capturing the movement trajectory of the ping-pong ball and racket; a 3D lidar is installed around or above the table to measure the distance and speed between the ball and the racket; an accelerometer and gyroscope are installed on the handle of the table tennis racket. The accelerometer can measure the acceleration and tilt angle of the racket, while the gyroscope can measure the angular velocity and rotation direction of the racket.
3. The computer-based virtual table tennis control system according to claim 1, characterized in that: in, The data preprocessing module includes data collection and integration unit, data cleaning and denoising unit, data standardization and normalization unit, feature extraction unit, data enhancement and expansion unit, and data partitioning unit; The data collection and integration unit is responsible for collecting raw data from the monitoring modules and integrating this data into a unified format and framework. The data cleaning and denoising unit is used to remove errors, outliers, duplicate data, and sensor noise from the raw data. The data standard and normalization unit is used to convert the data of sensors and measurement units to a unified scale; the feature extraction unit is used to extract the speed, acceleration and rotation of table tennis control from the pre-processed data; The data enhancement and expansion unit increases the diversity and size of the data set through data enhancement technology; the data partitioning unit is used to divide the preprocessed data set into training set, validation set and test set; Specifically, data interfaces and communication protocols are used to receive data streams from the monitoring module, and data formats are converted and synchronized. Kalman filtering and threshold setting are used to identify and correct abnormal data points and reduce the impact of noise. The minimum-maximum normalization method is used to map the data to a specific range and eliminate dimensional differences. Feature engineering technology and technical support from professionals are used to construct a table tennis feature set. The technology is enhanced through rotation, translation, scaling, and noise injection, and the original data is transformed to generate new training samples. The data set is divided into different subsets through stratified sampling, and the data is saved and managed through a database.
4. The computer-based virtual table tennis control system according to claim 1, characterized in that: The model creation module includes the model definition unit, loss function definition unit, optimizer definition unit, model training unit, model evaluation and testing unit, and model saving and loading unit; Among them, the convolution layer, pooling layer, and fully connected layer in the architecture of the convolutional neural network are defined through the model definition unit, and the convolution kernel size, step size, and padding of each layer are specified; the loss function definition unit is used to define the loss function used to calculate the error between the predicted value and the true value during the model training process; the optimizer definition unit is used to define the stochastic gradient descent used to update the model parameters; the model training unit is used to iteratively train the model using training data, calculate the gradient and update the model parameters through the backpropagation algorithm; the model evaluation and testing unit is used to evaluate and test the trained model using test data; the model saving and loading unit is used to save the trained model parameters and load the saved model parameters for inference or continued training.
5. The computer-based virtual table tennis control system according to claim 1, characterized in that: The adaptive control algorithm module includes a state perception unit, a prediction and decision-making unit, a real-time adjustment execution unit, and a feedback and optimization unit; Among them, the state perception unit is used to collect and analyze the player's batting action, ball path changes and opponent's action information; the prediction decision unit predicts the future trajectory of the ball and the player's possible batting action based on the current state and historical data; the prediction decision unit formulates a control strategy for adjusting the racket angle, speed and power according to the prediction results and current state; the real-time adjustment execution unit converts the control strategy into specific action instructions, and adjusts the racket angle, speed and power in real time.
6. The computer-based virtual table tennis control system according to claim 1, characterized in that: The decision and execution module includes a decision unit, an instruction generation unit, a control interface unit, an execution unit, a feedback monitoring unit, and an adjustment and optimization unit; Use the model and logical judgment in the model creation module to evaluate the effects of different strategies and select the best hitting plan. Generate angle adjustment, speed control and power adjustment signals based on the optimal strategy, establish a connection with the racket hardware through serial communication and wireless communication technology, execute control instructions by controlling the racket on the serving and receiving equipment, and monitor the angle, speed and power of the racket through the monitoring module.