Real-time technical and tactical identification and strategy recommendation method and system for sports competition

Abstract

The invention relates to the field of artificial intelligence, computer vision, deep learning and sports information processing, and discloses a sports competition real-time technology and tactics identification and strategy recommendation method and system. Comprising the following steps: acquiring and processing competition video data, player motion data, physiological data and competition event data in real time, and generating a unified time sequence data stream; identifying actions, positions and interaction events of players based on the time sequence data stream, and dividing a match into continuous tactical stages; in each tactical stage, constructing a time-space behavior graph by taking players, actions and events as nodes and taking an interaction relationship as an edge; inputting the time-space behavior graph into a graph neural network, aggregating node features through a graph attention mechanism, and outputting an embedded vector representing a current tactical form; and calculating the similarity between the embedded vector and a historical tactical sample library, and generating a tactical recommendation result. According to the method, the modeling accuracy of player behaviors and tactical forms in a complex dynamic game scene is improved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, computer vision, deep learning, and sports information processing, and in particular to a method and system for real-time tactical identification and strategy recommendation in sports competitions. Background Technology

[0002] As the level of athletic competition continues to rise, traditional methods of video review, data collection, and tactical summarization based on manual analysis can no longer meet the demands for rapid and accurate tactical decision-making on the field. This is especially true in team-based competitive sports, where the competition involves a large number of complex dynamic interactions, unexpected events, and tactical execution behaviors, which places extremely high demands on real-time capture, modeling, understanding, and response.

[0003] Existing systems mostly focus on single-modal analysis (such as video, GPS trajectory, or heart rate), lacking the ability to process multi-source heterogeneous data uniformly, and are also difficult to extend from individual behavior to modeling and decision support for overall tactical coordination. At the same time, existing models have poor domain adaptability, mainly targeting a few sports such as football and basketball, and are difficult to generalize to other competitive scenarios. Summary of the Invention

[0004] This invention provides a real-time tactical decision support system and method for sports competitions. Through unified alignment of multi-source heterogeneous data, behavioral graph modeling, tactical structure identification and strategy generation, it assists coaches or analysts in achieving cross-project, universal, and intelligent tactical analysis and real-time strategy recommendation.

[0005] To achieve the above-mentioned objectives, the present invention includes the following steps: The time-series data stream acquisition steps involve real-time acquisition and processing of match video data, player movement data, physiological data, and event data to generate a unified time-series data stream. The spatiotemporal behavior graph construction steps involve identifying player actions, positions, and interaction events based on the time-series data stream, dividing the match into continuous tactical phases, and constructing a spatiotemporal behavior graph within each tactical phase using players, actions, and events as nodes and interaction relationships as edges. The tactical feature extraction step involves inputting the spatiotemporal behavior graph into a graph neural network, aggregating node features through a graph attention mechanism, and outputting an embedding vector representing the current tactical form. The tactical recommendation production step involves calculating the similarity between the embedded vector and the historical tactical sample library to generate tactical recommendation results.

[0006] A second aspect of the present invention provides a real-time sports competition tactics and strategy recommendation system, comprising: The time-series data stream acquisition unit acquires and processes match video data, player movement data, physiological data, and event data in real time, generating a unified time-series data stream. The spatiotemporal behavior graph construction unit identifies player actions, positions, and interaction events based on the time-series data stream, and divides the game into continuous tactical phases; within each tactical phase, a spatiotemporal behavior graph is constructed using players, actions, and events as nodes and interaction relationships as edges. The tactical feature extraction unit inputs the spatiotemporal behavior graph into the graph neural network, aggregates node features through a graph attention mechanism, and outputs an embedding vector representing the current tactical form. The tactical recommendation production unit calculates the similarity between the embedded vector and the historical tactical sample library, and generates tactical recommendation results.

[0007] A third aspect of the present invention provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a circuit; the at least one processor invokes the instructions in the memory to cause the electronic device to execute the above-described method for real-time sports competition tactical identification and strategy recommendation.

[0008] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described method for real-time sports competition tactical identification and strategy recommendation.

[0009] The present invention has the following beneficial effects: This invention systematically integrates video, sensor, and game semantic information by introducing a unified multimodal data alignment mechanism and graph-structured behavior modeling method, thereby improving the accuracy of modeling player behavior and tactical patterns in complex dynamic game scenarios and providing a general and scalable path for sports intelligent decision analysis.

[0010] This invention constructs a spatiotemporal behavioral graph and combines it with a graph neural network for tactical collaboration analysis. This effectively identifies high-frequency collaboration patterns and key tactical changes, and generates real-time tactical suggestions through a historical strategy sample matching mechanism. This enhances the relevance and explanatory power of strategy recommendations, and significantly improves the speed and scientific nature of coaching teams' decision-making responses in actual combat. Attached Figure Description

[0011] Figure 1 A flowchart of a real-time sports competition tactics identification and strategy recommendation method provided in an embodiment of the present invention. Detailed Implementation

[0012] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0013] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The first embodiment of the real-time sports competition tactics identification and strategy recommendation method in this invention includes: Multimodal data acquisition and synchronous processing The system deploys 6-8 fixed high-definition cameras at the competition venue to acquire multi-view video; it also uses drones for aerial photography to fill blind spots and improve spatial observation depth. Each player wears an IMU module (100Hz) to collect acceleration, angular velocity, and attitude angle, a GPS device (10Hz) to provide coordinate data, and a heart rate sensor (5Hz) to collect physiological load data.

[0014] Algorithm alignment mechanisms (such as the TimeSync time synchronization module) are applied to various types of data to normalize all inputs into a unified time-series data stream: in, Represents visual data frames, Represents IMU motion data, Represents GPS motion data, This represents heart rate / fitness data. For the flow of game events (scoring, fouls, etc.) Visual recognition and segmentation analysis This step uses the real-time adaptive semantic segmentation network ARTHuS, combined with the Mask-RCNN teacher network, to generate semantic masks for specific game scenarios in real time, identify objects such as players, boundaries, balls, and referees, and output corresponding label sequences: Extract key player information and categorize their actions: A target tracking network is used to generate the target's trajectory path over time. In this embodiment, the ARTHuS semantic segmentation network is used to process the game images to achieve real-time pixel-level recognition of targets such as players, boundaries, balls, and referees in the game scene.

[0015] For the raw match video frames collected by the system First, a pre-trained teacher network (In this embodiment, Mask R-CNN is used) to generate pseudo-label masks as reference segmentation information, denoted as: This mask marks the pixel regions in the image that belong to the "human body" category, and uses a venue mask to remove irrelevant areas such as audience seats and billboards. This mask is then used to guide the training of the student network TinyNet, a lightweight student network.

[0016] The student network employs the following mechanism for training and iteration: The system initializes a pair of networks: an online segmentation network. With the corresponding training network And create an empty dataset. ; The system acquires images and outputs them within the first 5 minutes. Labeling, constructing the first training set ; Will For training One epoch, with the optimization objective being pixel-level cross-entropy loss; Will Parameter assignment To obtain a segmentation network that is updated in real time ; Repeat steps 3-4, and use a FIFO mechanism to replace the oldest training frame with the newest frame to construct... .

[0017] This process runs continuously, updating and iterating the student network every 200 frames, allowing the segmentation network to gradually adapt to changes in the colors, venue, and team uniforms of the current competition, achieving scene-specific "adaptation".

[0018] In this embodiment, for each frame The output of the processing is: Pixel mask This is used for the next step of attitude estimation; Attitude key point set The result is extracted from the mask by OpenPose. Action tags Input via BiLSTM Obtained through sequence classification; The final trajectory vector is: The trajectory of each player is used for subsequent graph construction, field control modeling, and tactical scoring analysis.

[0019] Spatiotemporal behavior modeling Divide the game into multiple tactical phases For each stage, the behavioral entities are structured into a spatiotemporal behavioral graph: in: : Point set (players, actions, events); Edge set (interactions such as passing, deployment, and interception); Estimate the area of ​​control each player has on the field: In this embodiment, the Manchester City Football Club Premier League match dataset is used. Each match contains an average of 1718 events. The system successfully segmented each match into approximately 95 tactical phases, with an average phase length of approximately 7.2 events. The original event sequence is divided into multiple continuous and clearly structured tactical phases according to the segmentation rules: If there is a time interval between the two events Second; Or, possession of the ball may change from one side to the other (such as during a change of sides, loss of possession, or dead ball). This is considered a tactical phase transition point. Therefore, the match is divided into multiple consecutive tactical phases: Each stage It is a continuous subsequence of events, representing an offensive attempt, a counterattack, or a defensive organization.

[0020] In each tactical phase Internally, construct the graph structure: in: Node set This includes player nodes (indicating their appearance in this phase), event nodes (such as "high tackle", "long shot attempt", etc.), and action nodes (such as "short pass" and "wing advance"). edge set : Represents the interactions between players (such as passing), the association between players and events, and the temporal logic between events.

[0021] Node features include, but are not limited to: Spatial location: Coordinates of an event or player; Participation metric: The number of edges connected to a node; Local temporal position: The relative temporal position of an event within a stage.

[0022] Edge features include, but are not limited to: Types: Collaboration (passing, covering), Pressing (pressing), Linkage (transition from midfield to forward); Duration: The time interval between two events; Spatial transfer vector: The vector of movement of the ball's position between events.

[0023] For example, if player A passes the ball from the left side of midfield to player B on the right wing of the forward line, then there exists an edge in the diagram: Once structured, this behavior can be used in the subsequent graph neural network inference stage.

[0024] To depict the players' tactical pressure and defensive range in space, a Voronoi partitioning model is used to construct the Dominant Region: For time Set of all player positions , define the first The control area for each player is: in This refers to the site's spatial area.

[0025] The system performs area integration on the controlled field region and calculates its value. As an indicator of a player's spatial control during this stage .

[0026] Graph Neural Network Collaborative Modeling The structured tactical behavior graph is input into a graph neural network to perform node information aggregation and high-order tactical recognition. The spatiotemporal behavior graph obtained in step S103 is represented as: in Indicates at time A collection of nodes consisting of players, tactical events, and action points. It indicates the interaction between them, such as passing, screening, pressing, etc.

[0027] For each node in the graph initial representation Multi-layer graph neural network modeling is performed. Graph Attention (GAT) is used for information aggregation. in, These are the weight coefficients between adjacent nodes calculated using the attention mechanism. The weight matrix is ​​a learnable matrix. For example, ReLU is an activation function.

[0028] Finally, the tactical form display vector is obtained by embedding the entire tactical map using the Readout operation: In this embodiment, for a Champions League quarter-final match, data from multiple attacking phases of both teams are selected as input, and each phase is constructed with a number of nodes. The number of sides is The graph structure is shown. In the experiment, a 3-layer GAT network with 4 attention heads was used, resulting in a final output embedding dimension of 256.

[0029] After training, the embedding vectors It can be used as input for tactical matching and strategy generation in subsequent stages.

[0030] Tactical rating and strategy recommendations For all player nodes appearing in the tactical phase diagram, let each player be denoted as Three scoring indicators are set according to actual needs: Field Control Index (SC), Execution Efficiency (EF), and Synergy Index (SY), and they are combined into a player behavior scoring vector: in: Estimated based on the cumulative area of ​​the player's controlled area during a specific time period, in units of... ; : Represents the ratio of the time a player actually takes to complete a key action to the model's expected value; : Represents the degree of collaboration of a player node in the graph, including its edge density with other players, the proportion of events it participates in, etc., with a value range of [0,1].

[0031] The specific calculation process for these three indicators is explained below: Site Control Index (SC) This metric is based on player trajectory data and Voronoi's control area. The system calculates this at every moment. The field control index is calculated by averaging or accumulating the area of ​​the field controlled by a player over a tactical phase. This index reflects the player's ability to cover space on the field during that phase.

[0032] in : Index of the player (or the tracked target). : No. Each tactical phase (time interval). :stage The set of discrete moments within a given time period. : Number of time steps within a stage. At any moment Based on the entire player position set, the data is divided using Voronoi (or dominant region) and assigned to the players. The controlled area. :area Geometric area (unit: m) 2 Among them, trajectory points Indicates player At any moment The site coordinates are output by the target tracking network; control area This indicates that the player is based on the Voronoi method. The spatial extent of the division is obtained by the following formula: Let the playing field be a planar region Ω, and there be N players. The position of each player i at time t is... That is, the area formed by all points that are closer to player i than to other players.

[0033] Execution efficiency (EF) This indicator is obtained from the time-series information and action classification output. The system first determines the actual time taken for a player to complete a key action (such as passing or shooting). And compared with the expected time set in advance or obtained from historical statistics. The two are compared. The ratio of the two is the execution efficiency: in, This indicates the expected completion time for this type of action, which is set a priori. Indicates player The actual time required to complete the action is obtained from the data alignment and action recognition in steps S1–S2. If the speed is faster than expected, it indicates that the player's action is completed faster than expected; conversely, it indicates that the execution efficiency is low.

[0034] Synergy Index (SY) This indicator is based on the topological characteristics of the spatiotemporal behavioral graph and is calculated directly using observable statistics of the graph structure.

[0035] in Players represented in the spatiotemporal behavior diagram The node. Indicates player node The degree (i.e. the number of edges connected to it, which reflects the frequency of interaction). This indicates the number of events the player participated in. This indicates the total number of events in this tactical phase. and For normalized weight coefficients, satisfying . and Equal weighting or expert calibration can be used.

[0036] Players The synergy index is a normalized scalar value in the range of [0,1]. The higher the value, the higher the player's participation and centrality in team tactical coordination.

[0037] Define the following scoring function for each player: in Spatial control index; Execution efficiency; Synergy Index For weights.

[0038] Substituting the behavioral score vector into the linear weighting function yields the player's overall score: in The values ​​are weighting coefficients, and in this embodiment, they are 0.4, 0.3, and 0.3.

[0039] Next, to match the optimal tactical strategy, the graph embedding representation of the current tactical stage is used. The input is fed into a pre-trained Transformer matching network, which constructs a historical tactical sample database: Each sample Each vector is a 256-dimensional vector with an labeled tactical strategy category. The matching strategy is performed using the following formula: in is the cosine similarity function used to calculate the similarity between the current embedding and historical samples.

[0040] Finally, a set of recommendation strategies is output based on the matching results: Each strategy contains structured content, including tactical name, suitable formation, positioning diagram, execution priority, and visualization generation parameters.

[0041] In this embodiment, during the ball possession phase in the 58th minute of the match, the system outputs the following recommendation: Formation suggestion: "Switch to a 4-3-3 formation and move the defense forward 7 meters"; Player suggestion: "Replace player number 10 with player number 7, who has a higher fitness rating." Collaborative strategy: "Speed ​​up the passing rhythm of the three players in the midfield and forward positions, and attempt to break through on the left side."

[0042] The strategy is displayed graphically in real time on the coach analysis interface, and its response time is recorded as 2.3 seconds. The coach adoption rate is 87.2% more accurate than that of manual strategies.

[0043] Step S106: Human-computer interaction and visualization Through heat map Path streamline diagram Collaborative graph Tactical situations can be displayed through various methods. Coaches can input requirements using interactive terminals via voice / touch. and provide feedback on new strategy suggestions. : In this embodiment, the following three types of interactive situational diagrams are generated using the identified tactical phases and player behavior maps: Heat map Statistics on a player or team during a certain period of time The frequency of space occupancy within the space is smoothed using a Gaussian kernel function: in Indicates the first One player The position coordinates at that moment.

[0044] Path streamline diagram Tracking players In time interval Trajectory path within: The direction and speed of its movement are indicated by arrows in the diagram.

[0045] Collaborative Relationship Diagram Using players as nodes and actions such as passing, assisting in defense, and blocking as edges, an interaction graph structure is constructed, and the node size (participation) and edge weight (cooperation intensity) are dynamically updated.

[0046] Each recommendation strategy The source is visualized through a graph structure: Show the tactical phases from the tactical sample library that are most similar to the current scenario. Including its graph structure and scoring vector and impact factors; The system displays the nodes and event paths that inform the recommendation strategy, along with supplementary text explanations.

[0047] For example: "The suggestion to move midfielder A to the high left area is based on the fact that his coordination index in that position has increased by 21% in similar scenarios in the past, and directly led to two effective attacks." Coaches can input tactical requirements via voice commands or touchscreen. ,For example: "The opponent is applying too much pressure; should we adjust our mid-lane push strategy?" The system performs semantic parsing on the input and then invokes the policy filter: in: Candidate selection strategy based on semantic keywords in the question (such as "oppression" and "middle"); : Policy embedding vector set, processed by Transformer on tactical phase graph Obtained by matching with historical strategies; : Back The most relevant strategy recommendation.

[0048] In the analysis of a professional football match live stream, the coaching team used a touch interface to access heatmaps and path maps for a specific period. They discovered that center forward B's activity area had deviated from the planned tactical path over the past 10 minutes. The system suggested replacing B with substitute C and implementing a cross-field strategy. After a voice inquiry asking "How to strengthen the organization in the backfield?", the system returned three strategic suggestions, providing a visual explanation combining the intensity of pressing in the attacking third, the trend of possession rate, and historical tactical evolution curves. The coach confirmed one of the suggestions and immediately implemented it via touch.

[0049] The above describes the method for real-time sports competition tactics identification and strategy recommendation in the embodiments of the present invention. The following describes the device for real-time sports competition tactics identification and strategy recommendation in the embodiments of the present invention: The time-series data stream acquisition unit acquires and processes match video data, player movement data, physiological data, and event data in real time, generating a unified time-series data stream. The spatiotemporal behavior graph construction unit identifies player actions, positions, and interaction events based on the time-series data stream, and divides the game into continuous tactical phases; within each tactical phase, a spatiotemporal behavior graph is constructed using players, actions, and events as nodes and interaction relationships as edges. The tactical feature extraction unit inputs the spatiotemporal behavior graph into the graph neural network, aggregates node features through a graph attention mechanism, and outputs an embedding vector representing the current tactical form. The tactical recommendation production unit calculates the similarity between the embedded vector and the historical tactical sample library, and generates tactical recommendation results.

[0050] Example 1 In this embodiment, the experimental conditions are set as follows: System: Ubuntu 20.04, Software: Python 3.9, Processor: Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz, Memory: 502GB, Graphics Processor: NVIDIA RTXA6000x10.

[0051] In this embodiment, a full video of a real match is selected, offensive segments are extracted, multimodal data streams are constructed, and steps S101 to S106 are completed sequentially in the system.

[0052] In this embodiment, the input video data resolution is 1920×1080, and the frame rate is 25fps. The action events are provided by the referee system, including the time and spatial coordinates of each pass, shot, and foul. A frame alignment mechanism is used to establish a unified timestamp sequence, generating 1372 frames of synchronization data packets.

[0053] The ARTHuS semantic segmentation network is used as a real-time human detector, and initial distillation is performed based on the teacher network Mask-RCNN. The system iterates and updates the student network TinyNet for 5 rounds in a 20-frame distillation cycle. Player keypoints are extracted using OpenPose, and action classification is performed using Bi-LSTM. In the system, a tactical phase is first identified as consisting of four consecutive actions: "advance-back pass-left cross-header". All player nodes and key event nodes are encoded into a behavior graph. The generated graph contains 23 nodes and 59 edges. The Voronoi method is used to estimate the control area. A three-layer GAT neural network is used to aggregate the node information of the graph, with node dimensions of 128, 128, and 256. Finally, a tactical embedding vector is output through a Readout operation. .

[0054] The following metrics were calculated for each of the 11 players on the field: Average controlled area (unit: m²) 2 ) The ratio of the time taken to complete a tactical maneuver to the expected duration Graph structure centrality index (edge ​​number normalized to one) Based on the current tactical embedding Matching against 112 tactical structures in the historical tactical sample set, and encoding using a Transformer structure, the Top-1 cosine similarity match result is "side-mid lane coordinated advance + mid lane diagonal run". The system suggests the following: Move player number 7 to the wing to create space for a drive through the middle.

[0055] The right-back pushes forward to provide support and apply pressure.

[0056] It is recommended to use a high-crossing strategy in the next round.

[0057] Using heatmaps Display player movement density and path map Indicates positioning trend, collaborative chart It indicates the passing / defensive relationship.

[0058] The coach used a touch interface to input voice commands during halftime: What were the reasons for the failure of the mid-lane penetration in the previous stage? The system returns 3 analysis explanations and automatically switches to the structure diagram + rating diagram + recommendation interface.

[0059] This invention also provides an electronic device, which can vary significantly due to differences in configuration or performance. It may include one or more processors (e.g., one or more processors) and memory, and one or more storage media (e.g., one or more mass storage devices) for storing applications or data. The memory and storage media may be temporary or persistent storage. The program stored in the storage media may include one or more modules, each module including a series of instruction operations on the electronic device. Furthermore, the processor may be configured to communicate with the storage media and execute the series of instruction operations stored in the storage media on the electronic device.

[0060] The electronic device may also include one or more power supplies, one or more wired or wireless network interfaces, one or more input / output interfaces, and / or one or more operating systems, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that the electronic device structure in this embodiment does not constitute a limitation on the electronic device itself, and may include more or fewer components, or combinations of certain components, or different component arrangements.

[0061] This invention provides an electronic device structure that can vary significantly depending on configuration or performance. It may include one or more central processing units (CPUs) (e.g., one or more processors) and memory, and one or more storage media (e.g., one or more mass storage devices) for storing applications or data. The memory and storage media can be temporary or persistent storage. The program stored in the storage media may include one or more modules, each module including a series of instruction operations on the electronic device. Furthermore, the processor may be configured to communicate with the storage media and execute the series of instruction operations stored in the storage media on the electronic device.

[0062] The electronic device may also include one or more power supplies, one or more wired or wireless network interfaces, one or more input / output interfaces, and / or one or more operating systems, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that the structure of the electronic device does not constitute a limitation on the electronic device itself, and may include more or fewer components than described above, or combine certain components, or have different component arrangements.

[0063] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the aforementioned method.

[0064] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0065] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0066] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A sports game real-time tactics identification and strategy recommendation method, characterized in that, The method comprises the following steps: a time series data stream acquisition step of acquiring and processing match video data, player movement data, physiological data and event data in real time to generate a unified time series data stream; a spatio-temporal behavior graph construction step of identifying player actions, positions and interaction events based on the time series data stream, and dividing the match into continuous tactical stages; in each tactical stage, a spatio-temporal behavior graph is constructed with players, actions and events as nodes and interaction relationships as edges; a tactical feature extraction step of inputting the spatio-temporal behavior graph into a graph neural network, aggregating node features through a graph attention mechanism, and outputting an embedding vector representing the current tactical form; a tactical recommendation production step of calculating the similarity of the embedding vector with a historical tactical sample library to generate a tactical recommendation result.

2. The method for real-time skill and tactic recognition and strategy recommendation in sports games according to claim 1, characterized in that, The spatio-temporal behavior graph construction step comprises: The semantic segmentation network SegmentNet is used to perform pixel-level recognition on the players, the ball, and the boundary of the court in the original game video frame v t , and output a corresponding label sequence Extracting a set of player key points and performing action classification to obtain an action label of the player at time t generating the trajectory path of a target in time using a target tracking network: The game is divided into a number of tactical phases Φ = {φ1, φ2,..., φ k}, and for each phase the behavior entities are structured into spatio-temporal behavior graphs: wherein: Point sets, including players, actions, events; Edge set, including pass, deploy, intercept.

3. The method of real-time skill and strategy recognition and strategy recommendation for sports games according to claim 1, characterized in that, The tactical feature extraction step comprises: based on the structure of the spatio-temporal behavior graph, applying a graph neural network to perform node information aggregation and representation update: wherein denotes the feature representation of node v at the l-th layer, is a set of neighboring nodes of node v, is an attention weight from node u to node v, indicating the importance of the neighboring relationship in the current tactical context, W (l) is a trainable weight matrix, and σ is a nonlinear activation function; the final output tactical co-attention graph embedding representation is used for subsequent prediction and identification; ultimately obtaining a display vector of the overall tactical form: z t = Readout({h v}).

4. The method of real-time skill and strategy recognition and strategy recommendation for sports games according to claim 1, characterized in that, The tactical recommendation production step comprises: Based on historical matches, a tactical sample library T = {τ1, …, τ n} is constructed, and a Transformer module is introduced for matching: Output the optimal strategy suggestion set R = {r1, r2, …, r m}.

5. The method for real-time skill and tactic recognition and strategy recommendation in sports games according to claim 1, characterized in that, The method further comprises a player tactical scoring step: defining a behavior score vector for each player: S i = [SC i , EF i , SY i ] wherein: SC i Indicates the field control cumulative area: where i is the index of the player, Φ k = [t s , t e ]: k-th tactical phase; T k = {t s , t s +1,..., t e}: set of discrete time instants within the phase Φ k . |T k |=t e -t s +1: number of time steps within a phase; a Voronoi or a dominant region partitioning based on the set of all player positions at time t, attributed to player i; Region Geometric area of the field where the trajectory point represents the field coordinate of player i at time t, output by the target tracking network; Control Region represents the spatial range divided for player i based on the Voronoi method, obtained by the following formula: Let the field be a planar region Ω, with N players, each player i having a position at time t i.e. the region of all points closer to player i than to any other player. EF i The action completion duration / expected duration, obtained through timing information and action classification output; first determine the actual time consumed by the player to complete a certain key action And compare it with the expected time consumption set in advance or obtained from historical statistics The ratio of the two is the execution efficiency: wherein, represents the desired completion duration of such action, set a priori; represents the duration in which the player i actually completes the action. SY i For edge density, tactical node participation rate; based on the topological features of the space-time behavior graph, directly use the observable statistics of the graph structure to calculate: where v i : the node representing player i in the spatio-temporal behavior graph; deg(v i ) denotes the degree of player node v i , EventCount(i) denotes the number of events that player i participated in, N events denotes the total number of events in the tactical phase; w1 and w2 are normalization weight coefficients satisfying w1+w2=1; w1 and w2 are assigned by equal weight or expert calibration; defining the following score function for each player: S i = λ1 · SC i + λ2 · EF i + λ3 · SY i where SC i : space control index; EF i : execution efficiency; SY i : synergy index, λ i is a weight.

6. The method for real-time skill and tactic recognition and strategy recommendation in sports games according to claim 1, characterized in that, The method further comprises a heat map display step: The heat map H(x, y, t) counts the spatial occupancy frequency of a player or a team at time t, and uses a Gaussian kernel function for smoothing: where χ i (t), y i (t) denotes the position coordinates of the i-th player at time t.

7. The method for real-time skill and strategy recognition and strategy recommendation in sports games according to claim 1, characterized in that, The method further comprises a path streamline diagram display step: Path flow line graph L i (t) trace the trajectory path of player i in time interval [t1, t2]: and the movement direction and speed thereof are represented by arrows in the diagram.

8. A sports game real-time tactics identification and strategy recommendation system, characterized in that, The system comprises: a time series data stream acquisition unit that acquires and processes match video data, player movement data, physiological data and event data in real time to generate a unified time series data stream; a spatio-temporal behavior graph construction unit that identifies player actions, positions and interaction events based on the time series data stream, and divides the match into continuous tactical stages; in each tactical stage, a spatio-temporal behavior graph is constructed with players, actions and events as nodes and interaction relationships as edges; a tactical feature extraction unit that inputs the spatio-temporal behavior graph into a graph neural network, aggregates node features through a graph attention mechanism, and outputs an embedding vector representing the current tactical form; a tactical recommendation production unit that calculates the similarity of the embedding vector with a historical tactical sample library to generate a tactical recommendation result. 9.An electronic device comprising a memory and at least one processor, the memory having instructions stored therein; the at least one processor invokes the instructions in the memory to cause the electronic device to perform the steps of the sports match real-time tactical identification and strategy recommendation method of any one of claims 1-7.

10. A computer-readable storage medium having stored thereon instructions, the instructions comprising, the instructions, when executed by the processor, implement the steps of the sports match real-time tactical identification and strategy recommendation method of any one of claims 1-7.

Images