Table tennis drop point analysis and calculation system, method and device based on computer vision

Abstract

The invention discloses a table tennis drop point analysis and calculation system, method and device based on computer vision in the technical field of physical training and match analysis. The method comprises the following steps: acquiring a video picture of table tennis motion; detecting coordinate points of a table tennis ball in the video picture through a neural network; checking a neural network detection result, and judging whether coordinate points of the table tennis are successfully obtained or not; when the coordinate point of the table tennis ball is not obtained, detecting the outline of the table tennis ball in the video picture through an image processing method; when the coordinate point of the table tennis ball is obtained or the outline of the table tennis ball is detected, performing table tennis ball trajectory tracking by using Kalman filtering and an SAM2 segmentation algorithm; and determining a table tennis drop point through a machine learning method based on a trajectory tracking result. According to the method, the table tennis motion picture is collected and subjected to dynamic ISP processing, then the deep learning model is used for detecting the position of the table tennis, multi-ball independent recognition is achieved through the multi-ball segmentation tracking algorithm, and detection, tracking and drop point analysis of the table tennis are achieved.

Description

Technical Field

[0001] The invention relates to a table tennis ball landing point analysis and calculation system, method and device based on computer vision, and belongs to the technical field of sports training and game analysis. Background Art

[0002] As China's national sport, table tennis has become one of the most popular sports. With the popularity of table tennis and the continuous improvement of technical level, people's demand for table tennis training is also increasing. In daily training, the training method of controlling the landing point is widely used, because the accuracy of the landing point is directly related to the score and victory in the game. Therefore, the precise detection and analysis of the landing point of table tennis has gradually become the focus of technical research and development.

[0003] Traditional table tennis landing point analysis methods usually rely on manual records by coaches or athletes. This method is not only time-consuming, but also easily interfered by subjective factors, and it is difficult to meet modern efficient and accurate training needs. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a table tennis ball landing point analysis calculation system, method and device based on computer vision. The present invention overcomes the shortcomings of the prior art and significantly improves the real-time, accuracy and convenience of table tennis ball landing point analysis.

[0005] To achieve the above object, the present invention is implemented by adopting the following technical solutions: In a first aspect, the present invention provides a table tennis ball landing point analysis and calculation method based on computer vision, comprising: Get video footage of table tennis; Detect the coordinates of the table tennis ball in the video screen through a neural network; Check the results of the neural network detection to determine whether the coordinates of the ping-pong ball are successfully obtained; In response to the failure to obtain the coordinate points of the table tennis ball, detecting the outline of the table tennis ball in the video picture by an image processing method; In response to obtaining the coordinate points of the table tennis ball or detecting the outline of the table tennis ball, tracking the trajectory of the table tennis ball using Kalman filtering and SAM2 segmentation algorithm; Based on the results of trajectory tracking, the landing point of the table tennis ball is determined by machine learning methods.

[0006] Furthermore, the video images of the table tennis movement are acquired by a camera, and the camera dynamically adjusts image parameters in real time through a hardware dynamic ISP processor in combination with a computer vision algorithm, and performs automatic focusing by calculating the object distance between the table tennis ball and the camera during dynamic adjustment.

[0007] Furthermore, the formula for dynamic adjustment is: ; ;

[0008] Where: E is the exposure value; C is the contrast; v is the movement speed, the unit is m / s; and are initial exposure and contrast, respectively; and is the adjustment coefficient; The maximum speed supported by the camera, in m / s; Indicates the real-time change of the picture brightness; The calculation formula of the object distance is: ; in: is the focal length of the camera, in m, d is the object distance, in m; D is the actual diameter of the ping-pong ball, in m; is the diameter of the ping-pong ball in the image.

[0009] Furthermore, the outline of the table tennis ball in the video image is detected by an image processing method, including: Binarize the video image to distinguish the foreground and background; Remove noise in the foreground through erosion and dilation operations; The Canny algorithm is used to perform edge detection on the foreground from which noise has been removed to extract the outline of the ping-pong ball.

[0010] Furthermore, the trajectory of the table tennis ball is tracked by using the Kalman filter and the SAM2 segmentation algorithm, including: firstly, the accuracy of the SAM2 segmentation algorithm is optimized by defining a loss function, multiple targets in the video screen are segmented, and then the trajectory of the table tennis ball is tracked by using the Kalman filter, wherein: The loss function is: ; ; ; Where: L is the loss function; and are the loss weight coefficients respectively; and are the cross entropy loss and IOU loss; is the true label; is the category probability predicted by the model; C and D are the segmentation mask areas.

[0011] Furthermore, the Kalman filter is used to track the trajectory of the table tennis ball, including: predicting the position and speed of the table tennis ball, calculating the Kalman gain, and updating the state using the Kalman gain, wherein: The state prediction formula is: ; ; in: is the state prediction vector, containing the position of the ping-pong ball and speed ; is the system state estimate obtained by updating all observations at time k-1; A is the state transfer matrix, which represents the state update of the system; B is the control input matrix; is the control input, is the state covariance prediction matrix; is the error covariance matrix of the state estimate at time k-1; is the process noise covariance matrix, representing the noise impact of the prediction; The Kalman gain calculation formula is: ; in: is the Kalman gain; is the observation matrix, which maps the state space to the observation space; is the observation noise covariance matrix, reflecting the measurement error; The state update formula is: ; ; in: is the updated state estimate; is the actual observation value of the current frame, that is, the position of the ping-pong ball detected by the neural network; is the identity matrix; is the covariance matrix after update at time k.

[0012] Furthermore, the landing point of the table tennis ball is determined by a machine learning method, including: data preparation and annotation for model training, model selection and design for the regression model, model training of the regression model using training data, coordinate prediction through the trained model, and finally optimization and error correction. The regression model uses the mean square error as the loss function, and the expression is: ; Where: MSE is mean square error; is the number of samples in the batch; is the actual coordinate of sample i; The coordinates of sample i predicted by the model.

[0013] In a second aspect, the present invention provides a table tennis ball landing point analysis and calculation system based on computer vision, comprising: Image receiving module: used to obtain video images of table tennis; Neural network module: used to detect the coordinate points of the table tennis ball in the video screen through the neural network; Detection and judgment module: used to check the results of neural network detection and determine whether the coordinate points of the table tennis ball are successfully obtained; Image processing module: used for detecting the outline of the table tennis ball in the video picture by using an image processing method in response to failure to obtain the coordinate point of the table tennis ball; Trajectory tracking module: used for tracking the trajectory of the table tennis ball by using Kalman filtering and SAM2 segmentation algorithm in response to obtaining the coordinate points of the table tennis ball or detecting the outline of the table tennis ball; Landing point determination module: used to determine the landing point of the table tennis ball based on the results of trajectory tracking through machine learning methods.

[0014] In a third aspect, the present invention provides a table tennis ball landing point analysis and calculation device based on computer vision, comprising a processor and a storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of any of the methods described above.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the above methods when executed by a processor.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. When the present invention is deployed in a table tennis training field, a high-speed camera is used to collect table tennis motion images in real time. After dynamic ISP processing, a deep learning model is used to detect the position of the table tennis ball and a multi-ball segmentation and tracking algorithm is used to realize multi-ball independent recognition. The trajectory analysis and landing point determination module ensures the accuracy of the landing point from different perspectives with the support of multi-perspective data. The data is output after perspective transformation, and coaches or athletes can view the landing point information in real time through portable devices and quickly adjust training strategies.

[0017] Second, the present invention solves the problem of the dependence of the field layout on the camera's perspective through multi-perspective data collection and simulation training. The system can maintain high-precision landing point analysis at any camera position. When the camera cannot be set up at high altitude, the system uses simulation and multi-perspective data training to ensure that the coordinates of the table tennis ball's landing point can still be accurately determined in the picture taken by the coach or athlete's handheld device, providing users with greater freedom and flexibility.

[0018] 3. In each cycle of the Kalman filter, through the prediction and update steps, the system can gradually converge to the real trajectory of the table tennis ball, ensuring that when rapid changes occur in the trajectory of the table tennis ball, the system can still smoothly track the movement path of the table tennis ball. By introducing the combination of Kalman filtering and SAM2 segmentation algorithm, the present invention effectively improves the real-time tracking accuracy of table tennis balls in multi-ball scenarios, ensuring that even in the case of rapid movement and multiple target overlaps, the system can still accurately distinguish and track the independent trajectory of each ball. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings: Figure 1 A flowchart of a table tennis ball landing point analysis and calculation method based on computer vision provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0020] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0021] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise specified, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art to which the present invention belongs. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present invention.

[0022] Example 1: Please refer to Figure 1 The present invention provides a table tennis ball landing point analysis and calculation method based on computer vision, comprising: Step 1: Use a high-speed high-definition camera to capture table tennis video images. The system first uses a high-speed high-definition camera to capture table tennis video images, providing basic visual data input. The present invention uses an efficient hardware dynamic ISP (image signal processing) processor combined with advanced computer vision algorithms to adjust the camera's image parameters in real time to adapt to changing light environments and motion states. The dynamic ISP technology can adaptively adjust exposure, contrast, and white balance parameters according to the speed of table tennis, background color, and light changes, thereby improving picture clarity and detection accuracy, so that the system can still obtain high-definition table tennis images in complex environments, ensuring the stability and accuracy of the detection results.

[0023] During the dynamic ISP parameter adjustment process, automatic exposure and contrast adjustment are performed according to the real-time image content and motion speed. Assuming that the exposure value is E, the contrast is C, and the motion speed is v, the unit is m / s, the dynamic adjustment formula can be expressed as: ; ; in: and are initial exposure and contrast, respectively; and is the adjustment coefficient; The maximum speed supported by the camera; Indicates the real-time change in the brightness of the image. The adjustment algorithm responds in real time during the camera acquisition process to ensure the clarity of the table tennis ball under different lighting conditions.

[0024] In order to enhance the camera's autofocus capability, especially when the ISP is adjusted dynamically, autofocus can be achieved by calculating the distance between the ping-pong ball and the camera. Assuming that the position of the ping-pong ball in the image coordinates is (u, v) and the focal length of the camera is f, the estimation formula for the object distance d is: ; Where: D is the actual diameter of the table tennis ball, in meters; is the diameter of the ping-pong ball in the image, which is calculated through image processing.

[0025] Step 2: Use neural network detection to obtain the coordinate points of the ping pong ball in the video. The video is processed by the YOLOv8 deep learning neural network to detect and extract the coordinate points of the ping pong ball.

[0026] Step 3: Check whether the coordinate points are obtained. Check the results of the neural network detection to determine whether the coordinate points of the ping-pong ball are successfully obtained. If successful, directly enter the trajectory tracking stage; if unsuccessful, start the image processing method to detect the outline of the ping-pong ball.

[0027] Step 4: Start the image processing method to detect the outline of the ping-pong ball. When the neural network fails to detect the ping-pong ball, the system uses image processing technology to detect the outline, including: 1. Binarize the video image to distinguish the foreground and background; 2. Remove noise in the foreground through corrosion and dilation operations; 3. Use the Canny algorithm for edge detection to extract the outline of the ping-pong ball from the foreground; Step 5: Use Kalman filtering and SAM2 segmentation algorithm to track the trajectory of table tennis balls. Aiming at the actual needs of table tennis multi-ball training, the present invention adopts multi-target detection and tracking algorithms, and especially introduces advanced segmentation models such as SAM2 to cope with the tracking and differentiation problems in multi-ball scenarios. By performing multi-level segmentation and independent tracking of table tennis balls in the video screen, the system can accurately identify and distinguish the trajectories of multiple table tennis balls, effectively avoiding the tracking errors and missed detection problems caused by multi-ball overlap or cross-motion in other systems, so that coaches and athletes can also perform accurate drop point analysis in multi-ball training.

[0028] In multi-ball training and competition scenarios, in order to solve the problem of tracking and segmenting multiple table tennis balls, the present invention introduces the adaptive segmentation model SAM2 (Segment Anything Model 2). SAM2 is a target segmentation algorithm based on deep learning, which can accurately segment multiple targets in the video screen. It has the following technical features: Adaptive segmentation capability: The SAM2 model uses region extraction and object segmentation techniques to automatically identify multiple ping-pong balls in an image and assign unique labels. The model extracts features through deep convolutional neural networks (such as ResNet or Transformer architectures) to maintain independent segmentation and recognition of multiple balls in a dynamic background.

[0029] Real-time Update and Tracking: SAM2 can update the segmentation mask of each ping-pong ball in real time, maintaining the continuity of tracking in consecutive frames of the video. This allows the system to accurately distinguish the independent movement trajectory of each ball in a multi-ball scene.

[0030] Embedded Multi-Target Association: SAM2 uses an embedded multi-target association method to maintain independent tracking of each ping-pong ball even when the balls overlap and move at high speeds. It uses a matching mechanism based on Euclidean distance and spatial relationships to uniquely associate each target to the matching target in the adjacent frame.

[0031] When using the SAM2 segmentation algorithm, the accuracy of the segmentation model can be optimized by defining a loss function L. Commonly used loss functions include cross entropy loss and IOU loss (Intersection-over-union loss), the loss function L can be expressed as: ; ; ; in: and are the loss weight coefficients respectively; is the true label; is the category probability predicted by the model; C and D are segmentation mask areas. By combining the SAM2 algorithm, the present invention can achieve the effect of independent tracking of multiple balls in a multi-ball scene based on fine segmentation and efficient target tracking.

[0032] The Kalman filter is a recursive algorithm based on state estimation, which is suitable for tracking the motion trajectory of a target. In the table tennis trajectory tracking of the present invention, the Kalman filter is used to predict and correct the position and speed of the table tennis ball to maintain continuous tracking of fast-moving targets. The following are the detailed steps and formulas of the Kalman filter: a) State prediction The state prediction at time step k is given by: ; ; in: is the state prediction vector, containing the position of the ping-pong ball and speed ; is the system state estimate obtained by updating all observations at time k-1; A is the state transfer matrix, usually in the form of a unit matrix, representing the state update of the system; B is the control input matrix; is the control input, usually zero; is the state covariance prediction matrix; is the error covariance matrix of the state estimate at time k-1; is the process noise covariance matrix, representing the noise impact of the prediction.

[0033] b) Kalman gain calculation When an observation arrives, calculate the Kalman gain : ; in: is the observation matrix, which maps the state space to the observation space; is the observation noise covariance matrix, reflecting the measurement error.

[0034] c) Status Update The Kalman gain is used to update the state so that the predicted value is closer to the true value: ; ; in: is the updated state estimate; is the actual observation value of the current frame, that is, the position of the ping-pong ball detected by the neural network; is the identity matrix; is the covariance matrix after update at time k.

[0035] In each cycle of the Kalman filter, through the prediction and update steps, the system can gradually converge to the true trajectory of the ping-pong ball, ensuring that when rapid changes occur in the trajectory of the ping-pong ball, the system can still smoothly track the movement path of the ping-pong ball.

[0036] By introducing the combination of Kalman filtering and SAM2 segmentation algorithm, the present invention effectively improves the real-time tracking accuracy of table tennis in multi-ball scenarios, ensuring that the system can accurately distinguish and track the independent trajectory of each ball even in fast motion and overlapping of multiple targets.

[0037] Step 6: Determine the landing point of the table tennis ball through machine learning methods. The traditional landing point determination method relies on the sudden change of the slope of the motion trajectory, which is easy to produce errors in complex motion paths. The present invention uses machine learning methods to combine 20% of real training data with 80% of simulation data to train a multi-perspective and diversified data model for simulating the landing point characteristics at different angles. Through simulation and data expansion, the system of the present invention significantly improves the accuracy of landing point analysis. In particular, with the support of simulation data, the system can maintain high accuracy of landing point determination even in the face of unusual viewing angles and trajectory changes.

[0038] In the deep learning coordinate regression method, a convolutional neural network (CNN) model is used to predict the coordinates of the ping-pong ball on the actual table tennis table directly from the camera image. The key to this method is to build and train a regression model that can map the image input to the actual coordinates of the table tennis table, so as to achieve accurate landing point determination under different camera angles. The following is the detailed process of the coordinate regression method: 1. Data preparation and annotation First, a large amount of labeled data is needed to train the regression model so that it can accurately predict the landing point of the table tennis ball under different perspectives and backgrounds. Data preparation includes the following steps: Data collection: Collect table tennis motion videos under different camera angles, lighting conditions, and background changes. Both static and dynamic training data can be included to enhance the generalization ability of the model.

[0039] Data annotation: The collected images are annotated and the coordinates of the ping-pong ball in each image are converted to the actual coordinates of the landing point on the ping-pong table. For example, the actual coordinates corresponding to the position of the ping-pong ball in the image coordinate system are annotated in the dataset.

[0040] The labeled dataset should cover the positions of the ping-pong ball at various viewing angles and distances to ensure that the model can adapt to changes in real scenes.

[0041] 2. Model selection and design In coordinate regression, it is crucial to choose a suitable deep neural network model. Commonly used network architectures include lightweight convolutional neural networks such as ResNet, MobileNet, and EfficientNet, which have good performance in processing image feature extraction. The design of a coordinate regression model usually includes the following parts: Feature extraction module: Use pre-trained CNN to extract visual features of the table tennis ball, such as color, edge, and contour information.

[0042] Fully connected layer: The extracted features are input into the fully connected layer for regression calculation, and the actual coordinates of the ping-pong ball are directly output.

[0043] Loss function: The mean square error (MSE) is used as the loss function to measure the error between the predicted coordinates and the actual coordinates, which is defined as: ; Where: MSE is mean square error; is the number of samples in the batch; is the actual coordinate of sample i; The coordinates of sample i predicted by the model. This MSE-based loss function is very common in regression tasks and helps the model gradually converge to accurate coordinate predictions.

[0044] 3. Model Training During the model training phase, a large amount of training data is used to optimize the model to ensure its robustness in a variety of scenarios. The training process includes the following steps: Data augmentation: The dataset is expanded through data augmentation techniques such as random rotation, scaling, flipping, brightness adjustment, etc. to increase the generalization ability of the model.

[0045] Training strategy: Use an adaptive learning rate optimization algorithm (such as Adam or SGD) to update model parameters to speed up training and prevent overfitting.

[0046] Validation and tuning: Test the model on the validation set to observe how the model performs on unseen data, and optimize model performance by adjusting hyperparameters (such as learning rate, number of network layers, etc.).

[0047] During the training process, through multiple iterations and loss reduction, the model will gradually learn to map image coordinates from different perspectives to actual table tennis table coordinates.

[0048] 4. Coordinate prediction After the model training is completed, the image of the ping-pong ball taken by the camera is input into the regression model, and the model will directly output the actual coordinates of the ping-pong ball on the ping-pong table. The specific process of coordinate prediction is as follows: Image preprocessing: Scale the images captured by the camera to a size suitable for model input (such as 224×224 pixels) and normalize them to match the data distribution during training.

[0049] Feature extraction and regression: The model extracts the visual features of the ping-pong ball through the convolutional layer, performs coordinate regression in the fully connected layer, and outputs the actual coordinates of the ping-pong ball.

[0050] Real-time prediction: During the table tennis game, the coordinates of the image can be predicted every other frame to achieve real-time monitoring of the landing point.

[0051] 5. Optimization and Error Correction To improve the accuracy of coordinate predictions, optimization and error correction can be performed in the post-training stage: Sliding window averaging: Perform sliding window averaging on multiple prediction results over time to reduce errors caused by jitter or noise.

[0052] Fine-tuning the model: If the actual usage scenario is different from the training data, a small amount of field data can be used to fine-tune the model to further improve the prediction accuracy.

[0053] Advantages and application scenarios The advantage of deep learning coordinate regression is that it does not require perspective transformation matrix or additional geometric calculations, and can directly predict the coordinates of the table tennis table in a data-driven way. This method is particularly suitable for the following scenarios: Camera positions for different perspectives: The model automatically adapts to different camera angles and perspective changes without the need for complex perspective corrections.

[0054] Multi-site adaptability: The model trained based on data can be used in a variety of environments and has good robustness to multiple backgrounds and different lighting conditions.

[0055] Through deep learning coordinate regression, the landing point of the table tennis ball can be directly and quickly mapped to the actual field coordinates, which is suitable for training and game analysis scenarios with high real-time requirements.

[0056] The present invention solves the problem of the dependence of the field layout on the camera's perspective through multi-perspective data collection and simulation training. The system can maintain high-precision landing point analysis at any camera position. When the camera cannot be set up at high altitude, the system ensures that the coordinates of the table tennis ball's landing point can still be accurately determined in the picture taken by the coach or athlete's handheld device through simulation and multi-perspective data training, providing users with greater freedom and flexibility.

[0057] In order to meet the needs of rapid deployment in different scenarios, the present invention adopts an integrated miniaturized design, integrating the camera, data processing unit and display device into one, with small size, light weight, and easy to carry and move. This portable integrated design not only reduces the deployment and use costs of the system, but also provides great convenience for coaches and athletes to quickly set up and use it in different training venues.

[0058] In actual use, the system is deployed in a table tennis training field, and the table tennis sports images are collected in real time through a high-speed camera. After dynamic ISP processing, the system uses a deep learning model to detect the position of the table tennis ball and realizes independent identification of multiple balls through a multi-ball segmentation and tracking algorithm. The trajectory analysis and landing point determination module ensures the accuracy of the landing point from different perspectives with the support of multi-view data. The data is output after perspective transformation, and coaches or athletes can view the landing point information in real time through portable devices and quickly adjust training strategies.

[0059] Compared with the prior art, the present invention has the following advantages: 1. High-precision dynamic ISP processing: Even under complex lighting conditions, it can effectively improve image quality and ensure the accuracy of table tennis detection.

[0060] 2. Multi-ball tracking capability: Supports multi-ball training scenarios and avoids multi-ball tracking confusion and missed detection problems through the SAM2 segmentation model.

[0061] 3. High-precision landing point analysis: Through machine learning and a large amount of simulation data, the landing point accuracy under multiple perspectives is optimized, and the system's judgment ability is improved.

[0062] 4. Applicability to any camera position: The system can adapt to various camera positions and can achieve accurate landing point determination without the need for high-altitude equipment.

[0063] 5. Portability and easy deployment: The compact integrated design is suitable for rapid deployment to meet the needs of different sites and applications.

[0064] In summary, the present invention has significant advantages in the real-time, accuracy and convenience of table tennis landing point detection, and can be widely used in various scenarios such as table tennis training and game analysis, providing effective technical support for the precise training and scientific analysis of table tennis.

[0065] The present invention has the following application scenarios: 1. Physical training Smart training system: This technology can record the trajectory and landing point of table tennis balls in real time and perform automated analysis to help athletes and coaches get quick feedback. This is crucial for the training of table tennis players and can be used to evaluate the accuracy, speed and landing position of the ball, helping athletes optimize their movements and techniques.

[0066] Virtual coach: By integrating this technology into sports training systems, athletes can get personalized training advice through the virtual coaching system to improve training efficiency.

[0067] 2. Live broadcast and analysis of the game Real-time game analysis: This technology can be applied to live broadcasts of table tennis matches, helping viewers understand the ball's trajectory, landing point, and hitting direction in real time, improving the viewing experience. Combined with data analysis, it can also provide rich statistical data and technical analysis, such as the hitting speed and trajectory length of each round.

[0068] Automated referee system: Through high-precision landing point analysis, the system can make automatic judgments during the game and reduce the chance of manual misjudgment, which will greatly improve the fairness and efficiency of table tennis games.

[0069] 3. Sports science research Motion trajectory analysis: This technology can be used in sports science research to help researchers accurately analyze the motion trajectory, collision mechanics and landing point distribution of table tennis balls. This has important application value in understanding the technical characteristics of players, evaluating training effects and optimizing equipment design.

[0070] Data-driven decision support: Through long-term data accumulation, the system can provide scientific decision support for coaches and athletes, optimize training plans, and enhance athlete performance.

[0071] The invention has a wide range of application markets, covering many aspects from table tennis training, game analysis to audience experience improvement. With the continuous development of sports events, the demand for real-time data analysis and intelligent systems is increasing, which provides broad prospects for the application of this technology in the global sports market.

[0072] Embodiment 2: A table tennis ball landing point analysis and calculation system based on computer vision, which can implement the table tennis ball landing point analysis and calculation method based on computer vision described in Embodiment 1, including: Image receiving module: used to obtain video images of table tennis; Neural network module: used to detect the coordinate points of the table tennis ball in the video screen through the neural network; Detection and judgment module: used to check the results of neural network detection and determine whether the coordinate points of the table tennis ball are successfully obtained; Image processing module: used for detecting the outline of the table tennis ball in the video picture by using an image processing method in response to failure to obtain the coordinate point of the table tennis ball; Trajectory tracking module: used for tracking the trajectory of the table tennis ball by using Kalman filtering and SAM2 segmentation algorithm in response to obtaining the coordinate points of the table tennis ball or detecting the outline of the table tennis ball; Landing point determination module: used to determine the landing point of the table tennis ball based on the results of trajectory tracking through machine learning methods.

[0073] Embodiment 3: The embodiment of the present invention further provides a table tennis ball landing point analysis and calculation device based on computer vision, which can implement the table tennis ball landing point analysis and calculation method based on computer vision described in Embodiment 1, including a processor and a storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to perform the steps of the following method: Get video footage of table tennis; Detect the coordinates of the table tennis ball in the video screen through a neural network; Check the results of the neural network detection to determine whether the coordinates of the ping-pong ball are successfully obtained; In response to the failure to obtain the coordinate points of the table tennis ball, detecting the outline of the table tennis ball in the video picture by an image processing method; In response to obtaining the coordinate points of the table tennis ball or detecting the outline of the table tennis ball, tracking the trajectory of the table tennis ball using Kalman filtering and SAM2 segmentation algorithm; Based on the results of trajectory tracking, the landing point of the table tennis ball is determined by machine learning methods.

[0074] Embodiment 4: The embodiment of the present invention further provides a computer-readable storage medium, which can implement the table tennis ball landing point analysis and calculation method based on computer vision described in Embodiment 1, and stores a computer program thereon, which implements the steps of the following method when the program is executed by a processor: Get video footage of table tennis; Detect the coordinates of the table tennis ball in the video screen through a neural network; Check the results of the neural network detection to determine whether the coordinates of the ping-pong ball are successfully obtained; In response to the failure to obtain the coordinate points of the table tennis ball, detecting the outline of the table tennis ball in the video picture by an image processing method; In response to obtaining the coordinate points of the table tennis ball or detecting the outline of the table tennis ball, tracking the trajectory of the table tennis ball using Kalman filtering and SAM2 segmentation algorithm; Based on the results of trajectory tracking, the landing point of the table tennis ball is determined by machine learning methods.

[0075] It is known from common technical knowledge that the present invention can be implemented by other embodiments that do not deviate from its spirit or essential features. Therefore, the above disclosed embodiments are only illustrative in all respects and are not exclusive. All changes within the scope of the present invention or within the scope equivalent to the present invention are included in the present invention.

[0076] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0077] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that specifies the functions of a block or multiple blocks.

[0078] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0079] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A table tennis ball landing point analysis and calculation method based on computer vision, characterized in that: include: Get video footage of table tennis; Detect the coordinates of the table tennis ball in the video screen through a neural network; Check the results of the neural network detection to determine whether the coordinates of the ping-pong ball are successfully obtained; In response to the failure to obtain the coordinate points of the table tennis ball, detecting the outline of the table tennis ball in the video picture by an image processing method; In response to obtaining the coordinate points of the table tennis ball or detecting the outline of the table tennis ball, tracking the trajectory of the table tennis ball using Kalman filtering and SAM2 segmentation algorithm; Based on the results of trajectory tracking, the landing point of the table tennis ball is determined by machine learning methods; The video images of the table tennis movement are acquired by a camera, which uses a hardware dynamic ISP processor in combination with a computer vision algorithm to dynamically adjust image parameters in real time, and automatically focuses by calculating the object distance between the table tennis ball and the camera during dynamic adjustment; the formula for the dynamic adjustment is: ； ； Where: E is the exposure value; C is the contrast; v is the movement speed, the unit is m / s; and are initial exposure and contrast, respectively; and is the adjustment coefficient; The maximum speed supported by the camera, in m / s; Indicates the real-time change of the picture brightness; The calculation formula of the object distance is: ； in: is the focal length of the camera, in m, d is the object distance, in m; D is the actual diameter of the ping-pong ball, in m; is the diameter of the ping-pong ball in the image.

2. The table tennis ball landing point analysis and calculation method based on computer vision according to claim 1, characterized in that: The outline of the table tennis ball in the video image is detected by an image processing method, including: Binarize the video image to distinguish the foreground and background; Remove noise in the foreground through erosion and dilation operations; The Canny algorithm is used to perform edge detection on the foreground from which noise has been removed, and the outline of the ping-pong ball is extracted.

3. The table tennis ball landing point analysis and calculation method based on computer vision according to claim 1, characterized in that: Using Kalman filter The method uses a SAM2 segmentation algorithm to track the trajectory of a table tennis ball, including: firstly, optimizing the accuracy of the SAM2 segmentation algorithm by defining a loss function, segmenting multiple targets in a video screen, and then using a Kalman filter to track the trajectory of the table tennis ball, wherein: The loss function is: ； ； ； Where: L is the loss function; and are the loss weight coefficients respectively; and are the cross entropy loss and IOU loss; is the true label; is the category probability predicted by the model; C and D are the segmentation mask areas.

4. The table tennis ball landing point analysis and calculation method based on computer vision according to claim 3 is characterized in that: Using Kalman filter to track the trajectory of table tennis ball includes three steps: predicting the position and speed of table tennis ball, calculating Kalman gain, and updating the state using Kalman gain, among which: The state prediction formula is: ； ； in: is the state prediction vector, containing the position of the ping-pong ball and speed ; is the system state estimate obtained by updating all observations at time k-1; A is the state transfer matrix, which represents the state update of the system; B is the control input matrix; is the control input; is the state covariance prediction matrix; is the error covariance matrix of the state estimate at time k-1; is the process noise covariance matrix, representing the noise impact of the prediction; The Kalman gain calculation formula is: ； in: is the Kalman gain; is the observation matrix, which maps the state space to the observation space; is the observation noise covariance matrix, reflecting the measurement error; The state update formula is: ； ； in: is the updated state estimate; is the actual observation value of the current frame, that is, the position of the ping-pong ball detected by the neural network; is the identity matrix; is the covariance matrix after update at time k.

5. The table tennis ball landing point analysis and calculation method based on computer vision according to claim 1, characterized in that: The method of determining the landing point of a table tennis ball by machine learning includes: data preparation and annotation for model training, model selection and design for regression model, model training of regression model using training data, coordinate prediction through the trained model, and finally optimization and error correction. The regression model uses mean square error as the loss function, and the expression is: ； Where: MSE is mean square error; is the number of samples in the batch; is the actual coordinate of sample i; The coordinates of sample i predicted by the model.

6. A table tennis ball landing point analysis and calculation system based on computer vision, based on the table tennis ball landing point analysis and calculation method based on computer vision according to any one of claims 1 to 5, characterized in that: include: Image receiving module: used to obtain video images of table tennis; Neural network module: used to detect the coordinate points of the table tennis ball in the video screen through the neural network; Detection and judgment module: used to check the results of neural network detection and determine whether the coordinate points of the table tennis ball are successfully obtained; Image processing module: used for detecting the outline of the table tennis ball in the video picture by using an image processing method in response to failure to obtain the coordinate point of the table tennis ball; Trajectory tracking module: used for tracking the trajectory of the table tennis ball by using Kalman filtering and SAM2 segmentation algorithm in response to obtaining the coordinate points of the table tennis ball or detecting the outline of the table tennis ball; Landing point determination module: used to determine the landing point of the table tennis ball based on the results of trajectory tracking through machine learning methods.

7. A table tennis ball landing point analysis and calculation device based on computer vision, characterized in that: including processor and storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 5 are implemented.

Images