Real-time AI evaluation system for tennis based on multi-view visual feature fusion

Abstract

The application relates to the technical field of sports competition analysis, and discloses a tennis real-time AI evaluation system based on multi-view visual feature fusion, which comprises the following steps: collecting multi-angle video stream data of a tennis match; fusing multi-view data to generate space-time feature representation by using a time series graph convolution network; constructing a time series causal graph model based on the fused space-time features; identifying key turning points of the match by using graph structure analysis and event influence evaluation; verifying the influence of the turning points by using counterfactual reasoning technology; and generating personalized tactical suggestions in combination with a player feature model. By constructing a time series causal graph model and identifying key turning points, the application realizes the mining and understanding of deep causal mechanisms of the match, and provides more accurate technical and tactical analysis and decision support for coaches and players.

Description

Technical Field

[0001] This invention relates to the field of sports competition analysis technology, and more specifically, to a real-time AI evaluation system for tennis based on multi-view visual feature fusion. Background Technology

[0002] With the development of data science and computer vision technology, sports analysis is gradually shifting from traditional experience-based subjective analysis to data-driven scientific analysis. In high-speed, complex competitive sports like tennis, accurately capturing and analyzing key events and turning points in the game is crucial for improving athlete performance and game strategies.

[0003] Existing tennis match analysis technologies mainly focus on basic data statistics and single-view video analysis. These methods have the following obvious shortcomings: First, they lack effective integration of multi-view data, resulting in an incomplete understanding of the situation on the court; second, they only focus on single event identification and basic action analysis, lacking a deep understanding of the temporal causal relationships during the match; third, they cannot effectively identify and explain key turning points that affect the course of the match; fourth, in the complex and ever-changing match environment, they cannot accurately capture the dynamic causal chains between events; and fifth, they cannot provide coaches and players with in-depth tactical insights and targeted improvement suggestions.

[0004] Currently, while some studies attempt to analyze key events in sports competitions using statistical models or rule systems, these methods often struggle to capture complex temporal relationships and causal mechanisms, resulting in limited accuracy and a lack of interpretability, failing to provide in-depth tactical guidance. Traditional video analysis tools are often limited to a single perspective, making it difficult to comprehensively grasp the game scenario, leading to biased or inaccurate analysis results. Furthermore, existing technologies generally lack the ability for deep causal analysis, making it difficult to reveal the underlying reasons for changes in the course of a game, thus limiting their application value in high-level competitive sports.

[0005] Therefore, there is an urgent need for a real-time tennis assessment method that can integrate multi-perspective visual features, deeply explore the causal mechanisms of matches, accurately identify key turning points, and provide personalized tactical suggestions to meet the needs of modern competitive sports for scientific training and match analysis. Summary of the Invention

[0006] This invention provides a real-time AI evaluation method for tennis based on multi-view visual feature fusion, which solves the technical problems in related technologies such as difficulty in deeply understanding the causal relationship of the game, inability to accurately identify key turning points, and lack of targeted tactical suggestions.

[0007] This invention discloses a real-time AI evaluation method for tennis based on multi-view visual feature fusion, comprising the following steps: collecting multi-angle video stream data of tennis matches; fusing multi-view data using a temporal graph convolutional network to generate spatiotemporal feature representations; constructing a temporal causal graph model based on the fused spatiotemporal features, where nodes represent match events and edges represent potential causal relationships between events; identifying key turning points in the match using graph structure analysis and event impact assessment; verifying the impact of turning points using counterfactual reasoning techniques to generate a quantitative impact assessment; and generating personalized tactical suggestions by combining player feature models. The temporal causal graph model uses a dynamic causal strength function to quantify the causal relationships between events and forms a sparse causal relationship network through adaptive threshold filtering.

[0008] Furthermore, the temporal graph convolutional network includes spatial graph convolutional layers and temporal convolutional layers, wherein: the spatial graph convolutional layers are used to capture spatial relationships between nodes; the temporal convolutional layers are used to capture temporal dependencies; and the network training stability is improved through residual connections and layer normalization.

[0009] Furthermore, the temporal cause-effect graph model has a hierarchical structure, including: the bottom layer consists of basic event nodes, representing the hitting action, player movement, and ball landing point, etc.; the middle layer consists of tactical combination nodes, representing the tactical sequence formed by multiple consecutive basic events; and the top layer consists of tactical effect nodes, representing the result of tactical execution.

[0010] Furthermore, the steps for identifying key turning points in the competition include: calculating the in-degree, out-degree, and various centrality indicators of nodes; calculating the change in graph structural centrality after removing nodes; calculating the event influence index based on conditional probability and importance weight; ranking nodes based on their importance and event influence, and identifying events exceeding a threshold as key turning points.

[0011] Furthermore, the counterfactual reasoning technique includes: constructing counterfactual scenarios where the turning point does not occur or occurs in a different manner; removing or modifying corresponding nodes in the causal sequence graph; using a probability propagation algorithm to predict the probability distribution of events occurring in the counterfactual scenarios; and calculating the difference between the actual result and the counterfactual result, which is defined as the causal effect of the turning point.

[0012] Furthermore, the player characteristic model includes: technical characteristics, representing the distribution of serving positions, the distribution of hitting power, and the distribution of landing points; tactical characteristics, representing commonly used tactical combinations and strategic adjustments to deal with different opponents; and psychological characteristics, representing performance on key points and ability to cope with pressure.

[0013] Furthermore, the steps for generating personalized tactical recommendations include: identifying effective tactical sequences that lead to scoring through causal analysis; identifying key technical or tactical problems that lead to losing points; analyzing the opponent's technical and tactical patterns and countermeasures; and using a multi-objective optimization algorithm to comprehensively consider success rate, risk level, and player execution ability to generate the most suitable tactical recommendations.

[0014] Furthermore, the personalized tactical suggestions include: suggestions for adjusting the distribution of ball landing points; suggestions for optimizing serving strategies; suggestions for adjusting technical movements; and suggestions for controlling the pace of the game.

[0015] Furthermore, it also includes the step of forming visualized tactical decision support information, which includes: tactical effectiveness evaluation charts; advantage area heat maps; time-series tactical recommendations; and real-time decision prompts.

[0016] The beneficial effects of this invention are as follows: Through the innovative combination of multi-view visual feature fusion, temporal causal graph construction, and turning point identification algorithms, it solves the technical problem that traditional tennis analysis systems cannot deeply understand causal relationships and identify key turning points, achieving the following significant technical effects: 1. It enables the mining and understanding of deep causal mechanisms during the game, allowing coaches and players to understand the essential reasons behind the game's trajectory; 2. Through innovative turning point identification algorithms and counterfactual reasoning techniques, it accurately identifies and quantifies the impact of key turning points, providing interpretable causal analysis; 3. It supports millisecond-level real-time event identification and analysis, providing immediate tactical adjustment suggestions during the game; 4. It generates highly personalized tactical suggestions, directly serving the practical needs of coaches and players; 5. Through temporal graph convolution technology, it achieves effective fusion of multi-view video data, providing more comprehensive and accurate basic data for analysis. Attached Figure Description

[0017] Figure 1 This is an overall flowchart of the present invention;

[0018] Figure 2 This is a flowchart of the multi-view visual feature acquisition and fusion steps of the present invention;

[0019] Figure 3 This is a flowchart of the time-series cause-effect graph construction steps of the present invention;

[0020] Figure 4 This is a flowchart of the inflection point identification algorithm steps of the present invention;

[0021] Figure 5 This is a flowchart of the counterfactual reasoning and verification steps of the present invention;

[0022] Figure 6 This is a flowchart of the personalized tactical decision support steps of the present invention. Detailed Implementation

[0023] This implementation method is applied to the field of tennis match analysis, particularly for scenarios requiring a deep understanding of the causal relationships between events during a match and the identification of key turning points. Existing tennis analysis systems mainly suffer from the following technical problems:

[0024] It focuses only on single event recognition and basic action analysis, lacking a deep understanding of the temporal and causal relationships during the competition;

[0025] Unable to effectively identify and explain key turning points that influenced the course of the game;

[0026] In a complex and ever-changing competition environment, it is impossible to accurately capture the dynamic causal chain between events;

[0027] They are unable to provide coaches and players with in-depth tactical insights and targeted improvement suggestions.

[0028] This embodiment provides a real-time AI evaluation method for tennis based on multi-view visual feature fusion. By constructing a time-series causal graph model and implementing a turning point identification algorithm, it solves the above-mentioned technical problems, realizes the mining and understanding of the deep causal mechanism of the game, and provides coaches and players with more in-depth technical and tactical analysis and targeted improvement suggestions.

[0029] The method of this embodiment includes:

[0030] Step 1: Multi-view visual feature acquisition and fusion

[0031] A temporal graph convolutional network is used to process multi-angle video streams and generate a fused spatiotemporal feature representation.

[0032] Specifically, it includes the following sub-steps:

[0033] 1.1 Install high frame rate cameras at both ends of the tennis court to collect multi-angle video stream data of the match. Each camera records at a rate of 60 frames per second to ensure that the image clarity is not less than 1080p.

[0034] 1.2 Apply target detection algorithms to process the video stream, extract the position information and motion trajectory data of players, ball and field elements, and form a preliminary spatiotemporal coordinate sequence;

[0035] 1.3 Constructing the sequence graph structure G st =(V st E st ,F node ), where node V st Indicates the positions of the player and the ball, side E st F represents the interaction relationship between nodes. node This is the node feature vector. The node feature vector contains position coordinates, velocity vector, acceleration vector, and local visual features;

[0036] 1.4 Applying a Temporal Graph Convolutional Network (ST-GCN) to fuse multi-view data. This network uses temporal graph convolutional layers to aggregate and update node features at different time points. The convolution operation is defined as follows:

[0037]

[0038] in, Let N(v) represent the feature vector of node v in the l-th layer, and let N(v) be the set of neighbors of node v. v,u W is the normalization constant. (l) and b (l) σ is a learnable parameter, and σ is the activation function.

[0039] The temporal graph convolutional network in this embodiment consists of the following structure:

[0040] Spatial graph convolutional layer: captures the spatial relationships between nodes and performs convolution operations on the graph at each time step.

[0041] Temporal convolutional layer: Uses one-dimensional convolution to capture temporal dependencies, with a window size of 3.

[0042] Residual connections: Adding residual connections every two layers improves the training performance of deep networks.

[0043] Layer normalization: Apply layer normalization after each convolutional layer to improve training stability.

[0044] The network consists of nine graph convolutional layers. The input is the original node features, and the output is a fused spatiotemporal feature representation. In real-time tennis match analysis, the network processes input data at a rate of 30 frames per second, accurately capturing motion trajectories and relationships during complex movements such as rapid player movement and turns.

[0045] Through the above sub-steps, multi-view video data is transformed into a unified spatiotemporal graph feature representation, providing a foundation for subsequent temporal causal analysis. The output of this step is a spatiotemporal feature tensor that integrates multi-view visual information. Where N represents the number of nodes, T len D represents the time step and D represents the feature dimension.

[0046] Step 2: Constructing a Temporal Cause-Effect Graph

[0047] Based on the fused spatiotemporal feature tensor, a temporal causal inference algorithm is applied to generate a temporal causal graph model of tennis matches.

[0048] Specifically, it includes the following sub-steps:

[0049] 2.1 Based on the spatiotemporal feature tensor F output in step 1st An event recognition model is used to detect various key events in a tennis match, such as serves, returns, and points of attack, forming an event sequence E. seq ={e1,e2,...,e n}, each event e i Includes event type, occurrence time, and related characteristics;

[0050] 2.2 Constructing the Temporal Cause-Effect Graph Model G tc =(V e E c ,T dim ),in:

[0051] Node set V e This represents various events detected from the video. Each node contains the event type, timestamp, and feature vector edge set E. c Representing a potential causal relationship, by a directed edge (e i ,e j ) represents event e i Event e may occur j

[0052] T dim Represents the time dimension, used to record the timing information of events.

[0053] The temporal cause-effect graph model is implemented in a hierarchical structure in a real tennis match:

[0054] The bottom layer consists of basic event nodes, such as the hitting action, player movement, and ball landing point.

[0055] Middle layer: Composed of tactical combination nodes, representing a tactical sequence formed by multiple consecutive basic events.

[0056] High-level: Composed of tactical effect nodes, representing the results of tactical execution, such as points scored, points conceded, and shifts in advantage.

[0057] In practical applications, the model can capture tactical causal chains such as "the serve lands near the sideline → the opponent's return ball quality decreases → the attacker gains an offensive opportunity → a point is scored".

[0058] 2.3 The dynamic causal strength function is applied to calculate the strength of the causal relationship between events. The function is defined as follows:

[0059]

[0060] in:

[0061] I c (e i ,e j ,Δt) represents event e i For event ej causal influence strength

[0062] P(e j |e i ,Δt) represents the time e i Event e occurs within a time interval Δt after it occurs j Conditional probability of occurrence

[0063] exp(-β·Δt) is the time decay factor, and β is the decay coefficient.

[0064] α is the causal weighting coefficient.

[0065] In practical implementation, the conditional probability P(e j |e i Δt) is calculated using a time-conditional model based on an attention mechanism. This model considers event type, time interval, and contextual features and is trained using historical match data. The time decay coefficient β is dynamically adjusted according to different event types, with a larger β value (rapid decay) used for short-term tactical events and a smaller β value (slow decay) used for long-term strategic events.

[0066] 2.4 Based on the aforementioned causal strength function, a complete temporal causal graph is constructed, retaining edges with causal strength greater than a threshold θ in the graph to form a sparse causal relationship network. In practical applications, the threshold θ is set as an adaptive value, dynamically adjusted according to the stage of the competition and its importance.

[0067] The output of this step is a directed graph structure that represents the temporal causal relationships of events in a tennis match. Each edge has a corresponding causal strength value, which is used to quantify the degree of causal influence between events.

[0068] Step 3: Turning Point Recognition Algorithm

[0069] Based on the constructed temporal causal graph, graph structure analysis and event impact assessment algorithms are applied to identify and quantify key turning points in the competition.

[0070] Specifically, it includes the following sub-steps:

[0071] 3.1 Analyze the topological structure of the time-series cause-effect graph, paying particular attention to the following characteristics:

[0072] The in-degree and out-degree of a node reflect the scope and degree of influence of an event.

[0073] Node centrality includes degree centrality, betweenness centrality, and eigenvector centrality.

[0074] Community structure: Identifying densely related groups of events

[0075] In the context of a tennis match, this step pays special attention to potentially high-impact events such as consecutive points, saving break points in a service game, and crucial points in a tiebreak. Graph structure analysis is used to determine the position and importance of these events in the causal network.

[0076] 3.2 Calculate the importance index of key nodes, defined as the change in graph structural centrality after removing the node:

[0077] ΔC(G tc ,e)=|C(G tc )-C(G tc \e)|

[0078] in:

[0079] C(G tc ) represents the centrality measure of the original graph.

[0080] C(G tc \e) represents the centrality metric of the graph after removing node e.

[0081] Based on the above calculations, the most important nodes are identified:

[0082] e * =argmax e ΔC(G tc e)

[0083] This algorithm employs a combination of multiple centrality measures, including eigenvector centrality (evaluating the global influence of a node), betweenness centrality (evaluating the importance of a node as a "bridge"), and PageRank (evaluating the importance of a node in a directed graph). In actual tennis match analysis, this algorithm can accurately identify turning points such as "critical breakaways" and "consecutive points creating a psychological advantage."

[0084] 3.3 The impact of an event on the course of the game is assessed using event impact metrics, and event impact is defined as:

[0085]

[0086] in:

[0087] P(e j |e) indicates that event e occurs after event e. j Conditional probability of occurrence

[0088] w j For event e j Importance weight

[0089] α j Adjustment coefficient

[0090] In practical applications, the importance weight w j Predefined based on event type; for example, scoring points have a higher weight than normal rounds, and key balls have a higher weight than normal balls. Adjustment coefficient α. j The system will be dynamically adjusted based on the stage of the match and the score; for example, the impact of events in the deciding set will be given higher weight.

[0091] 3.4 Based on a comprehensive score of node importance and event influence, events are ranked, and events with scores exceeding a threshold τ are identified as key turning points in the competition, forming a set of turning points E. key ={e k |score(e k )>τ}.

[0092] In real-world applications, this algorithm can identify the following turning points:

[0093] Earning two break points in a close score

[0094] Using rare but effective tactical combinations on crucial plays

[0095] Successfully saved the ball and turned the tide under high pressure.

[0096] The output of this step is a set of quantified key turning points in the game, each containing event information, the time of occurrence, an importance score, and a description of its potential impact on the course of the game.

[0097] Step 4: Counterfactual Reasoning and Verification

[0098] By using counterfactual reasoning techniques, the identified turning points are verified and explained, generating causal explanations and quantitative impact assessments.

[0099] Specifically, it includes the following sub-steps:

[0100] 4.1 For each key inflection point e identified in step 3 k ∈E key Constructing counterfactual scenarios S cf (e k ), that is, hypothetical event e k Situations that do not occur or occur in different ways;

[0101] In tennis match analysis, counterfactual scenarios are constructed by considering multiple possibilities. For example, for a key service point, the following counterfactual scenario can be constructed:

[0102] What happens if the serve lands in different areas?

[0103] What if there is a difference in serve speed or spin?

[0104] What if we used different tactical combinations?

[0105] 4.2 Applying a counterfactual reasoning algorithm to simulate the game's development path under counterfactual scenarios. This algorithm is based on a time-series cause-effect graph model and is implemented through the following process:

[0106] Remove or modify node e in the time-series cause-effect graph k

[0107] Based on the modified graph structure, a probability propagation algorithm is used to predict the probability distribution of subsequent events.

[0108] Calculate the game state transition path in a counterfactual scenario

[0109] The mathematical expression of counterfactual reasoning is:

[0110] P(Y cf |do(X=x′),G tc )

[0111] in:

[0112] Y cf Representing the outcome variable in a counterfactual situation

[0113] do(X=x′) represents an intervention operation, setting the variable X to the counterfactual value x′.

[0114] G tc For time-series cause-effect graph models

[0115] The counterfactual reasoning algorithm in this implementation is based on a structured causal model and combines it with a Monte Carlo tree search method for path exploration. During execution, the algorithm maintains consistency between time and event logic, ensuring that the generated counterfactual scenarios conform to the rules and real-world constraints of a tennis match.

[0116] For example, when analyzing a key volley point, the algorithm simulates various possible outcomes if the player chooses a baseline drive instead of a volley and calculates the probability of this decision change affecting the match result.

[0117] 4.3 Calculate the difference between the actual result and the counterfactual result, and define it as the causal effect at the inflection point:

[0118] CE(e k )=M(Y actual )-M(Y cf )

[0119] in:

[0120] M(Y actual ) is a measure of the actual observed results.

[0121] M(Ycf ) is a measure of the outcome of predictions in counterfactual situations.

[0122] In practice, the outcome metric M includes multiple dimensions, such as scoring probability, changes in game advantage, and tactical effectiveness. These metrics are calculated using a prediction model trained on historical game data.

[0123] 4.4 Generate causal explanations and quantitative impact assessment reports for each inflection point, including:

[0124] A detailed description of the turning point (event type, time, relevant players).

[0125] Causal path analysis (how this turning point affected the subsequent development of the game)

[0126] Quantitative impact assessment (the degree of impact on the competition result)

[0127] Visualized causal chain display

[0128] In practical applications, this technology can generate the following analyses:

[0129] "Saving a break point in the 7th game of the third set increased the player's chances of winning by about 35%, making it the most crucial turning point of the match."

[0130] "Three consecutive precise serve placements forced the opponent to change their stance during the return, creating conditions for the subsequent winning points."

[0131] "Tactical adjustments that changed the rhythm of the shot at crucial moments disrupted the opponent's defensive rhythm and were a key factor in turning the tide of the game."

[0132] The output of this step is a set of causal explanation reports on inflection points. Each report contains detailed causal analysis and quantitative impact assessment, enabling coaches and players to clearly understand the impact mechanisms of key events in the game.

[0133] Step 5: Personalized Tactical Decision Support

[0134] Based on the analysis results of the aforementioned steps, combined with historical data, targeted tactical suggestions and decision support information are generated.

[0135] Specifically, it includes the following sub-steps:

[0136] 5.1 Construct a player's technical and tactical characteristics model, which includes the following dimensions:

[0137] Technical characteristics: distribution of serve position, distribution of hitting power, distribution of landing point, etc.

[0138] Tactical characteristics: commonly used tactical combinations, strategic adjustments for dealing with different opponents, etc.

[0139] Psychological characteristics: key performance indicators, stress coping abilities, etc.

[0140] The player technical and tactical characteristic model in this embodiment employs a hybrid feature representation method, combining statistical features and deep learning features. By analyzing players' historical match data, the model automatically extracts technical patterns and habits to construct personalized player profiles. This model can identify players' strengths and weaknesses, such as "preferring to serve in the T-zone" and "prone to unforced errors on backhand returns."

[0141] 5.2 Based on the results of time-series causal analysis and inflection point identification, combined with player characteristic models, identify the technical and tactical advantages and disadvantages in the current match, specifically including:

[0142] Effective tactical combinations: Identifying effective tactical sequences that lead to scores through causal analysis.

[0143] Weaknesses: Identify the key technical or tactical problems that led to lost points.

[0144] Opponent Mode: Analyze the opponent's tactics and strategies.

[0145] In practical applications, the system can identify tactical information in real time, such as "the opponent is prone to sending high-arc balls to the backhand area under pressure" and "errors are likely to occur after more than 5 consecutive baseline rallies", and generate targeted suggestions.

[0146] 5.3 Generate targeted tactical recommendations, including:

[0147] Suggestions for adjusting shot placement: Based on heatmap analysis, more effective shot placement areas are recommended.

[0148] Serving strategy optimization: Based on the opponent's receiving characteristics, it provides optimization suggestions for serving position, speed, and spin.

[0149] Technical adjustments: Specific improvement suggestions for key technical aspects

[0150] Game tempo control: Based on temporal causal analysis, this study provides strategic suggestions for controlling the game tempo.

[0151] The targeted tactical recommendations are generated using a multi-objective optimization algorithm that comprehensively considers success rate, risk level, and player execution ability. This algorithm is trained on a large amount of historical match data and can generate the most suitable tactical recommendations based on the current match situation and player characteristics.

[0152] In specific game scenarios, the system will generate the following suggestions:

[0153] "It is recommended to increase the depth of shots towards the opponent's backhand area to reduce the probability of the opponent's offensive transition."

[0154] "In crucial moments, we recommend employing a high-risk serving strategy, targeting the T-zone."

[0155] "For the second serve, it's recommended to increase spin rather than speed to improve the success rate of the serve."

[0156] 5.4 Generate visualized tactical decision support information, including:

[0157] Tactical effectiveness evaluation chart: Quantitatively displays the success rate and impact of different tactics.

[0158] Advantage area heatmap: Highlights advantageous areas on the field and areas that should be the focus of attack.

[0159] Timing and Tactical Recommendations: Suggestions for Tactical Adjustments at Different Stages of the Match

[0160] Real-time decision prompts: Provides concise and clear tactical hints during the game.

[0161] The visual decision support system employs a multi-screen interactive design, suitable for coaches' tablets on the sidelines and for players to view during breaks. Through color coding, dynamic charts, and concise text, the system provides an intuitive presentation of complex tactical information, ensuring that coaches and players can quickly understand and apply tactical advice.

[0162] The output of this step is a set of personalized tactical decision support information, including specific tactical suggestions, technical adjustment directions, and visual decision aids, which directly serve the practical needs of coaches and players.

[0163] This implementation method achieves the following significant technical effects through a series of innovative technical steps, including multi-view visual feature fusion, temporal causal graph construction, inflection point identification algorithm, counterfactual reasoning and verification, and personalized tactical decision support:

[0164] Deep Causal Mechanism Mining: Breaking through the limitations of traditional tennis analysis systems that only focus on surface phenomena, this study constructs a time-series causal graph model to uncover and understand the deep causal mechanisms in the course of a match, enabling coaches and players to understand the essential reasons behind the match's trajectory.

[0165] High-precision turning point identification: Through innovative turning point identification algorithms, the system can accurately identify key turning points that affect the course of a game, and provide interpretable causal analysis through counterfactual reasoning technology, enabling coaches and players to clearly understand the impact mechanism of key events.

[0166] Real-time analysis and decision support: The system supports millisecond-level real-time event recognition and analysis, and can provide instant tactical adjustment suggestions during the game, meeting the needs of real-time tactical adjustments during the game and significantly improving the scientific level of training and games.

[0167] Personalized tactical optimization: Based on in-depth analysis of player characteristics and understanding of the causal mechanism of the game, the system can generate highly personalized tactical suggestions, including adjustments to the distribution of ball landing points, optimization of serving strategies, and adjustments to technical movements, directly serving the practical needs of coaches and players.

[0168] Multi-dimensional data fusion: Through temporal graph convolution technology, the system effectively fused video data from multiple perspectives, overcoming the limitations of single-perspective data and providing more comprehensive and accurate basic data for subsequent analysis.

[0169] Overall, the real-time AI evaluation method for tennis based on multi-view visual feature fusion provided in this implementation has achieved a technological breakthrough in the field of tennis match analysis, and can provide coaches and players with valuable tactical analysis and decision support in a more in-depth, comprehensive and real-time manner.

[0170] This paper demonstrates the application process and effects of this implementation method through a real-world case study analyzing a high-level tennis match. The analysis focuses on the men's singles quarterfinals of an international tennis tournament, involving two players ranked in the top 20 of the world, A and B.

[0171] This application was conducted on a standard indoor hard court, with four high-definition cameras installed at each end of the court. The cameras captured data at a frame rate of 60 frames per second, with a resolution of 1920×1080 pixels. The camera layout is as follows:

[0172] Table 1. Layout and Functions of Multi-view Cameras

[0173]

[0174]

[0175] Multi-view visual feature fusion

[0176] The system first acquires video streams from eight cameras, extracting and fusing multi-view visual features in real time. Taking a typical serve-to-score rally as an example, the system processing flow is shown in Table 2:

[0177] Table 2 Example of Multi-view Visual Feature Fusion Processing Flow

[0178]

[0179] The graph structure contains 9 nodes (the parts of the two players and the ball), each node has a 15-dimensional feature vector, and the fused feature tensor has a dimension of 9×120×64 (number of nodes × time step × feature dimension).

[0180] Construction of time-series cause-effect graphs

[0181] Based on the fused spatiotemporal features, the system identified 28 key events and constructed a time-series causal graph. Table 3 shows some of the key events and their causal relationships:

[0182] Table 3 Examples of key events in the competition and their causal relationships

[0183]

[0184] Turning point identification and counterfactual reasoning

[0185] The system analyzes the time-series cause-effect graph and identifies five key inflection points. The two most critical inflection points and their counterfactual inference results are shown in Table 4.

[0186] Table 4. Results of Key Turning Point Identification and Counterfactual Reasoning

[0187]

[0188] The counterfactual reasoning of TP2 by the system is as follows: Suppose that player B uses a different serving strategy at the break point (serving towards the backhand instead of the actual forehand), the simulation results show that the probability of player B saving the break point in this game will decrease from the actual 78% to about 31%, which will have an impact of about 41% on the final winning percentage.

[0189] This implementation method was applied to the analysis of 10 tennis matches of different levels, and the main technical effectiveness verification results are as follows:

[0190] Verification of the accuracy of turning point identification

[0191] Table 5 Comparison of Turning Point Identification Accuracy with Existing Technologies

[0192]

[0193] 4.3.2 Real-time performance verification

[0194] Table 6 System Real-Time Performance Indicators

[0195]

[0196] Experimental results show that this implementation can complete the entire process from video input to tactical suggestion generation within 200 milliseconds, meeting the needs of real-time analysis of tennis matches. Compared with existing technologies, this system improves the accuracy of turning point identification by 21.8%-25.6% and the ability to identify key tactical decision points by 26.1%-31.7%, providing coaches and players with more timely and accurate tactical analysis and decision support.

Claims

1. A real-time AI evaluation method for tennis based on multi-view visual feature fusion, characterized in that, Includes the following steps: Collect multi-angle video stream data of tennis matches; Temporal graph convolutional networks are used to fuse multi-view data to generate spatiotemporal feature representations; A temporal causal graph model is constructed based on fused spatiotemporal features, where nodes represent competition events and edges represent potential causal relationships between events; Identify key turning points in the competition using graph structure analysis and event impact assessment; Verify the impact of turning points using counterfactual reasoning techniques to generate a quantitative impact assessment; By combining player characteristic models, personalized tactical suggestions can be generated; Among them, the temporal causal graph model uses a dynamic causal strength function to quantify the causal relationship between events and forms a sparse causal relationship network through adaptive threshold filtering; The steps for identifying key turning points in the competition include: calculating the in-degree, out-degree, and various centrality indices of nodes; calculating the change in graph structural centrality after removing nodes; calculating the event influence index based on conditional probability and importance weight; ranking nodes based on their importance and event influence, and identifying events exceeding a threshold as key turning points. The dynamic causal strength function calculates the strength of the causal relationship between events, and the function is defined as follows: in, Indicates an event Regarding the event The strength of causal influence Indicates in the event After it happened Events within a time period The conditional probability of occurrence The time decay factor, The attenuation coefficient is... The causal weighting coefficient; The importance index of a key node is defined as the change in the graph's structural centrality after removing that node: in, This represents the centrality metric of the original graph. Indicates the removal of a node. The centrality measure of the following figure; Among them, the event influence index is defined as: in, Indicates an event Post-event The conditional probability of occurrence For the event Importance weights This is the adjustment coefficient.

2. The method according to claim 1, characterized in that, The temporal graph convolutional network includes spatial graph convolutional layers and temporal convolutional layers, wherein: Spatial graph convolutional layers are used to capture the spatial relationships between nodes; Temporal convolutional layers are used to capture temporal dependencies; Furthermore, network training stability is improved through residual connections and layer normalization.

3. The method according to claim 1, characterized in that, The time-series cause-effect graph model has a hierarchical structure, including: The bottom layer consists of basic event nodes, representing the hitting action, player movement, and ball landing point; The middle layer consists of tactical combination nodes, representing a tactical sequence formed by multiple consecutive basic events; The high-level section consists of tactical effect nodes, representing the results of tactical execution.

4. The method according to claim 1, characterized in that, The counterfactual reasoning techniques include: Construct counterfactual scenarios where the turning point does not occur or occurs in a different way; Remove or modify the corresponding nodes in the time-series cause-effect graph; The probability propagation algorithm is used to predict the probability distribution of events occurring in counterfactual scenarios. The difference between the actual outcome and the counterfactual outcome is calculated and defined as the causal effect of the inflection point.

5. The method according to claim 1, characterized in that, The player characteristic model includes: Technical characteristics, representing the distribution of serve positions, the distribution of hitting power, and the distribution of landing points; Tactical characteristics, representing commonly used tactical combinations and strategic adjustments for dealing with different opponents; Psychological characteristics indicate key performance indicators and stress coping abilities.

6. The method according to claim 1, characterized in that, The steps to generate personalized tactical recommendations include: Identify effective tactical sequences that lead to high scores through causal analysis; Identify the key technical or tactical issues that led to the loss of points; Analyze the opponent's tactics and strategies; A multi-objective optimization algorithm is used to generate the most suitable tactical suggestions by comprehensively considering success rate, risk level and player execution ability.

7. The method according to claim 1, characterized in that, The personalized tactical recommendations include: Suggestions for adjusting the distribution of shot placement; Suggestions for optimizing serving strategies; Suggestions for adjusting technical maneuvers; Suggestions for controlling the pace of the game.

8. The method according to claim 1, characterized in that, It also includes the step of generating visualized tactical decision support information, which includes: Tactical effectiveness evaluation chart; Heat map of advantageous areas; Timing tactics recommendations; Real-time decision prompts.

9. A real-time AI evaluation system for tennis based on multi-view visual feature fusion, characterized in that, It is used to perform the real-time AI evaluation method for tennis based on multi-view visual feature fusion as described in any one of claims 1-8, comprising: The multi-view video acquisition module is used to acquire multi-angle video stream data of tennis matches; The spatiotemporal feature fusion module is used to fuse multi-view data using a temporal graph convolutional network to generate spatiotemporal feature representations. The temporal causal graph construction module is used to construct a temporal causal graph model based on fused spatiotemporal features; The turning point identification module is used to identify key turning points in a competition by utilizing graph structure analysis and event impact assessment. The counterfactual reasoning module is used to verify the impact of turning points and generate quantitative impact assessments. The tactical decision support module is used to combine player characteristic models to generate personalized tactical suggestions and visualized decision support information.

Images