A Dynamic Sports Data Visualization System and Method

Abstract

The present application provides a dynamic sports data visualization system and method, which relates to the technical field of data visualization analysis, and captures the competition behavior data of athletes in sports events in real time; determines multiple motion interaction nodes of an athlete in the current sports event, and determines the motion evolution trend of each motion interaction node according to the motion tendency trajectory of the athlete; conducts feedback evaluation on the real-time state data to obtain the motion feedback strategy of the athlete, and determines the motion linkage attribute of the athlete from the motion feedback strategy and the motion posture information of the athlete in the current sports event; inputs the motion evolution trend and the motion linkage attribute into a preset motion prediction model for real-time training, and visually displays the output motion prediction data. The present application can perform linkage analysis on the dynamic sports data of athletes in a data environment with real-time capture ability and multi-modal perception basis, so as to improve the effectiveness of visual processing of collaborative interactive sports dynamic data.

Description

Technical Field

[0001] This application relates to the technical field of data visualization analysis. More specifically, this application relates to a dynamic sports data visualization system and method. Background Art

[0002] Data visualization analysis refers to the conversion, presentation, and interactive operation of structured or unstructured raw data in a graphical way, so as to achieve an intuitive perception and cognitive understanding of data features, laws, trends, and anomalies. In the field of sports, especially in the analysis of dynamic sports data, various behavioral data of athletes during training or competitions often have the characteristics of high dimensionality, strong time series, and multi-variable coupling. Traditional numerical tables or static charts are difficult to comprehensively reflect their dynamic evolution process and internal structure.

[0003] However, existing dynamic sports data visualization methods generally have problems such as single data source, lack of spatio-temporal information, lack of real-time performance, and lack of intelligent correlation analysis ability, making it difficult for the system to comprehensively reflect the action details and collaborative characteristics of athletes in high-speed dynamics, resulting in difficult identification of key action nodes, lagging prediction of movement trends, and lack of pertinence in training feedback, thus severely restricting their application effects in professional competitive analysis and intelligent training guidance. Therefore, how to perform linkage analysis on the dynamic sports data of athletes in a data environment with real-time capture ability and multi-modal perception basis to improve the effectiveness of collaborative interactive sports dynamic data visualization processing is a problem faced by the industry. Summary of the Invention

[0004] This application provides a dynamic sports data visualization system and method, which can perform linkage analysis on the dynamic sports data of athletes in a data environment with real-time capture ability and multi-modal perception basis to improve the effectiveness of collaborative interactive sports dynamic data visualization processing.

[0005] In a first aspect, this application provides a dynamic sports data visualization method, and the data visualization method includes the following steps:

[0006] Based on computer vision technology, the action postures of athletes in sports events are captured in real time to obtain the competition behavior data of athletes in sports events;

[0007] Extract the key action information of athletes in sports events from the competition behavior data, perform posture recognition on the key action information to obtain multiple motion interaction nodes of athletes in the current sports event, and determine the action evolution trend of each motion interaction node according to the action tendency trajectory of the athletes;

[0008] Obtain the real-time status data of the athlete in the current sports event, conduct feedback evaluation on the real-time status data to obtain the athlete's motion feedback strategy, and then determine the action linkage attribute of the athlete based on the motion feedback strategy and the motion posture information of the athlete in the current sports event;

[0009] Input the action evolution trend and the action linkage attribute into a preset action prediction model for real-time training, and visually display the output action prediction data.

[0010] In this embodiment, the competition behavior data is a structured data set describing the action postures and spatio-temporal trajectories of athletes during the competition.

[0011] In this embodiment, performing pose recognition on the key action information to obtain multiple motion interaction nodes of the athlete in the current sports event specifically includes:

[0012] Obtain the motion posture features in the key action information;

[0013] Based on a pre-trained three-dimensional convolutional pose recognition model, perform motion trajectory clustering analysis on the motion posture features to detect candidate interaction nodes that meet the kinematic constraint conditions of the athlete in the current sports event;

[0014] Combine the rule library of the current sports event to resolve conflicts for the candidate interaction nodes, and generate multiple motion interaction nodes of the athlete in the current sports event.

[0015] In this embodiment, determining the action evolution trend of each motion interaction node according to the action tendency trajectory of the athlete specifically includes:

[0016] Set the action tendency trajectory of the athlete based on the rule library of the current sports event;

[0017] Determine the dynamic displacement range corresponding to the action of the athlete according to the action tendency trajectory;

[0018] Determine the trajectory evolution characteristics when the athlete's action changes through the dynamic displacement range;

[0019] Determine the action evolution trend of each motion interaction node from the trajectory evolution characteristics.

[0020] In this embodiment, the action evolution trend represents the process of the athlete smoothly transitioning from one action state to another during the movement.

[0021] In this embodiment, the real-time status data refers to the data on the physiological state and motion behavior of the athlete obtained in real time through sensors and vision systems during the competition.

[0022] In this embodiment, determining the action linkage attribute of the athlete based on the motion feedback strategy and the motion posture information of the athlete in the current sports event specifically includes:

[0023] Obtain the motion posture information of the athlete in the current sports event;

[0024] Extract the motion trajectory features of the athlete from the motion posture information;

[0025] Determine the action delay deviation of the athlete according to the motion trajectory features;

[0026] Determine the action constraint conditions of the athlete in the current sports event according to the motion feedback strategy;

[0027] Determine the action linkage attribute of the athlete through the action delay deviation and the action constraint conditions.

[0028] In this embodiment, inputting the action evolution trend and the action linkage attribute into a preset action prediction model for real-time training specifically includes:

[0029] Generate a dynamic evolution input sequence according to the probability distribution and trend verification index in the action evolution trend;

[0030] Through the spatio-temporal convolution-attention hybrid network layer of the preset action prediction model, perform multi-scale feature fusion and dynamic weight allocation on the dynamic evolution input sequence, and calculate the causal association strength between each linkage node;

[0031] Perform incremental adversarial training on the preset action prediction model according to the causal association strength, and output action prediction data.

[0032] In this embodiment, the action prediction data refers to the prediction result of the action prediction model on the change of the athlete's action state within a future time window.

[0033] In a second aspect, the present application provides a dynamic sports data visualization system for executing a dynamic sports data visualization method. The data visualization system includes:

[0034] A data acquisition module, configured to perform real-time capture of the action postures of athletes in a sports event based on computer vision technology, and obtain the competition behavior data of the athletes in the sports event;

[0035] A posture recognition module, configured to extract the key action information of the athlete in the sports event from the competition behavior data, perform posture recognition on the key action information, obtain multiple motion interaction nodes of the athlete in the current sports event, and determine the action evolution trend of each motion interaction node according to the action tendency trajectory of the athlete;

[0036] A feedback evaluation module, configured to obtain real-time status data of an athlete in a current sports event, perform feedback evaluation on the real-time status data to obtain a sports feedback strategy of the athlete, and further determine an action linkage attribute of the athlete based on the sports feedback strategy and the motion posture information of the athlete in the current sports event;

[0037] A visualization display module, configured to input the action evolution trend and the action linkage attribute into a preset action prediction model for real-time training, and perform visualization display on the output action prediction data.

[0038] The technical solution provided by the embodiments disclosed in this application has the following beneficial effects:

[0039] Based on computer vision technology, the motion postures of athletes in a sports event are captured in real time to obtain the competition behavior data of athletes in the sports event; key action information of the athletes in the sports event is extracted from the competition behavior data, and posture recognition is performed on the key action information to obtain multiple motion interaction nodes of the athletes in the current sports event. The action evolution trend of each motion interaction node is determined according to the action tendency trajectory of the athletes; real-time status data of the athletes in the current sports event is obtained, feedback evaluation is performed on the real-time status data to obtain a sports feedback strategy of the athletes, and further the action linkage attribute of the athletes is determined based on the sports feedback strategy and the motion posture information of the athletes in the current sports event; the action evolution trend and the action linkage attribute are input into a preset action prediction model for real-time training, and the output action prediction data is visually displayed.

[0040] It can be seen from this application that the accuracy of dynamic sports data analysis and the decision-making assistance ability are improved; among them, by capturing competition behavior data, non-contact high-precision real-time acquisition of the complex posture actions of athletes is realized, ensuring the comprehensiveness, dynamics and continuity of data sources; by performing posture recognition on key action information, the key posture change nodes in the motion process can be accurately located, the structured evolution logic behind the motion behavior can be identified, and the accuracy and timeliness of action recognition and behavior understanding are improved; by analyzing real-time status data and integrating real-time physiological status and motion posture information, the constraint relationship between motion strategies and action responses is dynamically calculated, the intelligent recognition of action linkage attributes is realized, and the understanding ability of the visualization system for the cooperation mechanism is improved; by visually displaying action prediction data, real-time prediction and visual presentation of action trends can be realized, the forward-looking and interactivity of sports data analysis are improved, and at the same time, multi-dimensional prediction results are used to drive the three-dimensional visualization process, and the action coordination change process and the causal relationship between nodes are presented in real time, enhancing the interactive display efficiency and linkage analysis effectiveness of sports dynamic data.

[0041] In summary, the technical solution adopted in this application can perform a linkage analysis on the dynamic sports data of athletes in a data environment with real-time capture capabilities and multi-modal perception as the basis, so as to improve the effectiveness of visual processing of sports dynamic data for collaborative interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0043] Figure 1 is a flowchart of a method for visualizing dynamic sports data provided by the present application;

[0044] Figure 2 is a schematic flowchart for determining key action information provided by the present application;

[0045] Figure 3 is a schematic flowchart for determining a motion feedback strategy provided by the present application;

[0046] Figure 4 is a module structure diagram of a system for visualizing dynamic sports data provided by the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0048] The embodiment of the present application provides a dynamic sports data visualization system and method. The core is to capture the action postures of athletes in sports events in real time based on computer vision technology to obtain the competition behavior data of athletes in sports events; extract the key action information of athletes in sports events from the competition behavior data, perform posture recognition on the key action information to obtain multiple motion interaction nodes of athletes in the current sports event, and determine the action evolution trend of each motion interaction node according to the action tendency trajectory of the athletes; obtain the real-time state data of athletes in the current sports event, perform feedback evaluation on the real-time state data to obtain the motion feedback strategy of the athletes, and then determine the action linkage attribute of the athletes from the motion feedback strategy and the motion posture information of the athletes in the current sports event; input the action evolution trend and the action linkage attribute into a preset action prediction model for real-time training, and visually display the output action prediction data.

[0049] Embodiment 1. To better understand the above technical solution, the above technical solution will be described in detail below in combination with the accompanying drawings of the specification and specific implementation manners. Refer to Figure 1 As shown, this figure is an exemplary flowchart of a dynamic sports data visualization method according to this embodiment of the present application. The data visualization method includes the following steps:

[0050] In step S1, capture the action postures of athletes in sports events in real time based on computer vision technology to obtain the competition behavior data of athletes in sports events.

[0051] Specifically, deploy multiple high-frame-rate industrial cameras inside the stadium to perform multi-angle coverage shooting on the competition area, collect the continuous image sequence of athletes during the competition, input the image data in the continuous image sequence into the convolutional neural network model, perform key point recognition on the human body in the continuous image sequence, and extract the two-dimensional coordinate information of joints such as the head, torso, and limbs; under the condition of multi-camera synchronization, combined with the multi-view geometry principle and time synchronization algorithm, the three-dimensional space reduction of the two-dimensional key points can be performed to obtain the three-dimensional action postures of athletes in each time frame; then through time series integration, the posture information of continuous frames is constructed into an action trajectory sequence to form a structured behavior data including elements such as spatial position, time index, and action amplitude, and this structured behavior data is used as the competition behavior data of athletes in sports events.

[0052] It should be noted that in the present application, the competition behavior data is a structured data set describing the action postures and spatio-temporal trajectories of athletes during the competition.

[0053] In step S2, key action information of athletes in a sports event is extracted from the competition behavior data, pose recognition is performed on the key action information to obtain multiple motion interaction nodes of the athletes in the current sports event, and the action evolution trend of each motion interaction node is determined according to the action tendency trajectory of the athletes.

[0054] Preferably, in this embodiment, key action information of athletes in a sports event is extracted from the competition behavior data, with reference to Figure 2 As shown, this figure is a schematic flowchart for determining key action information in some embodiments of the present application. The determination of key action information in this embodiment can be implemented by the following steps:

[0055] In step S21, during the parsing of the athlete's action characteristics, action sequence basic data including spatio-temporal coordinates and motion parameters is extracted from the competition behavior data;

[0056] In step S22, based on a preset motion mechanics model, pose matching is performed on the action sequence basic data, and an action deviation index between each action node and a standard action template is calculated;

[0057] In step S23, a three-dimensional motion trajectory anomaly map is constructed according to the action deviation index to identify the action anomaly area during the athlete's action execution process;

[0058] In step S24, through the association mapping between the action anomaly area and the event rule library, key action information of athletes in a sports event is generated.

[0059] In specific implementation, first, in the competition behavior data, the continuous action sequences of athletes are extracted in chronological order; the content included in each frame includes: 3D spatial coordinates, where the 3D spatial coordinates include: the X, Y, and Z coordinates of body joint points, action speed, acceleration, attitude angle, action frequency, etc. Through the fusion of the pose estimation module and sensors using IMU or GPS, the pose data of each frame is calibrated and complemented to form a basic data set of action sequences containing spatio-temporal coordinates and motion parameters. Then, a preset motion mechanics model is introduced. This preset dynamics model sets a group of standard action templates based on the standard technical actions of specific projects, which can be: the four-stage action trajectories of approach run, takeoff, flight, and landing in long jump; the basic data of the athlete's action sequence is input into this motion mechanics model, and the spatial trajectories, joint angles, duration, action rhythm, and other parameters of each action node are compared with those of the corresponding template action. The Euclidean distance, cosine similarity, or dynamic time warping algorithm is used to calculate the deviation between each action node and the standard action template, and this deviation is used as the action deviation index. Then, all the action nodes in the action deviation index are mapped into 3D space, and a 3D motion trajectory anomaly map is generated with the error value as the color or height variable; in the 3D motion trajectory anomaly map, the areas where the error significantly deviates from the standard interval are marked as "action anomaly areas", where the standard interval can be determined according to the average value of all action deviation indexes, which will not be elaborated here. The 3D trajectory map is visualized through color coding (such as red representing large deviation) or voxel stacking to clearly display the distribution of abnormal points in the technical execution of the motion trajectory. Finally, the action anomaly areas are mapped into the event rule library and matched with the key action nodes marked in the event rule library, where the key action nodes marked in the event rule library can be: shooting release point, hurdle takeoff point, etc., to identify whether the anomalies in the action anomaly areas occur at positions with tactical or rule significance; if the match is successful, this key action node is used as key action information, and its type, position, time, deviation degree, and other indexes are recorded; it is also possible to implement automatic matching in combination with the rule label system, such as judging whether key action information is formed by the overlap of key point labels and time windows.

[0060] It should be noted that in this application, the spatio-temporal coordinates represent a set of information describing the specific positions of each key part of the athlete's body in a three-dimensional space and the time points at which they occur during the competition; the motion parameters represent a set of information quantitatively describing the changes in the motion state; the basic action sequence data refers to a set of time series data of the continuous action states of the athlete; the preset motion mechanics model represents a model describing the characteristics of each key stage of the standard technical action in dimensions such as spatio-temporal, mechanics, and angle; the action node represents a key position with structural changes during the action process; the standard action template represents a standardized technical action data model for reference and comparison; the action deviation index refers to a numerical value measuring the degree of difference between the athlete's actual action and the standard action; the three-dimensional motion trajectory anomaly atlas represents a graphical result generated by fusing the spatial trajectory of the athlete's executed action and the deviation index; the action anomaly area refers to the spatio-temporal distribution section in the athlete's action trajectory that has a significant deviation from the standard action; the event rule library represents a knowledge system defining the key action nodes, tactical requirements, and scoring criteria in sports; the key action information refers to the action nodes that have an impact on the competition process and are technically significant for analysis.

[0061] In this embodiment, the posture recognition of the key action information to obtain multiple motion interaction nodes of the athlete in the current sports event can be implemented by the following steps:

[0062] Obtain the motion posture characteristics in the key action information;

[0063] Based on a pre-trained three-dimensional convolutional posture recognition model, perform motion trajectory clustering analysis on the motion posture characteristics to detect candidate interaction nodes that meet the kinematic constraint conditions of the athlete in the current sports event;

[0064] Combine the rule library of the current sports event to resolve conflicts of the candidate interaction nodes and generate multiple motion interaction nodes of the athlete in the current sports event.

[0065] In specific implementation, first, extract the posture features related to action execution, namely motion posture features, from the recognized key action information. The motion posture features include: Spatial posture features: the skeleton structure composed of the three-dimensional coordinates of key points; Dynamic features: the speed, acceleration, joint angle changes, etc. of the key point trajectories; Posture stability indicators: such as the body center of gravity offset rate, action symmetry, etc. In the process of extracting motion posture features, a time series window can be used to perform sliding sampling on several frames of data before and after the action nodes and unify them into a standardized tensor format. Then, input the extracted motion posture features into a trained three-dimensional convolutional neural network model. Among them, the three-dimensional convolutional neural network model can be I3D, ST-GCN, which is not limited here; this three-dimensional convolutional neural network model can process the information in both the time and space dimensions and identify the patterns in the posture changes; at the output layer, combine the K-Means clustering algorithm to cluster the action segments with similar patterns in the continuous trajectory and identify the candidate nodes with interactive significance, that is, candidate interaction nodes. Among them, the candidate interaction nodes need to meet the requirements that the trajectory structure has stable continuity; the action state conforms to specific kinematic constraints. Among them, the specific kinematic constraints can be the joint range of motion, body balance posture. Finally, match and compare the candidate interaction nodes with the rule library of the current sports event, including checking whether the candidate interaction nodes appear in the specified area of the competition and whether they conform to the definition of tactical actions in the sport. Among them, the definition of tactical actions can be passing the ball, receiving the ball, taking off, changing direction; when multiple candidate interaction nodes overlap, conflict or have repeated meanings in time or space, conflict resolution is carried out according to the priorities, tolerance ranges or action labels set in the rule library, and only the valid and non-redundant candidate interaction nodes are retained. The valid and non-redundant candidate interaction nodes are used as motion interaction nodes; among them, conflict resolution can adopt: the time priority rule (retain the one that occurs first), the space separation principle (only retain the action points with independent functions) and the action label fusion strategy (merge similar actions into composite interaction points).

[0066] It should be noted that in this application, the motion posture features refer to the skeleton structure and dynamic parameters reflecting the key actions in the three-dimensional space; the pre-trained three-dimensional convolutional posture recognition model is represented; the three-dimensional convolutional posture recognition model is a deep learning model that can process time-space-posture information and is used to analyze the key change patterns in continuous motion; the kinematic constraint conditions represent the action range, speed, angle and other limiting conditions allowed by the sports physiology and structure; the candidate interaction nodes refer to the potential key action points that meet the kinematic constraint conditions and may have tactical or rule significance in the athlete's action trajectory; the rule library of the current sports event represents the standard knowledge system that defines the key technical actions, spatial regions and tactical relationships in the current sports event; the motion interaction nodes refer to the key points associated with the competition rules during the sports process, such as "passing the ball → receiving the ball", "taking off → landing".

[0067] In this embodiment, determining the action evolution trend of each motion interaction node according to the action tendency trajectory of the athlete can be achieved by the following steps:

[0068] Set the action tendency trajectory of the athlete based on the rule library of the current sports event;

[0069] Determine the dynamic displacement range corresponding to the athlete's action according to the action tendency trajectory;

[0070] Determine the trajectory evolution characteristics when the athlete's action changes through the dynamic displacement range;

[0071] Determine the action evolution trend of each motion interaction node from the trajectory evolution characteristics.

[0072] Specifically, in implementation, first, based on the rule library of the current sports event and the athlete's historical action data, set the action tendency trajectory of the athlete. The rule library of the current sports event includes information such as standard action sequences, action conversion requirements, and tactical characteristics for different sports events; for example, in the basketball event, the rule library will set the paths of common actions such as shooting, passing, and dribbling. By analyzing the athlete's historical action data, deduce the action tendency trajectory of the athlete in a certain situation, that is, the probability, starting position, ending position, and transition mode of the athlete performing a specific action. Then, in combination with the action tendency trajectory, use a physical model or a data-driven method to determine the dynamic displacement range of the athlete when performing an action; for example, use a gait analysis model to calculate the movement range of the athlete in actions such as running and jumping, combine the athlete's speed, acceleration, and the field size defined in the rule library, estimate the spatial displacement limit of each action stage, and use the spatial displacement limit as the dynamic displacement range. Then, by combining the dynamic displacement range and the action tendency trajectory, analyze the trajectory evolution characteristics of the athlete when performing an action. The time series analysis method, such as the hidden Markov model or the trajectory prediction model based on neural networks, can be used to model the athlete's movement trajectory. Using the dynamic displacement range limit, divide the athlete's trajectory evolution process into different stages (such as the preparation stage, execution stage, and recovery stage), and calculate key characteristics such as the trajectory change rate, acceleration change, and turning frequency in each stage. Use the calculated characteristics as the trajectory evolution characteristics. Finally, combine the trajectory evolution characteristics with the time and spatial positions of the motion interaction nodes to calculate the action evolution trend of each motion interaction node, which can be achieved by methods such as regression analysis and prediction models, such as the trajectory prediction model based on deep learning, to predict the transition mode between action nodes. For example, in a football game, there is an acceleration stage between the action nodes of the athlete from starting to receiving the ball, and the trajectory evolution trend analysis can reveal the key conversion points during the athlete's acceleration process.

[0073] It should be noted that in this application, the action tendency trajectory represents the action path and transformation mode adopted by an athlete during a competition according to the rule library, historical behavior data, and the current competition situation; the dynamic displacement range represents the spatial range of the change in the positions of various joints or the overall position of the athlete's body when performing a certain action; the trajectory evolution feature represents the change feature of the trajectory from the initial state to the end state during the athlete's action process; the action evolution trend represents the process of the athlete smoothly transitioning from one action state to another during the movement process.

[0074] In step S3, the real-time state data of the athlete in the current sports event is obtained, the real-time state data is subjected to feedback evaluation to obtain the athlete's motion feedback strategy, and then the action linkage attribute of the athlete is determined from the motion feedback strategy and the motion posture information of the athlete in the current sports event.

[0075] Specifically, the real-time state data of the athlete in the current sports event can be obtained in the following way: First, a multi-sensor fusion system is arranged, which consists of wearable devices (such as IMU sensors, accelerometers, gyroscopes) worn by the athlete and a camera network on the field; the sensors collect the athlete's physiological data (such as heart rate, breathing rate, muscle activity, etc.) and kinematic data (such as acceleration, speed, angle change, etc.) in real time, and the collected data can be transmitted wirelessly to the data processing center; at the same time, the cameras on the field capture the action postures of the athlete, and computer vision technology is used for real-time posture estimation to obtain the joint point positions of the athlete (such as head, limbs, torso, etc.) and their position changes in three-dimensional space, and the sensor data and visual data are fused, and data synchronization technology (such as time series synchronization, timestamp matching, etc.) is used to integrate the physiological state, motion trajectory, and posture information of the athlete. Finally, through the data processing and analysis platform, the collected data is converted into a real-time state representation of the athlete, including indicators such as current speed, position, action accuracy, fatigue degree, etc., that is, the real-time state data of the athlete in the current sports event.

[0076] It should be noted that in this application, the real-time state data refers to the data about the athlete's physiological state and motion behavior obtained in real time through sensors and the visual system during the competition.

[0077] Preferably, in this embodiment, the real-time state data is subjected to feedback evaluation to obtain the athlete's motion feedback strategy, referring to Figure 3 As shown in the figure, which is a schematic flowchart of determining the motion feedback strategy in some embodiments of this application. The motion feedback strategy in this embodiment can be implemented by the following steps:

[0078] In step S31, based on the exercise performance index and injury risk level in the real-time status data, a dynamic feedback evaluation network model of the athlete is constructed;

[0079] In step S32, the dynamic feedback evaluation network model is quantified by fuzzy logic to generate action feedback evaluation indicators;

[0080] In step S33, according to the action feedback evaluation indicators, exercise strategy matching is performed to obtain a candidate strategy set;

[0081] In step S34, the exercise feedback strategy of the athlete is selected through the candidate strategy set.

[0082] When specifically implemented, first, by analyzing the real-time status data of the athlete, the exercise performance index and injury risk level are extracted. Among them, the exercise performance index includes speed, explosive power, and endurance, and the injury risk level includes joint pressure, fatigue degree, and overload. The exercise performance index and injury risk level are obtained by combining sensor data and a calculation model. Next, a dynamic feedback evaluation network model is constructed. This model is based on a multi-layer neural network or regression analysis and combines the historical data of the athlete to update the evaluation results of the exercise performance index and injury risk level in the real-time status data in real time. Then, in the dynamic feedback evaluation network model, fuzzy logic technology is applied to quantify the evaluation results. The fuzzy logic system can handle the uncertainty and ambiguity in the athlete's status data, such as the volatility of the exercise performance index or injury risk level. By setting fuzzy rules, for example, "If the heart rate is too high and the fatigue degree is too large, then the injury risk is high", the athlete's status data is converted into quantified action feedback evaluation indicators, such as action quality scores, fatigue indices, recovery period suggestions, etc. Then, through a matching algorithm, where the matching algorithm can be similarity-based matching, decision trees, or optimization algorithms, the current status of the athlete is compared with the preset exercise strategies. Each strategy corresponds to a different athlete status, such as high-efficiency training strategies, recovery strategies, protection strategies, etc. The process of strategy matching considers factors such as the athlete's physiological state, game progress, and venue conditions to find the most suitable exercise strategy for the current status, that is, the candidate strategy set. Finally, according to the candidate strategy set, considering factors such as the athlete's historical performance, current game situation, and the coach's tactical requirements, the final strategy selection is made, that is, the exercise feedback strategy. The final exercise feedback strategy can be determined through priority-based selection, maximum benefit selection, or weight evaluation methods; for example, if the athlete's fatigue degree is relatively high, then a recovery or mild training strategy is preferentially selected; if the athlete's status is good, then a high-intensity training or offensive tactical strategy can be selected.

[0083] It should be noted that in this application, the motion efficiency index represents the efficiency and performance of quantifying the actions performed by athletes; the injury risk level represents the risk level of evaluating the injuries suffered by athletes; the dynamic feedback evaluation network model refers to a multi-index fusion calculation model for real-time evaluation of the motion efficiency and injury risk status of athletes; the action feedback evaluation index refers to a quantitative parameter reflecting the action quality and fatigue degree of athletes; the candidate strategy set refers to a set of training and competition strategies selected for athletes suitable for the current situation; the motion feedback strategy refers to an optimized strategy selected by athletes according to real-time state data and competition requirements.

[0084] In this embodiment, the determination of the action linkage attribute of an athlete from the motion feedback strategy and the motion posture information of the athlete in the current sports event can be implemented by the following steps:

[0085] Obtain the motion posture information of the athlete in the current sports event;

[0086] Extract the motion trajectory characteristics of the athlete from the motion posture information;

[0087] Determine the action delay deviation of the athlete according to the motion trajectory characteristics;

[0088] Determine the action constraint conditions of the athlete in the current sports event according to the motion feedback strategy;

[0089] Determine the action linkage attribute of the athlete through the action delay deviation and the action constraint conditions.

[0090] In specific implementation, first, three-dimensional pose data of the athlete is obtained through a computer vision system. The computer vision system is based on joint modeling of multiple cameras and uses pose estimation algorithms such as OpenPose, MediaPipe, or 3D convolutional neural network models to output the position coordinates of each joint of the athlete in three-dimensional space in real time. The data of these position coordinates forms a continuous pose sequence, and the continuous pose sequence is used as the motion pose information. Next, based on the motion pose information, trajectory modeling is performed on the joint coordinate points, and parameters such as their position, speed, acceleration, and angular velocity are extracted. The Kalman filter can be used to smooth the trajectory, the Fourier transform can be used to extract frequency features, and the spline curve can be used to fit the motion trajectory form. The extracted frequency features and the motion trajectory form fitted by the spline curve are used as the motion trajectory features. Again, the motion trajectory features are aligned with the standard action trajectory on the time axis, and dynamic time warping or cross-correlation analysis is used to calculate the time offset of each key joint action. The calculated time offset is used as the action delay deviation of the athlete. Then, the restrictions on aspects such as action intensity, angle range, and rhythm frequency in the motion feedback strategy are parsed, and this restriction is used as the action constraint condition for the athlete in the current sports event. For example, if the motion feedback strategy is "reduce the load on the knee joint", the maximum flexion and extension angle of the knee joint is limited to 80% of the standard action. This limit standard can be set according to expert experience or sports physiological index management, and is not limited here. Finally, combining the action delay deviation and the action constraint condition, the time synchronization and structural cooperation relationship between multiple joints are analyzed. Multi-channel time series clustering is used to determine whether the joint actions are within the allowed delay range and meet the action standards set by the motion feedback strategy at the same time. If more than two joints always cooperate within the range allowed by the motion feedback strategy, it can be defined as "high linkage", that is, the action linkage attribute of the athlete is obtained.

[0091] It should be noted that in this application, the motion pose information refers to the sequence of pose data of each joint point of the athlete in three-dimensional space within continuous time; the motion trajectory feature refers to the quantitative description reflecting the joint movement path and its change rate; the action delay deviation refers to the time error between the timing performance of the key part in the actual action trajectory and the ideal action template; the action constraint condition refers to the control parameter for the execution of the index during the motion process; the action linkage attribute refers to the collaborative relationship in time and space among multiple joint parts of the body during the execution of the action.

[0092] In step S4, the action evolution trend and the action linkage attribute are input into a preset action prediction model for real-time training, and the output action prediction data is visually displayed.

[0093] In this embodiment, the real-time training of inputting the action evolution trend and the action linkage attribute into a preset action prediction model can be implemented by the following steps:

[0094] Generate a dynamic evolution input sequence according to the probability distribution and trend verification index in the described action evolution trend;

[0095] Through the spatio-temporal convolution-attention hybrid network layer of the preset action prediction model, perform multi-scale feature fusion and dynamic weight allocation on the dynamic evolution input sequence, and calculate the causal association strength between each linkage node;

[0096] Perform incremental adversarial training on the preset action prediction model according to the causal association strength, and output action prediction data.

[0097] When specifically implemented, first, extract the time series features between action nodes from the identified action evolution trend, including the probability distribution and trend verification index in the action evolution trend. Among them, the probability distribution includes the possibility of an action state evolving to the next state. Construct the probability distribution and trend verification index in the action evolution trend into a time-continuous input sequence, retain the time series dependence relationship and action state labels, so that it has a structured input format that can be used for deep model learning. Then, input the generated dynamic evolution input sequence into the preset action prediction model. This action prediction model adopts a spatio-temporal convolution-attention hybrid network structure, that is, combines a temporal convolutional neural network and a multi-head attention mechanism; in the spatio-temporal convolution part, extract multi-scale local features of the action evolution between the time axis and space coordinates, and dynamically allocate the influence weights of different nodes on the prediction target in the attention layer. Especially for the input action linkage attribute, the action prediction model will focus on the causal influence path between each linkage node and calculate its causal association strength, that is, the influence weight between two nodes under historical data. This influence weight can be represented by Granger causality or information flow intensity, which is not limited here. Finally, in the model training process, adopt an incremental adversarial training mechanism, that is, the model trains while running and gradually updates its parameters. This incremental adversarial training mechanism combines mini-batch online learning and adversarial perturbation samples to improve robustness. When each new input data arrives, adjust the model attention weight according to the current causal association strength, and then use the error between the real data and the prediction result to update the model parameters in reverse. The action prediction model outputs a set of action prediction data every time it iterates a batch of data, including the action state evolution results at several future moments, that is, the action prediction data is obtained.

[0098] It should be noted that in this application, the probability distribution represents a set of probability weights describing the evolution of a certain action state to subsequent states; the trend verification index is a quantitative parameter used to measure the trend stability and prediction confidence; the dynamic evolution input sequence refers to the action trend and probability feature data organized in chronological order; the spatio-temporal convolution-attention hybrid network layer of the preset action prediction model represents a deep neural network structure that fuses temporal convolution (extracting time dependencies) and attention mechanism (capturing key linkage relationships); the causal association strength refers to the degree of influence of a motion node on the state evolution of another node; the action prediction data refers to the prediction results of the action prediction model regarding the changes in the athlete's action state within a future time window.

[0099] In specific implementation, the visualization display of the output action prediction data can be achieved in the following way: that is, the output action prediction data is structured, and the key parameters included are extracted, including the prediction time, action category, three-dimensional joint coordinates, speed change value, action probability confidence, etc. Then, based on the three-dimensional skeleton driving technology, the predicted joint coordinates are mapped into the human body posture model to form a continuous virtual action sequence. This process can be realized with the help of three-dimensional graphics engines such as Three.js or Unity, and the predicted actions are made smooth and visible through animation interpolation and smooth transition algorithms; in addition, to improve the information expression density, trajectory lines, action trend arrows and color coding can be superimposed in the three-dimensional view, such as using the depth of color to represent the action intensity and using dynamic lines to represent the action path direction. At the same time, combined with the confidence index of the action node, the uncertainty area of the prediction result is marked through a heat map or transparency change. Finally, the prediction animation, numerical indicators and trend charts are presented through an interactive front-end interface to complete the visualization display of dynamic sports data.

[0100] Thus, in this application, the accuracy of dynamic sports data analysis and the decision-making assistance ability are improved; among them, by capturing game behavior data, non-contact high-precision real-time acquisition of the complex posture actions of athletes is realized, ensuring the comprehensiveness, dynamics and continuity of the data source; by performing posture recognition on key action information, the key posture change nodes in the movement process can be accurately located, the structured evolution logic behind the movement behavior can be recognized, and the accuracy and timeliness of action recognition and behavior understanding are improved; by analyzing the real-time state data and fusing the real-time physiological state and movement posture information, the constraint relationship between the movement strategy and the action response is dynamically calculated, the intelligent recognition of the action linkage attribute is realized, and the understanding ability of the visualization system for the cooperation mechanism is improved; by visualizing the action prediction data, the real-time prediction and visualization presentation of the action trend can be realized, the forward-looking and interactivity of sports data analysis are improved, and at the same time, the three-dimensional visualization process is driven by multi-dimensional prediction results, and the action coordination change process and the causal relationship between nodes are presented in real time, enhancing the interactive display efficiency and linkage analysis effectiveness of sports dynamic data.

[0101] In summary, the technical solution adopted in this application can perform associated analysis on the dynamic sports data of athletes in a data environment with real-time capture capabilities and multi-modal perception basis to improve the effectiveness of visual processing of sports dynamic data for collaborative interaction.

[0102] Embodiment 2. This application provides a dynamic sports data visualization system. Referring to Figure 4 as shown, this figure is a module structure diagram of a dynamic sports data visualization system according to this embodiment of this application. The data visualization system includes:

[0103] A data acquisition module 100, configured to perform real-time capture on the action postures of athletes in a sports event based on computer vision technology to obtain the competition behavior data of athletes in the sports event;

[0104] A posture recognition module 200, configured to extract key action information of athletes in a sports event from the competition behavior data, perform posture recognition on the key action information to obtain multiple motion interaction nodes of the athletes in the current sports event, and determine the action evolution trend of each motion interaction node according to the action tendency trajectory of the athletes;

[0105] A feedback evaluation module 300, configured to obtain the real-time state data of athletes in the current sports event, perform feedback evaluation on the real-time state data to obtain the sports feedback strategy of the athletes, and further determine the action linkage attribute of the athletes from the sports feedback strategy and the motion posture information of the athletes in the current sports event;

[0106] A visualization display module 400, configured to input the action evolution trend and the action linkage attribute into a preset action prediction model for real-time training, and visually display the output action prediction data.

[0107] This application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of this application. It should be understood that each process and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of processes and / or blocks in the flowcharts and / or block diagrams can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0108] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The storage medium includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc memories, magnetic disk memories, tape memories, or any other medium that can be used to carry or store data and is computer-readable.

[0109] It should also be noted that the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent in such process, method, commodity or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, commodity or device including the element.

Claims

1. A method for visualizing dynamic sports data, characterized in that, The data visualization method includes the following steps: Based on computer vision technology, the action postures of athletes in sports events are captured in real time to obtain the competition behavior data of athletes in sports events; Extract the key action information of athletes in sports events from the competition behavior data, perform posture recognition on the key action information, obtain multiple motion interaction nodes of athletes in the current sports event, and determine the action evolution trend of each motion interaction node according to the action tendency trajectory of the athletes; Obtain the real-time state data of athletes in the current sports event, perform feedback evaluation on the real-time state data to obtain the motion feedback strategy of the athletes, and then determine the action linkage attribute of the athletes from the motion feedback strategy and the motion posture information of the athletes in the current sports event; Input the action evolution trend and the action linkage attribute into a preset action prediction model for real-time training, and visually display the output action prediction data; Among them, inputting the action evolution trend and the action linkage attribute into a preset action prediction model for real-time training specifically includes: Generate a dynamic evolution input sequence according to the probability distribution and trend verification index in the action evolution trend; Through the spatio-temporal convolution-attention hybrid network layer of the preset action prediction model, perform multi-scale feature fusion and dynamic weight allocation on the dynamic evolution input sequence, and calculate the causal association strength between each linkage node; Perform incremental adversarial training on the preset action prediction model according to the causal association strength, and output action prediction data; Among them, determining the action linkage attribute of the athletes from the motion feedback strategy and the motion posture information of the athletes in the current sports event specifically includes: Obtain the motion posture information of the athletes in the current sports event; Extract the motion trajectory features of the athletes from the motion posture information; Determine the action delay deviation of the athletes according to the motion trajectory features; Determine the action constraint conditions of the athletes in the current sports event according to the motion feedback strategy; Determine the action linkage attribute of the athletes through the action delay deviation and the action constraint conditions, where the action linkage attribute refers to the temporal and spatial coordination relationship of multiple joint parts of the body during the execution of actions.

2. The dynamic sports data visualization method according to claim 1, wherein The competition behavior data is a structured data set describing the action postures and spatio-temporal trajectories of athletes during the competition process.

3. The dynamic sports data visualization method according to claim 1, characterized in that, Performing posture recognition on the key action information to obtain multiple motion interaction nodes of athletes in the current sports event specifically includes: Obtain the motion posture features in the key action information; Based on a pre-trained three-dimensional convolutional posture recognition model, perform motion trajectory clustering analysis on the motion posture features to detect candidate interaction nodes that meet the kinematic constraint conditions of athletes in the current sports event; Resolve conflicts for the candidate interaction nodes in combination with the rule base of the current sports event to generate multiple motion interaction nodes of athletes in the current sports event.

4. A dynamic sports data visualization method according to claim 1, characterized in that, Determining the action evolution trend of each motion interaction node according to the action tendency trajectory of the athletes specifically includes: Set the action tendency trajectory of the athletes based on the rule base of the current sports event; Determine the dynamic displacement range corresponding to the athlete's action according to the action tendency trajectory; Determine the trajectory evolution characteristics when the athlete's action changes through the dynamic displacement range; Determine the action evolution trend of each motion interaction node from the trajectory evolution characteristics.

5. A dynamic sports data visualization method according to claim 1, characterized in that The action evolution trend represents the process of the athlete smoothly transitioning from one action state to another during the movement process.

6. The dynamic sports data visualization method according to claim 1, characterized in that The real-time state data refers to the data about the athlete's physiological state and movement behavior obtained in real time through sensors and vision systems during the competition.

7. A dynamic sports data visualization method according to claim 1, characterized in that, The action prediction data refers to the prediction results of the action prediction model on the change of the athlete's action state within a future time window.

8. A dynamic sports data visualization system for performing a dynamic sports data visualization method according to any one of claims 1 to 7, characterized in that, The data visualization system includes: A data acquisition module, configured to capture the action postures of athletes in sports events in real time based on computer vision technology, and obtain the competition behavior data of athletes in sports events; A posture recognition module, configured to extract the key action information of athletes in sports events from the competition behavior data, perform posture recognition on the key action information to obtain multiple motion interaction nodes of the athlete in the current sports event, and determine the action evolution trend of each motion interaction node according to the action tendency trajectory of the athlete; A feedback evaluation module, configured to obtain the real-time state data of the athlete in the current sports event, perform feedback evaluation on the real-time state data to obtain the athlete's motion feedback strategy, and further determine the action linkage attribute of the athlete from the motion feedback strategy and the motion posture information of the athlete in the current sports event; A visualization display module, configured to input the action evolution trend and the action linkage attribute into a preset action prediction model for real-time training, and visually display the output action prediction data.

Images