Pick ball motion trail tracking and tactical decision auxiliary system

Abstract

The invention discloses a Pick ball motion trail tracking and tactical decision-making auxiliary system which adopts a high-frame-rate camera, a micro sensor installed on a ball and an athlete motion sensor worn on an athlete to respectively collect video image data, ball motion data and athlete motion data in the Pick ball motion process. Precise positioning and trajectory tracking are carried out on a ball through a target detection and tracking algorithm based on deep learning, a tactical knowledge model is constructed by combining site information and athlete action data and applying a decision tree and a neural network algorithm, real-time tactical decision suggestions are provided for athletes and coaches, and the training efficiency is improved. And the competitive level and the training effect of the Pick ball sports can be improved.

Description

[Technical field]

[0001] The present invention relates to the field of sports assistance technology, and in particular to a pickleball motion trajectory tracking and tactical decision-making assistance system. [Background Technology]

[0002] In the field of sports science, the study of ball sports trajectories has always been an important topic of concern. It is of key significance for in-depth understanding of the physical nature of ball sports and improving the competitive level of athletes. Many studies at home and abroad have achieved fruitful results in the study of trajectories of ball sports such as table tennis, football, golf, tennis, and volleyball. In the early days, the influence of basic forces such as gravity and air resistance on ball flight was mainly explored based on the principles of classical mechanics. As the research deepened, it gradually focused on factors such as ball rotation and the Magnus effect (Asai et al., 2020; Ding et al., 2023; Kajiyama, 2024; Tamaki et al., 2024). In terms of research methods, it has gradually expanded from theoretical analysis to various means such as experimental research and computer simulation. For example, by establishing a mechanical model to reveal the mechanism of force on the ball's trajectory, radar testing, computational fluid dynamics (CFD), and computer vision technology are used to measure and simulate the ball's trajectory (Mehta et al., 2001; Lynn et al., 2021; Cai, 2022; Tamada et al., 2020; Zhu et al., 2022; Sohara et al., 2022).

[0003] However, existing research has certain limitations. On the one hand, in order to facilitate calculation and analysis, theoretical models often simplify the actual situation and ignore factors such as ball surface roughness and air turbulence, resulting in deviations between the model and reality. On the other hand, although CFD and computer vision technologies are widely used, their scalability in different ball sports is limited. For example, the impact of complex environments (such as differences in field materials) on ball trajectories has not been studied in depth.

[0004] At the same time, tactical analysis also occupies a core position in sports data research, especially in racket-and-net events, which is crucial for athletes to develop winning strategies. Although related studies use visualization to display tactical usage rates, scoring rates, and design analysis frameworks, most of them use two-dimensional visualization, which has problems such as lack of three-dimensional perception of batting attributes, easy visual confusion, and overlapping and interweaving of key features (such as some table tennis and tennis tactical analysis studies). Although there are attempts to apply immersive technology, they tend to be simple visualization of a few cases rather than large-scale data pattern analysis (such as some basketball tactical studies).

[0005] In recent years, there have been new developments in badminton tactical analysis, such as using virtual reality technology to explore the laws of three-dimensional trajectories, but there are still deficiencies such as the failure to reveal the correlation between continuous shot sequences. However, the emerging sport of pickleball has relatively little research on its motion trajectory, and there is currently no complete system for tracking the motion trajectory of pickleball and assisting in tactical decision-making. The present invention aims to fill this gap by integrating advanced technical means to achieve accurate tracking of the motion trajectory of pickleball and scientific assistance in tactical decision-making, overcome the many deficiencies of existing research in ball motion trajectory and tactical analysis, and provide strong support for the development of pickleball.

[0006] References:

[0007] Asai, Takeshi, Kaoru Kimachi, Richong Liu, Masaaki Koido, Masao Nakayama, John Eric Goff, and Sungchan Hong. 2020. "Measurements of the Flight Trajectory of a Spinning Soccer Ball and the Magnus Force Acting on It" Proceedings 49, no.1:88. https: / / doi.org / 10.3390 / proceedings2020049088Cai,GL2022. A method for predicting the trajectory of table tennis in multirotation state based on binocular vision. Computational Intelligence and Neuroscience, 2022(1),8274202.

[0008] Ding,Z.,Jiang,S.,Wang,X.,&Wei,Z.(2023).The Influence of Magnus Effect on The Flying Distance of Baseball and Golf.Highlights in Science,Engineeringand Technology,38,821-828.

[0009] Kajiyama,Y.(2024).Analytic solutions for drag and Magnusforces.EuropeanJournal of Physics,45(5),055002.

[0010] Lynn,L.P.,Hendricks,K.D.,Houser,A.,McCullers,L.,Hughes,J.,Dunning,D.W.,Soto,G.,Britton,Z.,Mills,A.,Turnbull,S.D.,Keller,B.G.,Civale,A.,Kintzler,B.,Weber,R.,Woodruff,B.,&Corbin,P.(2021).Analyzing the SideForce ona Baseball Using Hawk-Eye Measurements.https: / / api.semanticscholar.org / CorpusID:235641247

[0011] Mehta,R.D.,&Pallis,J.M.(2001).Sports ball aerodynamics:effectsofvelocity,spin and surface roughness.Minerals,Metals and MaterialsSociety / AIME,Materials and Science in Sports(USA),185-197.

[0012] Sohara,K.,&Sugaya,Y.(2022).3-D Tennis Ball Trajectory EstimationfromMulti-view Videos and Its Parameterization by Parabola Fitting.InVISIGRAPP(5:VISAPP)(pp.769-776).

[0013] Tamada, Chandan Varma & Sabbavarapu, Sai Ganesh. (2020). Flow Visualization of Footballs to Analyze the Factors Affecting their Aerodynamic Performance Using CFD. 5.462 - 475. 10.38124 / IJISRT20SEP296.

[0014] Tamaki, S., Yamagata, S., & Hashizume, S. (2024). Spin measurement system for table tennis balls based on asynchronous non - high - speed cameras. International Journal of Computer Science in Sport, 23(1), 37 - 53.

[0015] Zhu, J., Lin, J., Yin, B., & Pan, Z. (2022, June). Trajectory reconstruction system of moving target applied in volleyball match. In 2022 2nd International Conference on Information Technology and Contemporary Sports (TCS) (pp. 6 - 9). IEEE.

[0016] Comparative documents:

[0017] CN 110727826 A, 2020.01.24

[0018] CN 111950407 A, 2020.11.17

[0019] CN 106310660 A, 2017.01.11

[0020] CN 109091837 A, 2018.12.28

[0021] CN 109499068 A, 2019.03.22

[0022] CN 110383324 A, 2019.10.25

[0023] CN 112085761 A, 2020.12.15

[0024] CN 207895727 U, 2018.09.21

[0025] US2009275371 A1, 2009.11.05

[0026] US2016342318 A1, 2016.11.24

[0027] US2021069574 A1, 2021.03.11

[0028] US2021142066 A1, 2021.05.13

[0029] Ye Shuannan et al. A review of immersive visualization. Journal of Computer-Aided Design and Computer Graphics. 2021, Vol. 33 (Issue 4), pp. 497-507.

[0030] CN 113610010 B (An immersive badminton tactics visual analysis system) [Summary of the invention]

[0031] The present invention overcomes the shortcomings of the prior art and provides a pickle ball motion trajectory tracking and tactical decision-making assistance system to achieve the functions of pickle ball motion trajectory tracking and tactical decision-making suggestion.

[0032] To achieve the above object, the present invention adopts the following technical solutions:

[0033] A pickleball motion trajectory tracking and tactical decision-making assistance system, characterized in that: it includes

[0034] A data acquisition module is used to collect video image data, ball movement data and athlete action data during the ball movement process;

[0035] The trajectory tracking module is connected to the data acquisition module. Based on the deep learning target detection and tracking algorithm, it detects and identifies the ball in the video image, predicts and corrects the position of the ball, and obtains the ball motion trajectory data;

[0036] The data integration module is connected to the data acquisition module and the trajectory tracking module respectively, and is used to integrate the ball motion trajectory data, the field information in the video image data and the player action data to obtain comprehensive data;

[0037] The tactical decision module is connected to the data integration module and is used to classify tactics according to the comprehensive data. At the same time, a prediction model is used to predict the tactical effect with tactical features as input and game results as output to obtain tactical decision suggestions;

[0038] The output module is connected to the tactical decision module and is used to output the tactical decision suggestions in a visual manner.

[0039] The pickle ball motion trajectory tracking and tactical decision-making support system as described above is characterized in that: the data acquisition module includes a camera for collecting video image data, a pickle ball motion sensor installed on the pickle ball to collect the ball's motion data, and an athlete motion sensor worn on the athlete to collect the athlete's motion data.

[0040] The pickle ball motion trajectory tracking and tactical decision-making support system as described above is characterized in that: a convolutional neural network is used in the trajectory tracking module to detect and identify the ball in the video image; and a Kalman filter algorithm is used in the trajectory tracking module to predict and correct the position of the ball.

[0041] The pickleball motion trajectory tracking and tactical decision-making support system as described above is characterized in that the convolutional neural network uses Faster R-CNN or YOLO algorithm for target detection.

[0042] The pickle ball motion trajectory tracking and tactical decision-making support system as described above is characterized in that a Kalman filter algorithm predicts the position of the ball at the next moment based on the position and speed of the ball at the previous moment, and makes corrections based on the current detection results.

[0043] The pickleball motion trajectory tracking and tactical decision-making support system as described above is characterized in that: the field information in the data integration module includes net height information and non-volley zone rule information, and the athlete action data includes athlete hitting action data, moving pace data and body posture data.

[0044] The pickle ball motion trajectory tracking and tactical decision-making support system as described above is characterized in that: a decision tree algorithm is used in the tactical decision module to construct a tactical classification model to classify tactics according to comprehensive data; a neural network algorithm is used in the tactical decision module to construct a tactical effect prediction model to predict tactical effects.

[0045] The pickle ball motion trajectory tracking and tactical decision-making support system as described above is characterized in that the decision tree algorithm selects the optimal features for node splitting by calculating information gain or Gini index to construct a tactical classification model.

[0046] The pickleball motion trajectory tracking and tactical decision-making support system as described above is characterized in that the neural network algorithm adopts a multi-layer perceptron and optimizes the model weights and biases through a back-propagation algorithm.

[0047] A pickleball motion trajectory tracking and tactical decision-making assistance system as described above is characterized in that: tactical decision suggestions include recommended tactical types, tactical implementation timing and expected tactical effects; visualization methods include superimposing tactical prompt information on the video screen and generating tactical analysis charts or reports.

[0048] The beneficial effects of the present invention are:

[0049] The present invention integrates ball motion trajectory data, video image data and athlete action data, and relies on advanced technical means and scientific algorithm models such as decision tree algorithm to build a tactical classification model and neural network algorithm to build a tactical effect prediction model, so as to achieve accurate tracking of the pickle ball motion trajectory, deeply integrate multi-source data such as venue information and athlete motion data, and then output accurate, real-time and highly targeted tactical decision-making suggestions for athletes and coaches, ultimately achieving a significant improvement in the competitive level and training effect of pickle ball. [Drawings]

[0050] Figure 1 It is a schematic diagram of the model structure of the present invention. [Specific implementation method]

[0051] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings.

[0052] It should be noted that all directional indications (such as up, down, left, right, front, back...) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly. In addition, the descriptions of "preferred", "sub-preferred", etc. in the present invention are only used for descriptive purposes, and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined as "preferred" or "sub-preferred" may explicitly or implicitly include at least one such feature.

[0053] like Figure 1 As shown, a pickleball motion trajectory tracking and tactical decision-making assistance system includes

[0054] The data acquisition module is used to collect video image data, ball motion data and athlete motion data during the pickleball movement process. The data acquisition module includes a camera that collects video image data, a pickleball motion sensor installed on the pickleball to collect the ball motion data, and an athlete motion sensor worn on the athlete to collect the athlete motion data. The pickleball motion sensor is a micro sensor.

[0055] Preferably, in the deployment and setting of high frame rate cameras, the key point of equipment selection is that when selecting a high frame rate camera, it is necessary to focus on its frame rate, resolution and ability to cope with different lighting conditions. In terms of frame rate, select a camera with a frame rate of no less than 240 frames per second to ensure that it can capture subtle changes in the high-speed movement of the pickleball, such as the deformation of the ball at the moment of rapid hitting and the rapid turning during flight. The resolution is set to 3840×2160 or higher, which helps to more clearly distinguish the relative position relationship between the ball and the field, the athlete, and the texture characteristics of the ball surface. For adaptability to lighting conditions, select a camera with a high dynamic range (HDR) function, such as a sports camera of a well-known brand, which can accurately capture the movement details of the ball in different lighting environments such as direct strong light (such as outdoor sunny game scenes) or shadow areas (such as corners of the field), effectively reduce the image overexposure or dark phenomenon caused by lighting problems, and ensure the stability and clarity of image quality.

[0056] The venue layout strategy is to arrange the camera positions reasonably according to the characteristics of the pickleball court, especially the standard doubles court. Placing cameras at the four corners of the court can fully cover the horizontal and vertical range of the court, ensuring that the movement trajectory of the ball in the edge area of ​​the court is captured. At the same time, a camera is installed at the midpoint of each of the two sidelines, which can better focus on the movement of the ball in the center area of ​​the court and effectively make up for the shooting blind spots that may exist in the center of the court for the corner camera. For example, when the ball is interacting with a fast short-distance hit in the center area of ​​the court, the camera at the midpoint of the sideline can provide a clearer and more accurate perspective. The camera is connected to the local server through a high-speed network cable (such as Category 6 and above Ethernet cable). During the connection process, it is necessary to set appropriate network transmission parameters such as IP address allocation, subnet mask, gateway settings, etc. according to the transmission requirements of the camera and the receiving capacity of the server to ensure that the video image data can be transmitted to the trajectory tracking module in real time and stably, avoiding jams or interruptions during data transmission, and providing a reliable data source for subsequent real-time analysis of the ball's movement trajectory.

[0057] The calibration process is that after the camera is installed, it must be accurately calibrated. First, adjust the focal length. According to the size of the venue and the shooting distance, set the focal length of the camera to clearly capture the movement range of the ball. You can place test balls at different locations in the venue to observe whether the ball in the camera image is clear, and adjust the focal length repeatedly until the best effect is achieved. In terms of viewing angle correction, use professional calibration tools (such as laser levels) to ensure that the camera's shooting angle is horizontal and vertical to the plane of the venue to avoid tilting or distorting the image, which affects the position judgment of the ball. Color balance calibration requires testing under different lighting conditions (such as natural light during the day, evening afterglow, indoor lighting, etc.), adjusting the color parameters of the camera so that the color of the captured image matches the actual scene, ensuring that the color of the ball and the color of the venue markings can be accurately presented, so that the subsequent ball recognition algorithm based on color features can run effectively.

[0058] Preferably, when installing and transmitting data on the micro sensor on the ball, the sensor selection is to select a sensor combination that combines a high-precision MEMS inertial measurement unit (IMU) with a micro GPS module, and ensure that its measurement accuracy meets the needs of pickleball motion analysis. The IMU's acceleration measurement accuracy must reach ±0.01g or above, and the angular velocity measurement accuracy must reach ±0.1° / s or above, in order to accurately capture the acceleration and rotation of the ball during movement. The positioning accuracy of the micro GPS module should be within 0.1 meters, and it can accurately provide the position information of the ball in outdoor venues. The sensor is firmly fixed in the ball, while avoiding adding extra weight to the ball or changing the distribution of the ball's center of gravity, to ensure that the ball's motion performance after the sensor is installed is basically the same as when it is not installed, and does not affect the athlete's normal hitting and the ball's flight trajectory.

[0059] The communication setting and data encryption measures are that the sensor and the data integration module communicate wirelessly through low-power Bluetooth 5.0 technology. In the communication setting process, in addition to setting the transmission frequency to 20 times per second, it is also necessary to optimize the transmission power and receiving sensitivity of the Bluetooth signal to ensure that the communication between the sensor and the data integration module is stable and reliable during the high-speed movement of the ball. For example, according to the size of the venue and the signal interference, the transmission power is adjusted within a suitable range (such as between 0-10dBm) to ensure the timeliness and continuity of data transmission, while avoiding signal interference with other devices or interference by other devices. To ensure the security of data transmission, the AES (Advanced Encryption Standard) encryption algorithm is used to encrypt the transmitted data. At the sensor end, the collected data is encrypted before being sent; at the data integration module end, the received data is decrypted. Only devices with the correct decryption key can interpret the data, effectively preventing the data from being stolen or tampered with during transmission, and ensuring the integrity and confidentiality of the data.

[0060] Preferably, in the collection of athlete motion data, the selection and wearing method of the sensor is to equip the athlete with a high-precision motion sensor (such as a nine-axis sensor), and the measurement accuracy and sampling frequency of the sensor must meet the requirements of pickleball sports analysis. The acceleration measurement accuracy reaches ±0.05g or more, the angular velocity measurement accuracy reaches ±0.1° / s or more, and the sampling frequency is not less than 100Hz, ensuring that the athlete's posture changes and movement details during rapid movement can be accurately captured. In terms of wearing method, the sensors are fixed to the athlete's wrist, elbow, shoulder, waist, knee, ankle and other key parts, and an adjustable elastic bandage or a special fixing clamp is used to ensure that the sensor is firmly worn during exercise and will not be displaced or fall off due to the athlete's strenuous exercise. At the same time, it is necessary to ensure that the sensor is worn accurately and fits closely to the joints of the human body. For example, the wrist sensor should be worn about 2-3 cm above the wrist joint to accurately measure the wrist motion parameters.

[0061] The trajectory tracking module is connected to the data acquisition module. Based on the deep learning target detection and tracking algorithm, it detects and identifies the ball in the video image, and predicts and corrects the position of the ball to obtain the ball motion trajectory data. Among them, the trajectory tracking module uses a convolutional neural network to detect and identify the ball in the video image. The convolutional neural network uses the Faster R-CNN or YOLO algorithm for target detection; the trajectory tracking module predicts and corrects the position of the ball through the Kalman filter algorithm. The Kalman filter algorithm predicts the position of the ball at the next moment based on the position and speed of the ball at the previous moment, and corrects it in combination with the current detection result.

[0062] Preferably, in the training and optimization of the CNN convolutional neural network model, the training set construction and preprocessing is to collect a large amount of video image data containing various pickleball sports scenes to construct a training set. The data sources include different levels of competitions (such as amateur competitions, professional leagues, etc.), different venue types (such as indoor wooden floor venues, outdoor plastic venues, etc.), different weather conditions (such as sunny days, cloudy days, rainy days, etc.) and different athlete levels (beginners, intermediate players, advanced players) Competition videos. In the collection process, ensure the quality and diversity of the video image, covering various batting methods (such as forehand shots, backhand shots, high-distance shots, flat shots, cutting balls, etc.) and the movement state of the ball (such as high-speed flight, low-speed arc balls, rapidly falling balls, etc.). The collected video image data is preprocessed, first of all, the image is cropped to remove irrelevant background areas in the video (such as the audience seats, the environment outside the venue, etc.), and only retain the effective area containing the ball and the athlete, reducing the amount of data while improving the efficiency of model training. Then, image normalization is performed to normalize the image pixel values ​​to a specific range (such as between 0 and 1) to enhance the stability and consistency of the data and facilitate subsequent model training.

[0063] The model construction and parameter optimization are to use the YOLOv5 algorithm to build a convolutional neural network (CNN) model, and combine it with the SE-Net attention mechanism. When building the model, the network structure is reasonably designed according to the requirements of the ball target detection task. For example, the convolution kernel size of the convolution layer is set to 3×3 and 5×5, the number is 32, 64, etc., and the step size is 1 or 2. Through the combination of convolution kernels of different sizes, the local features and overall features of the ball can be effectively extracted. The pooling layer uses the maximum pooling operation, the pooling window size is 2×2, and the step size is 2, which reduces the data dimension and the amount of calculation. The number of neurons in the fully connected layer is designed according to the classification task and feature dimension, such as 128 or 256, etc., to integrate the feature information extracted by the previous convolution layer and pooling layer. During the training process, the stochastic gradient descent (SGD) optimization algorithm is used, combined with the momentum method (Momentum) and the learning rate decay strategy (such as cosine annealing learning rate decay). The momentum coefficient of the momentum method is set to 0.9 to help the model converge faster and avoid falling into the local minimum. The cosine annealing learning rate decay strategy dynamically adjusts the learning rate according to the number of training rounds. The initial learning rate is set to 0.001, and the learning rate is gradually reduced as the number of training rounds increases, so that the model can adjust parameters more finely in the later stage of training, thereby improving the accuracy and generalization ability of the model.

[0064] The application of data enhancement and transfer learning is to use a variety of data enhancement techniques to expand the training data set. The random cropping technology randomly crops parts of the image while keeping the ball intact, increasing the model's ability to recognize the ball at different positions and scales; the rotation operation randomly rotates the image within a certain angle range (such as ±30°) to simulate the situation of the ball flying at different angles; the flip operation includes horizontal and vertical flipping of the image, so that the model can adapt to the movement of the ball in different directions; the brightness adjustment randomly changes the brightness of the image within a certain range (such as ±0.2) to enhance the model's ability to recognize the ball under different lighting conditions. At the same time, the transfer learning technology is used to migrate the model parameters pre-trained in other similar ball sports (such as tennis, badminton, etc.) or target detection tasks to this model. For example, by loading the YOLOv5 model parameters pre-trained on a large-scale tennis image dataset and then fine-tuning it on the pickleball dataset, only the parameters of some network layers (such as the last few fully connected layers) need to be retrained, which greatly speeds up the convergence of the model and improves the training effect. At the same time, the prior knowledge of the pre-trained model is used to enhance the model’s ability to extract ball features and classification accuracy.

[0065] Preferably, in the adaptive adjustment of Kalman filter parameters, the initialization parameter determination is to set the initial state vector and the initial covariance matrix according to the initial position and velocity estimation value of the ball when the system is initialized. The initial position can be determined by the sensor measurement value or video image analysis of the ball at the moment of serving, and the velocity estimation value can be estimated according to the player's hitting power and the initial movement direction of the ball. For example, for the case of a faster serving speed (such as a serving speed of about 30m / s), the initial velocity estimation value can be set to a corresponding range, and the initial covariance matrix is ​​reasonably set according to the uncertainty of the sensor measurement error and the position and velocity estimation to reflect the uncertainty degree of the initial state, providing a basis for the accurate operation of the subsequent Kalman filter algorithm.

[0066] The real-time parameter adjustment mechanism is to monitor the changes in the motion state of the ball in real time as the ball moves. The acceleration data of the ball is obtained in real time through the micro sensor on the ball (such as IMU), and the velocity and position changes of the ball are calculated using the acceleration data. The velocity and position changes of the ball are compared with the position and velocity predicted by the Kalman filter algorithm to calculate the state estimation error. At the same time, the estimated value of the measurement error is determined according to the accuracy and stability of the sensor measurement. According to the state estimation error and the measurement error, the process noise covariance matrix and the measurement noise covariance matrix are automatically adjusted using the update formula of the Kalman filter algorithm. For example, when the ball is disturbed by a large external force during the movement (such as encountering strong winds or colliding with the boundary of the field), resulting in a sudden change in the motion state, the state estimation error will increase. At this time, the Kalman filter algorithm will increase the process noise covariance matrix accordingly to improve the adaptability to the new state; when the sensor measurement data fluctuates, the measurement error estimate will change accordingly, and the algorithm will adjust the measurement noise covariance matrix to ensure a reasonable degree of trust in the measurement data, thereby ensuring the accuracy of trajectory tracking, so that the algorithm can accurately track the motion trajectory of the ball in real time.

[0067] The data integration module is connected to the data acquisition module and the trajectory tracking module respectively, and is used to integrate the ball motion trajectory data, the venue information in the video image data, and the athlete action data to obtain comprehensive data. Among them, the venue information in the data integration module includes the net height information and the non-volley zone rule information, and the athlete action data includes the athlete's hitting action data, moving step data, and body posture data.

[0068] Preferably, in the precise measurement and storage of venue information, the measurement tools and methods are selected in that when using a high-precision laser rangefinder and an electronic level to measure the venue information, the measurement accuracy of the laser rangefinder must be within ±1mm to ensure the accuracy of the measurement of dimensions such as the height of the net. For the measurement of the height of the net, multiple measurements are performed at both ends and the middle position of the net of the venue (such as 5 measurements at each position), and the average value is taken as the final measurement result, and it is accurate to the millimeter level. During the measurement process, ensure that the measuring light of the laser rangefinder is perpendicular to the net to avoid measurement errors caused by measurement angle deviations. The electronic level is used to measure the flatness and slope of the venue. By placing a level at different positions on the venue, measuring the height difference of each point on the venue, calculating the slope value of the venue, and ensuring that the slope of the venue meets the competition standards (such as the slope does not exceed 0.5%). For the measurement of the boundary of the non-intercepting area, a high-precision GPS positioning device is used, and its positioning accuracy should be within 0.1 meters. Continuous positioning and drawing are performed along the boundary to obtain the precise coordinate points of the boundary to ensure the accuracy of the non-intercepting area range.

[0069] The data storage structure design is to store the measured venue information in the local SQLite database and design a reasonable data table structure. Create a table named "venue_info" containing fields "id" (venue information unique identifier), "net_height" (net height), "court_slope" (venue slope), "non_volley_zone_boundary" (non-volley zone boundary coordinate point set, stored in a specific format, such as JSON format), etc. When storing the net height data, the height value accurate to millimeters is directly stored in the corresponding field; the venue slope value is stored in percentage form; the non-volley zone boundary coordinate point set stores the longitude and latitude information of each point in sequence. Through this data storage structure design, it is convenient to quickly read and call data. For example, when the tactical decision module needs to obtain venue information, it can quickly and accurately obtain the required venue information through a simple SQL query statement (such as "SELECT*FROM venue_info WHERE id=[specific venue ID]"), thereby improving the system operation efficiency.

[0070] Preferably, in the synchronization of athlete motion data, the video motion capture system setting and synchronization technology is to combine the Vicon video motion capture system based on multiple cameras (such as 4-6 cameras), and reasonably arrange the cameras to ensure that the entire competition venue can be covered without blind spots. The resolution of the camera is set to 1920×1080 or higher, and the frame rate is not less than 120Hz to clearly capture the three-dimensional motion trajectory and action details of the athlete. During the system setting process, each camera is calibrated, including internal parameter calibration (such as focal length, principal point position, etc.) and external parameter calibration (such as the position and posture of the camera) to ensure that the coordinate system between multiple cameras is unified, which is convenient for subsequent data fusion. High-precision time synchronization technology (such as network time protocol (NTP) or precision time protocol (PTP)) is used to accurately align the data collected by the motion sensor with the data of the video motion capture system on the timeline.

[0071] The tactical decision module is connected to the data integration module and is used to classify tactics according to the comprehensive data. At the same time, a prediction model is used to predict the tactical effect with tactical features as input and game results as output to obtain tactical decision suggestions. Among them, the tactical decision module uses the decision tree algorithm to build a tactical classification model to classify tactics according to the comprehensive data. The decision tree algorithm selects the optimal features for node splitting by calculating information gain or Gini index to build a tactical classification model; the tactical decision module uses a neural network algorithm to build a tactical effect prediction model to predict the tactical effect. The neural network algorithm uses a multi-layer perceptron and optimizes the model weights and biases through the back propagation algorithm.

[0072] Preferably, in the construction and improvement of the decision tree model, data collection and preprocessing are to collect a large amount of pickleball game data, which should cover the performance of athletes of different levels (amateurs, professionals, etc.) in various game scenarios (championships, friendly matches, training matches, etc.). The data content includes the ball landing area (accurate to the division of the field, such as the front court, midfield, back court, and left and right half courts, etc.), the distance between the athletes (calculated by sensors on the athletes or video analysis, accurate to the centimeter level), the hitting angle (based on the horizontal and vertical directions, accurate to degrees), the hitting force (estimated based on the acceleration of the ball measured by the sensor, accurate to a certain force value range), the rotation speed of the ball (obtained by the sensor on the ball or video analysis, accurate to revolutions per second) and other tactical features, as well as the corresponding game results (such as scoring, losing points, successful offense, successful defense, etc.). In the data preprocessing stage, data cleaning is first performed to remove data records with obvious errors or abnormalities, such as data with a hitting angle that exceeds a reasonable range (such as greater than 180 degrees or less than -180 degrees). For missing values, appropriate methods are used to process them, such as filling the mean or median according to the distribution characteristics of the data. Then, feature normalization is performed to map feature values ​​of different ranges to a specific interval (such as between 0 and 1). For example, the batting force value is converted to the interval of 0 to 1 through the normalization formula, so that different features have equal importance in model training and improve the model training effect.

[0073] Model training and optimization is to build a tactical classification model based on the improved C4.5 decision tree algorithm. When calculating the information gain ratio, a penalty factor is introduced, and the calculation of the penalty factor is determined according to the importance of the feature and the data distribution. For example, for features that appear too frequently or too infrequently in the data, the penalty factor is appropriately increased to reduce its impact on the model and avoid overfitting the model to the training data. In the process of building a decision tree, the optimal decision tree depth is selected by training and evaluating decision trees of different depths (such as using cross-validation technology to divide the data set into training set, validation set and test set, training decision trees of different depths on the training set, evaluating the model performance on the validation set, and selecting the depth with the best performance). At the same time, the minimum number of samples for leaf nodes is determined (according to the amount of data and data distribution, select the appropriate number of samples, such as 10 to 20 samples, etc.), optimize the decision tree model structure, and improve the accuracy of tactical classification. After the model training is completed, the model is evaluated on the test set. The evaluation indicators include accuracy, recall rate, F1 value, etc., to ensure that the model has good generalization ability on new data.

[0074] Preferably, in the neural network model design and training optimization, the model structure design is to use a deep neural network (DNN) to build a tactical effect prediction model, using a multi-layer convolutional layer and a fully connected layer structure. The convolution layer is used to extract the complex relationship between tactical features, for example, by setting convolution kernels of different sizes (such as 3×3, 5×5) to capture tactical feature combinations of different scales. The number of convolution kernels in the first convolution layer can be set to 32. As the network depth increases, the number of convolution kernels gradually increases (such as 64 in the second layer, 128 in the third layer, etc.) to extract more advanced features. Pooling layers (such as maximum pooling, with a pooling window of 2×2) perform downsampling between convolution layers to reduce data dimensions and prevent overfitting. The fully connected layer is designed according to the prediction task. The number of neurons in the input layer is determined according to the dimension of the tactical features (such as the length of the feature vector after feature extraction). The hidden layer is set to 2-3 layers, and the number of neurons gradually decreases (such as 256 in the first layer, 128 in the second layer, 64 in the third layer, etc.). The output layer is determined according to the prediction results (such as 1 neuron for predicting scores and 2 neurons for predicting wins and losses). It is used to predict tactical effects and outputs the predicted values ​​of the game results (such as scoring probability, win and loss prediction, etc.) based on the tactical feature input.

[0075] The optimization of the training process lies in the use of stochastic gradient descent (SGD) combined with momentum and adaptive learning rate adjustment strategies (such as Adagrad) during model training. The momentum coefficient of the momentum method is set to 0.9 to help the model converge faster and avoid falling into local minima. Adagrad automatically adjusts the learning rate based on the historical gradient information of the parameters. The initial learning rate is set to 0.001. During the training process, the learning rate will automatically adjust according to the change of the gradient, so that the learning rate can adjust the parameters more finely in the later stage of training, improving the stability and accuracy of model training. At the same time, set reasonable training parameters, such as the number of training rounds (epochs) is set to 100-200 rounds according to the size of the data set and the complexity of the model, and the batch size (batch size) is selected according to the hardware resources and the amount of data. Choose a suitable value (such as 32, 64 or 128, etc.). The Early Stopping method is used to monitor the loss function value on the validation set during training. When the validation set loss no longer decreases after several consecutive rounds (such as 5-10 rounds), training is stopped to prevent the model from being over-trained and ensure that the model can achieve good performance on both the training set and the test set.

[0076] The output module is connected to the tactical decision module and is used to output the tactical decision suggestions in a visual way. The tactical decision suggestions include the recommended tactical type, the timing of tactical implementation, and the expected tactical effect; the visualization method includes superimposing tactical prompt information on the video screen and generating tactical analysis charts or reports.

[0077] Preferably, in AR technology-assisted real-time decision-making, AR application development is to use augmented reality (AR) technology to provide real-time tactical decision assistance to athletes through AR glasses. Develop a special AR application, which first needs to establish a stable communication connection with the tactical decision module, and receive the tactical prompt information (such as tactical type, implementation timing, attack direction, defense area, etc.) generated by the tactical decision module in real time. In the program, the AR development framework (such as Unity combined with Vuforia, etc.) is used to convert the tactical prompt information into intuitive graphics and text forms. For example, when the system recommends offensive tactics, an arrow model pointing to the direction of attack is created in the AR scene, and the text prompt "At this time, it is recommended to attack, and the direction of attack is the bottom line on the left side of the opponent's field" is marked at a suitable position in the athlete's field of view. In order to ensure the stability and accuracy of AR display, the performance of AR applications is optimized, including image rendering algorithms, tracking and positioning algorithms, etc. In terms of image rendering, an efficient rendering pipeline is used to reduce image rendering delays and ensure that graphics and text can be displayed in the athlete's field of view in a timely and clear manner. The tracking and positioning algorithm uses vision-based simultaneous localization and mapping (SLAM) technology, combined with the cameras and sensors on the AR glasses, to track the athlete's head position and movement direction in real time, ensuring that tactical prompt information can be accurately superimposed on the corresponding position in the athlete's field of view without offset or shaking.

[0078] Interface design and user experience optimization will reasonably design the layout and style of the AR display interface according to the athletes' visual habits and sports needs. For example, tactical prompt information is displayed in the edge area of ​​the athlete's field of vision to avoid blocking the athlete's normal line of sight, while at the same time being able to quickly obtain information within the athlete's peripheral vision. Text prompts use concise and clear fonts and colors to ensure clear readability under different lighting conditions. Graphic prompts such as arrows use bright colors and appropriate sizes for easy identification. In addition, in order to facilitate athlete operation, AR applications can support voice interaction functions. Athletes can use voice commands to query more tactical information or confirm operations, etc., to improve the convenience and efficiency of athletes using the system during the game, so that they can quickly obtain tactical guidance and adjust game strategies in a timely manner without distracting their attention.

[0079] Preferably, in the generation of a detailed tactical analysis report, data visualization and report generation are to use data visualization libraries (such as Matplotlib, Seaborn in Python, etc.) and text processing tools to generate a detailed tactical analysis report based on the analysis results of the tactical decision module. The report content includes the basis for tactical decision-making (detailed display of data support based on model analysis), which is presented by drawing various charts, such as drawing a histogram of the ball's landing point distribution to intuitively display the frequency of the ball's landing points in different areas of the field; drawing a relationship diagram between the athlete's position and the ball's trajectory, using a line graph or a scatter plot to show the relationship between the athlete's position at different times and the ball's trajectory, etc. The implementation steps section specifically describes the athlete's action requirements when executing tactics (such as the specification of the batting action, the rhythm and direction of the moving steps, etc.), the coordination method (such as the position coordination between doubles players, the batting order, the filling strategy, etc.) and the timing (such as executing tactical actions when the ball reaches a specific position or the opponent is in a specific state, accurate to the distance or time node of the ball's flight, etc.). The expected effect is presented in the form of quantitative data (such as the probability of scoring given in percentage form, the probability of losing points is also quantified, and the quantitative evaluation of the impact of different tactics on the game situation after implementation, such as changes in ball possession, etc.) and text descriptions to present the possible results after the implementation of tactics, as well as the possible impact on the opponent (such as the opponent's possible return direction, tactical response strategy, etc.). In addition, the report also includes targeted operational suggestions for athletes. According to the athlete's technical characteristics (such as hitting power, hitting accuracy, moving speed, endurance, etc.) and current status (such as physical condition, technical performance level, psychological state, etc.), it is recommended to adjust the range of hitting power (given in the form of specific force value range or percentage increase or decrease), optimize the position (such as the adjustment of the front and back court positions, and the movement suggestions for the left and right half courts, etc.).

[0080] The report format and interactive design will generate a tactical analysis report in HTML5 format, making it cross-platform and interactive, so that athletes and coaches can view and share it on various devices (such as computers, tablets, mobile phones, etc.). Add interactive elements to the report, such as clicking on the chart to view detailed data. When the athlete clicks on the ball's landing point distribution histogram, a pop-up window containing specific data for each landing area (number of landing points, proportion, etc.) will pop up; the folding and expanding function is convenient for viewing different parts of the content. For example, the tactical implementation steps section is set to be foldable, and athletes can expand and view detailed content as needed, improving the readability and practicality of the report. At the same time, the report can also support export functions, such as exporting to PDF format, so that athletes and coaches can view and archive it offline. The tactical analysis report generated in the above way can provide athletes and coaches with comprehensive, in-depth and intuitive tactical analysis information, helping them to better summarize game experience and formulate training plans and game strategies.

[0081] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. All equivalent structural changes made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or directly or indirectly applied in other related technical fields, are included in the patent protection scope of the present invention.

Claims

1. A pickleball motion trajectory tracking and tactical decision-making support system, characterized by: Included A data acquisition module is used to collect video image data, ball movement data and athlete action data during the ball movement process; The trajectory tracking module is connected to the data acquisition module. Based on the deep learning target detection and tracking algorithm, it detects and identifies the ball in the video image, predicts and corrects the position of the ball, and obtains the ball motion trajectory data; The data integration module is connected to the data acquisition module and the trajectory tracking module respectively, and is used to integrate the ball motion trajectory data, the field information in the video image data and the player action data to obtain comprehensive data; The tactical decision module is connected to the data integration module and is used to classify tactics according to the comprehensive data. At the same time, a prediction model is used to predict the tactical effect with tactical features as input and game results as output to obtain tactical decision suggestions; The output module is connected to the tactical decision module and is used to output the tactical decision suggestions in a visual manner.

2. The pickle ball motion trajectory tracking and tactical decision-making support system according to claim 1, characterized in that: The data acquisition module includes a camera for collecting video image data, a pickle ball motion sensor installed on the pickle ball to collect the ball's motion data, and an athlete motion sensor worn on the athlete to collect the athlete's motion data.

3. The pickle ball motion trajectory tracking and tactical decision-making support system according to claim 1, characterized in that: In the trajectory tracking module, a convolutional neural network is used to detect and identify the ball in the video image; in the trajectory tracking module, a Kalman filter algorithm is used to predict and correct the position of the ball.

4. The pickle ball motion trajectory tracking and tactical decision-making support system according to claim 3, characterized in that: Convolutional neural networks use Faster R-CNN or YOLO algorithms for target detection.

5. The pickle ball motion trajectory tracking and tactical decision-making support system according to claim 3, characterized in that: The Kalman filter algorithm predicts the position of the ball at the next moment based on its position and speed at the previous moment, and makes corrections based on the current detection results.

6. The pickle ball motion trajectory tracking and tactical decision-making support system according to claim 1, characterized in that: The venue information in the data integration module includes the net height information and the non-volley zone rule information, and the athlete action data includes the athlete's hitting action data, moving pace data and body posture data.

7. The pickle ball motion trajectory tracking and tactical decision-making support system according to claim 1, characterized in that: In the tactical decision module, the decision tree algorithm is used to build a tactical classification model to classify tactics according to comprehensive data; in the tactical decision module, the neural network algorithm is used to build a tactical effect prediction model to predict the tactical effect.

8. The pickle ball motion trajectory tracking and tactical decision-making support system according to claim 7, characterized in that: The decision tree algorithm selects the optimal features for node splitting by calculating information gain or Gini index to build a tactical classification model.

9. The pickle ball motion trajectory tracking and tactical decision-making support system according to claim 7, characterized in that: The neural network algorithm uses a multi-layer perceptron and optimizes model weights and biases through the back-propagation algorithm.

10. The pickle ball motion trajectory tracking and tactical decision-making support system according to claim 1, characterized in that: Tactical decision recommendations include recommended tactical types, tactical implementation timing, and expected tactical effects; visualization methods include superimposing tactical prompt information on the video screen and generating tactical analysis charts or reports.

Images