Football defense decision analysis method, system, equipment and medium

Abstract

The invention relates to the technical field of decision analysis, and discloses a football defense decision analysis method, system and device and a medium, and the method comprises the steps: obtaining a plurality of groups of court data, and carrying out the fusion of the data, and obtaining the fusion data corresponding to the data of each court; performing feature engineering processing on the multiple groups of fused data to obtain graph feature sequences, and labeling each graph feature sequence by using a preset feature tag to obtain a corresponding tag; training a preset defense analysis model according to the graph feature sequence of each piece of fusion data and the corresponding label to obtain a defense decision analysis model; and obtaining current court data, and obtaining a defense decision analysis result by using the defense decision analysis model. The defending decision analysis model is obtained through model training, the defending decision analysis model is used for effectively evaluating the ability of the defending team, defending styles and scenes are quickly classified, suggestions are provided for defending running, a real-time defending strategy adjustment and tactical optimization scheme is provided, and the competitiveness and efficiency of the team are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of decision analysis, and in particular to a football defense decision analysis method, system, equipment and medium. Background Art

[0002] Football is a typical collective intelligence confrontation sport, in which defense is an important factor for victory. With the continuous evolution of football tactics and the increase in the strength of opponents, defensive strategies are becoming increasingly diverse and sophisticated. In recent years, high-pressure tactics have been favored, while some teams have also succeeded in using low-position defense tactics. The evaluation and decision-making of defensive tactics have gradually become a hot topic, and there is an urgent need to solve this problem through data analysis.

[0003] Due to data limitations, traditional methods for evaluating and making decisions about soccer defense strategies remain at the statistical analysis stage, especially statistics on the number of steals, interceptions, blocked shots, etc. of specific players, visualization of heat maps of coverage areas, and the number of goals lost in games. This method can only macroscopically evaluate the role of defensive players in the entire game or season, but cannot provide a fine-grained evaluation within a specific time segment, let alone evaluate whether the defensive players' decisions are correct, making it difficult to analyze the defensive situation from a global perspective. Summary of the invention

[0004] In view of this, the present invention provides a football defense decision analysis method, system, device and medium to solve the problem that it is impossible to evaluate the defensive situation of the court in a fine-grained manner to effectively perform defense decision analysis.

[0005] In a first aspect, the present invention provides a football defense decision analysis method, the method comprising:

[0006] Acquire multiple sets of court data, and fuse each set of court data separately to obtain fused data corresponding to each court data;

[0007] Perform feature engineering on multiple groups of fused data to obtain graph feature sequences of each fused data, and use preset feature labels to label each graph feature sequence to obtain labels corresponding to each graph feature sequence;

[0008] According to the graph feature sequences of each fused data and their corresponding labels, the preset defense analysis model is trained to obtain a defense decision analysis model;

[0009] The current court data is obtained, and the defensive decision analysis model is used according to the current court data to obtain the defensive decision analysis result.

[0010] The football defense decision analysis method provided by the present invention obtains a defense decision analysis model through model training, and uses the defense decision analysis model to effectively evaluate the ability of the defensive team, quickly classify defense styles and scenarios, provide suggestions for defensive running, and provide real-time defense strategy adjustment and tactical optimization solutions, thereby improving the team's competitiveness and efficiency.

[0011] In an optional implementation, the stadium data includes: player data, game information data, stadium event data, and tracking data. Each set of stadium data is fused separately to obtain fused data corresponding to each stadium data, including:

[0012] Obtain player data, game information data, stadium event data, and tracking data according to a preset data structure;

[0013] Uniformly number the player data in each stadium data;

[0014] Align the timestamps of the course event data and the tracking data, and determine the tracking time range based on the timestamps of the course event data;

[0015] Determine tracking data according to the tracking time range, the tracking data being the ball position and the player position in each frame of the tracking image within the tracking time range;

[0016] The event data and tracking data of each court are fused to obtain the fused data corresponding to the data of each court.

[0017] The football defense decision analysis method provided by the present invention proposes a set of standardized data structures by unifying different data sources, unifies the player numbers in tracking data and event data, aligns timestamps, and further obtains fused data of event data and tracking data, which is conducive to overall analysis of the data and makes the analysis results more accurate.

[0018] In an optional implementation, each graph feature sequence is labeled using a preset feature label to obtain a label corresponding to each graph feature sequence, including:

[0019] Get the time range of the target graph feature sequence;

[0020] According to the timestamps of the fused data and the ball-controlling teams corresponding to different timestamps, the time range of the target graph feature sequence is divided into multiple time periods;

[0021] Dividing the target graph feature sequence into multiple ball holding segments according to multiple time periods;

[0022] Each ball holding segment is annotated using a preset feature label to obtain a label for each ball holding segment as a label corresponding to the target graph feature sequence.

[0023] The football defense decision analysis method provided by the present invention divides the graph features into multiple ball holding segments according to different ball holding teams, and labels each ball holding segment to obtain labels of each graph feature sequence, which is conducive to analyzing the stadium situation according to each graph feature sequence and the corresponding label, and accurately grasping the stadium characteristics at each stage.

[0024] In an optional embodiment, the labels include: defensive player pressure intensity labels, defensive player interception intensity labels, and defensive scenario labels, and the preset defensive analysis models include: preset pressure prediction models, preset interception prediction models, preset scenario classification models, and preset running generation models.

[0025] The football defense decision analysis method provided by the present invention labels different ball holding segments according to important information that affects the situation on the field, can use the labels to analyze the field situation in a fine-grained manner, set multiple sub-models, obtain multiple defense-related results, and comprehensively analyze the multiple results to more comprehensively grasp the current situation on the field.

[0026] In an optional implementation, a preset defense analysis model is trained according to the graph feature sequences of each fusion data and their corresponding labels to obtain a defense decision analysis model, including:

[0027] The defensive pressure prediction model is trained by using the feature sequences of each graph and the corresponding defensive player pressure intensity labels to obtain a defensive pressure prediction model. The prediction results of the defensive pressure prediction model are used to reflect the pressure contribution of the players of the defensive team to the offensive team in the continuous ball holding segments.

[0028] The preset interception prediction model is trained by using the feature sequences of each graph and the corresponding interception strength labels of the defensive players to obtain a defensive interception prediction model. The prediction results of the defensive interception prediction model are used to reflect the interception contribution of the players of the defensive team to the offensive team in the continuous ball holding segments.

[0029] Merging the image feature sequences, the defensive pressure prediction results output by the defensive pressure prediction model, and the defensive interception prediction results output by the defensive interception prediction model to obtain a new image feature sequence, using the new image feature sequence and the defensive scene labels to train a preset scene classification model to obtain a defensive scene classification model, the classification results of the defensive scene classification model being used to reflect the probability that the ball holding segment belongs to a preset defensive scene category;

[0030] Input the new image feature sequence into the defense scene classification model to obtain the defense scene classification result;

[0031] Divide the new image feature sequence into multiple scene sample data according to scene classification labels;

[0032] The preset running generation model is trained using scene sample data to obtain a defensive running generation model, which is used to reflect the movement trajectory of defensive players in continuous ball-holding segments. The defensive pressure prediction model, defensive interception prediction model, defensive scenario classification model, and defensive running generation model together constitute a defensive decision analysis model.

[0033] The football defense decision analysis method provided by the present invention takes multiple groups of stadium data as samples, trains the defense decision analysis model according to preset rules, obtains multiple sub-models, and uses the multiple sub-models to analyze the stadium data from different angles. The analysis process is more comprehensive and it is convenient to analyze the defense situation from a global perspective.

[0034] In an optional implementation, the defensive decision analysis results include: defensive pressure prediction results, defensive interception prediction results, defensive scenario classification results, and defensive running generation results;

[0035] According to the current court data, the defensive decision analysis model is used to obtain the defensive decision analysis results, including:

[0036] The current court data is fused to obtain the current fused data, and the current fused data is subjected to feature engineering processing to obtain the current graph feature sequence of the current fused data;

[0037] Inputting the current graph feature sequence into the defensive pressure prediction model to obtain the current defensive pressure prediction result;

[0038] Inputting the current graph feature sequence into the defensive interception prediction model to obtain the current defensive interception prediction result;

[0039] The current image feature sequence, the current defensive pressure prediction result, and the current defensive interception prediction result are combined to obtain the current new image feature sequence, which is input into the defensive scene classification model to obtain the current defensive scene classification result;

[0040] According to the current defense scene classification result, the defense running generation model is used to obtain the current defense running generation result.

[0041] The football defense decision analysis method provided by the present invention uses a defense decision model to comprehensively analyze the current court data, obtain analysis results from multiple different angles, automatically evaluate the pressure intensity and interception situation of the defensive players, further judge the defense scene and defense style, and then generate defense trajectories under different defense styles, providing defense decision suggestions for the current scene.

[0042] In a second aspect, the present invention provides a football defense decision analysis system, the system comprising:

[0043] A data processing module is used to obtain multiple sets of court data and fuse each set of court data to obtain fused data corresponding to each court data;

[0044] A label processing module is used to perform feature engineering on multiple groups of fused data to obtain a graph feature sequence of each fused data, and to label each graph feature sequence using a preset feature label to obtain a label corresponding to each graph feature sequence;

[0045] A defense analysis module is used to train a preset defense analysis model according to the graph feature sequences of each fusion data and their corresponding labels to obtain a defense decision analysis model;

[0046] The defensive decision module is used to obtain the current court data and use the defensive decision analysis model according to the current court data to obtain the defensive result.

[0047] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0048] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to cause a computer to execute the method of the first aspect or any corresponding embodiment thereof.

[0049] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the method of the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 is a flowchart of a football defense decision analysis method according to an embodiment of the present invention;

[0052] Figure 2 is a schematic diagram of a flow chart of preprocessing source data in a football defense decision analysis method according to an embodiment of the present invention;

[0053] Figure 3 is a schematic diagram of a process of labeling fusion data in a football defense decision analysis method according to an embodiment of the present invention;

[0054] Figure 4is a flow chart of another football defense decision analysis method according to an embodiment of the present invention;

[0055] Figure 5 is a schematic diagram of a process of performing decision analysis based on graph features in a specific embodiment of a football defense decision analysis method according to an embodiment of the present invention;

[0056] Figure 6 is a structural block diagram of a football defense decision analysis system according to an embodiment of the present invention;

[0057] Figure 7 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0059] The embodiments of the present invention provide a football defense decision analysis method, system, device and medium, which analyze the situation on the field by establishing a defense decision analysis model to effectively evaluate the ability of the defending team, quickly classify the defense style and scenario, provide suggestions for defense running, and provide real-time defense strategy adjustment and tactical optimization scheme.

[0060] According to an embodiment of the present invention, an embodiment of a football defense decision analysis method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0061] In this embodiment, a football defense decision analysis method is provided, which can be used in the above-mentioned computer system. Figure 1 is a flow chart of a football defense decision analysis method according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0062] Step S101, obtaining multiple sets of court data, and fusing each set of court data to obtain fused data corresponding to each set of court data.

[0063] Specifically, in order to achieve fine-grained evaluation and analysis of the stadium, it is necessary to obtain multiple sets of stadium data for analysis. The multiple sets of stadium data are historical data, which can be recorded and collected by manual or automated equipment. This is only an example, but not limited to this. When obtaining stadium data, a set of standardized data structures are obtained, such as: player data, game information data, tracking data, and stadium event data. To facilitate data processing, the data structure and attributes of each set of data are the same, such as Figure 2 As shown in the figure, it is the process of preprocessing the court data. For example, the attributes of the player data include: the player number in the tracking data, the player number in the court event data, the player's name, age, height, weight, on-court position, and team; the attributes of the game information data include: game time, score, court size, team number in the tracking event, team number in the event data, player number lookup dictionary, team attack direction; the attributes of the tracking data include: the position information of the ball and the player in each frame; the attributes of the court event data include: event type, event occurrence time, event half, event result, event execution team, event executor, event occurrence location, event destination, and the ball holding segment sequence number of the event, which are only examples, but not limited to this.

[0064] Step S102 , performing feature engineering processing on multiple groups of fused data to obtain graph feature sequences of each fused data, and labeling each graph feature sequence with a preset feature label to obtain a label corresponding to each graph feature sequence.

[0065] Specifically, the graph features of each fused data are fully connected to represent the court information, including: node features, which are used to record the player information at any time of the game, including player position, player speed, player height, player weight, on-court position, etc.; edge features, which are used to record the relationship between a pair of players, including whether a pair of players are teammates or opponents, and whether a pair of players are in offensive and defensive positions; global features, which are used to record information related to court events and game-related information, including: number of rounds, remaining time of the game, ball-controlling player, ball-controlling team, etc., for example only, but not limited to this. It should be noted that the process of feature engineering and labeling multiple sets of fused data with preset labels is parallel, in no particular order.

[0066] like Figure 3As shown in the figure, it is the process of labeling the fused data. The original data features required for labeling include: the original features of the tracking data at any time in the game, the original features of any stadium event that occurred in the game, the fusion features of the tracking data and the stadium event data, and the graph features obtained by feature engineering the original features of the tracking data and the stadium event data. Specifically, the fusion of the tracking data and the stadium event data adopts the method of unified player labeling and aligned timestamps. The unified player labeling refers to the unified numbering based on the player's on-field jersey number. Aligned timestamps refer to the automatic query and marking of the time of the stadium event data. Within a certain range before and after the time of the stadium event record, the location of the stadium event and the location of the initiator of the stadium event are calculated to find the best matching stadium event timestamp. The processing of the original features adopts the structure of the graph to store information. The data structure of the graph refers to the fusion result based on the tracking data and the stadium event data. The stadium situation corresponding to any stadium event in the game can be obtained. The high-level features under the corresponding stadium situation are recorded in the form of graph features: the node records the detailed features of each player when the stadium event occurs, the edge reveals the relationship between the players, and the global feature records the relevant attributes of the stadium event.

[0067] The process of labeling the fused data requires domain experts to label the clips, which can be obtained from the commentary / commentary / post-match analysis video, or the event data provided by the data company may mark certain events as stressful events, for example only, but not limited to this.

[0068] Step S103: training a preset defense analysis model according to the graph feature sequences of each fusion data and their corresponding labels to obtain a defense decision analysis model.

[0069] Specifically, based on the focus and hot demands of defense in football games (for example: organized defense, blocking shooting opportunities, active pressure and pressing, positional defense, etc.), a preset defense analysis model was designed. The preset defense analysis model was trained using the actual historical stadium data to obtain a defensive decision analysis model suitable for any stadium situation.

[0070] Step S104, obtaining current court data, and using a defense decision analysis model according to the current court data to obtain a defense decision analysis result.

[0071] Specifically, when analyzing the current court situation, the current court data is obtained according to the preset court data structure, and the current court data is input into the defensive decision analysis model to obtain the defensive decision analysis results, and the defensive decision analysis results are used to guide the defensive strategy of the defensive players.

[0072] The football defensive decision analysis method provided in this embodiment obtains a defensive decision analysis model through model training, and uses the defensive decision analysis model to effectively evaluate the capabilities of the defensive team, quickly classify defensive styles and scenarios, provide suggestions for defensive running, and provide real-time defensive strategy adjustments and tactical optimization solutions, thereby improving the team's competitiveness and efficiency.

[0073] In this embodiment, a football defense decision analysis method is provided, which can be used in the above-mentioned computer system. Figure 4 is a flow chart of a football defense decision analysis method according to an embodiment of the present invention. Figure 4 As shown, the process includes the following steps:

[0074] Step S201, obtaining multiple sets of court data, and fusing each set of court data to obtain fused data corresponding to each set of court data.

[0075] Specifically, the court data includes: player data, game information data, court event data, and tracking data. The above step S201 includes:

[0076] Step S2011, obtaining player data, game information data, stadium event data, and tracking data according to a preset data structure.

[0077] Specifically, the situation on the field is complicated, and the data structures of different data sources are different. When analyzing data, the data structure should be consistent with the input of the defensive decision analysis model, but the data source cannot do this. Therefore, according to the different data characteristics, the data sources need to be unified into the same data structure to meet the requirements of defensive decision analysis, and the data is obtained according to the preset data structure between different data sources. The preset data structure includes four groups of data: player data, game information data, tracking data, and field event data. Specifically, the player data records the player's own information of a game (name, age, player number, etc.), the game information data records the basic information of the game (game time, score, field size, etc.), the tracking data records the position information of the ball and the player in each frame, and the field event data records the attributes of each important event (type, destination, time, executor, etc. of a certain event).

[0078] Step S2012: uniformly number the player data in each stadium data.

[0079] Specifically, in the same data source, tracking data and court event data are collected separately, so the labels of a player in the two sets of data are often different. For this problem, the same numbering can be performed based on the player's jersey number on the court. This is only an example and is not limited to this.

[0080] Step S2013, aligning the timestamps of the course event data and the tracking data, and determining the tracking time range according to the timestamps of the course event data.

[0081] Specifically, the player labels in the tracking data and the course event data are unified, and the timestamps of the tracking data and the course event data are aligned. The frame rate of the tracking data is 10 or 25, and the time unit of the course event data is second (that is, the frame rate is 1). Therefore, the tracking data and the course event data have different time scales (two events recorded very close to each other may correspond to the same tracking data timestamp). At the same time, the course event data is manually labeled and has labeling errors. Therefore, it is difficult for the course event data to reflect the time truly and finely. The tracking time range of the tracking data should be determined according to the timestamp of the course event data.

[0082] Step S2014, determining tracking data according to the tracking time range, where the tracking data is the ball position and the player position in each frame of the tracking picture within the tracking time range.

[0083] Specifically, within a preset time period before and after the event is recorded, the ball position and the event initiator's position are calculated when the event occurs, and the best matching timestamp is found. The time range of the tracking data is determined accordingly, and the ball position and the player position in each frame of the tracking image are recorded within the tracking time range. It should be noted that during the data collection process of the tracking data, the thermal noise of the electronic components, the electromagnetic noise of the communication system, the noise of the storage system, etc. make the tracking data inaccurate and fluctuate. The mean filling and linear interpolation methods can be used to smooth the data to achieve the effect of denoising.

[0084] Step S2015, fusing the event data of each course and the tracking data to obtain fused data corresponding to the data of each course.

[0085] Specifically, the stadium event data and the tracking data may be fused according to timestamps and divided into multiple groups of fused data according to timestamps, which is only an example but not limited to this.

[0086] The football defense decision analysis method provided in this embodiment proposes a set of standardized data structures by unifying different data sources, unifies the player numbers in tracking data and event data, aligns timestamps, and further obtains fusion data of event data and tracking data, which is conducive to overall analysis of the data and makes the analysis results more accurate.

[0087] Step S202 , performing feature engineering processing on the multiple groups of fused data to obtain graph feature sequences of each fused data, and labeling each graph feature sequence with a preset feature label to obtain a label corresponding to each graph feature sequence.

[0088] Specifically, the above step S202 includes:

[0089] Step S2021, obtaining the time range of the target graph feature sequence.

[0090] Specifically, for each graph feature sequence of fused data, each feature sequence is processed separately as a target feature sequence. First, the time range occupied by the target graph feature sequence is obtained, such as 12 o'clock to 12:03, which is only used as an example but not limited to this.

[0091] Step S2022, dividing the time range of the target graph feature sequence into multiple time periods according to the timestamps of the fused data and the ball-controlling teams corresponding to different timestamps.

[0092] Step S2023, dividing the target graph feature sequence into a plurality of ball holding segments according to a plurality of time periods.

[0093] Specifically, different players and teams may control the ball within the time range of the target graph feature sequence, so the time range of the target graph feature sequence is divided into multiple time periods according to the ball-controlling teams corresponding to different timestamps. The target graph feature sequence corresponding to each time period is regarded as a ball-controlling segment to record the ball-controlling team.

[0094] Step S2024: label each ball holding segment using a preset feature label to obtain a label for each ball holding segment as a label corresponding to the target graph feature sequence.

[0095] Specifically, based on the court conditions in each ball-holding segment, each ball-holding segment is labeled using the defensive player pressure intensity label, defensive player interception intensity label, and defensive scene label to obtain the labels corresponding to each graph feature sequence. The defensive player pressure intensity is determined by the position and speed of all players; the defensive player interception intensity is determined by the position and speed of all players; the defensive scene is divided into three dimensions: the aggressiveness of the defense, the stage of the defense facing the offense, and the position of the defense facing the offense holding the ball.

[0096] The football defense decision analysis method provided in this embodiment divides the graph features into multiple ball-holding segments according to different ball-holding teams, and labels each ball-holding segment to obtain a label for each graph feature sequence. This is conducive to analyzing the stadium situation according to each graph feature sequence and the corresponding label, and accurately grasping the stadium characteristics at each stage.

[0097] In some optional embodiments, the labels include: defensive player pressure intensity labels, defensive player interception intensity labels, and defensive scenario labels, and the preset defensive analysis models include: preset pressure prediction models, preset interception prediction models, preset scenario classification models, and preset running generation models.

[0098] The football defense decision analysis method provided in this embodiment labels different ball holding segments according to important information that affects the situation on the field, can use the labels to analyze the field situation in a fine-grained manner, set multiple sub-models, obtain multiple defense-related results, and comprehensively analyze multiple results to more comprehensively grasp the current situation on the field.

[0099] Step S203: training a preset defense analysis model according to the graph feature sequences of each fusion data and their corresponding labels to obtain a defense decision analysis model.

[0100] Specifically, for the graph feature sequence of each fusion data, the encoders of the preset pressure prediction model, the preset interception prediction model, the preset scene classification model, and the preset running generation model use a multi-layer multi-head graph attention neural network to encode it, and obtain the embedded feature sequence of the time-series graph features, and further extract the time-series information of the graph feature sequence to obtain a spatiotemporal coding, which helps each model to better understand and utilize the time and space structures in the data, thereby improving the performance and generalization ability of each model. The preset pressure prediction model, the preset interception prediction model, and the preset scene classification model are fully connected neural network structures, and the preset running generation model is a variational autoencoder recurrent neural network structure, which is only used as an example, but not limited to this. If the model adopts a neural network structure, the network parameters that need to be set are: the number of layers of the network hidden layer, the activation function of each neuron in the hidden layer, the loss function of the network, the initialization input, the sample size of batch training, and the random seed of network training. The example of the present invention uses a graph attention neural network + a recurrent neural network + a variational autoencoder neural network + a fully connected neural network as a model framework.

[0101] Specifically, the above step S203 includes:

[0102] Step S2031, using each graph feature sequence and its corresponding defensive player pressure intensity label, the preset pressure prediction model is trained to obtain a defensive pressure prediction model. The prediction result of the defensive pressure prediction model is used to reflect the pressure contribution of the players of the defensive team to the offensive team in the continuous ball holding segment.

[0103] Specifically, the model framework structure adopted by the preset pressure prediction model, the preset interception prediction model, and the preset scene classification model in this embodiment is consistent, including: a graph feature encoding module, a temporal feature extraction module, and a decoding module. First, it is necessary to encode the graph features after feature engineering: the graph neural network can well understand the features of the graph, and by transmitting node features, edge features, and global features, the initial features are mapped to a high-dimensional space, expressing the high-level semantic information of the graph, and obtaining the embedded features of the graph. The attention mechanism can characterize the node feature weights to adaptively pay attention to neighboring nodes, reduce network parameters through sparse connections, thereby reducing computational complexity and memory consumption, and thus using global information to guide the update and representation learning of node features. The multi-head attention mechanism has the following advantages: enhancing model representation capabilities, improving model robustness, stronger interpretability, and handling long-distance dependencies. Therefore, this embodiment uses a multi-layer multi-head graph attention neural network as a graph feature encoding module, inputs the graph features after feature engineering of a ball holding segment, and inputs the graph features corresponding to each moment in the ball holding segment into the graph attention neural network to obtain a graph embedded feature sequence.

[0104] Further, the time series information of the embedded feature sequence of the graph is feature extracted. The recurrent neural network has the characteristics of memory and shared parameters and is used to process time series information. The embedded feature sequence of the graph generated by the graph feature encoding module can be expressed as a time series matrix. Therefore, in this embodiment, a recurrent neural network, a gated neural unit, a long short-term memory network, etc. are used as a time series feature extraction module.

[0105] Based on the graph feature encoding module and the temporal feature extraction module, a spatiotemporal encoding can be obtained. The spatiotemporal encoding includes both the feature representation of nodes, edges, and the global in the graph, as well as the temporal relationship between the graph features processed by feature engineering at each moment in a ball-holding segment. Spatiotemporal encoding can help neural networks better understand and utilize the temporal and spatial structures in the data, thereby improving the performance and generalization ability of the model.

[0106] Finally, according to different task requirements, a decoding module is designed to implement a classification model or a regression model. The classification model is used to solve discrete output tasks, and the regression model is used to solve continuous output tasks. The three tasks to be solved in the example are predicting the defensive scene, predicting the pressure intensity of the defensive player, and predicting the interception intensity of the defensive player. Predicting the pressure intensity of the defensive player and predicting the interception intensity of the defensive player are continuous output tasks, which require a regression model; predicting the defensive scene is a discrete output task, which requires a classification model. This embodiment uses a fully connected network as a decoding module, and the network parameters are determined by the task scale and data volume.

[0107] Step S2032, using each graph feature sequence and its corresponding defensive player interception strength label, the preset interception prediction model is trained to obtain a defensive interception prediction model. The prediction result of the defensive interception prediction model is used to reflect the interception contribution of the players of the defensive team to the offensive team in the continuous ball holding segment.

[0108] Step S2033, merge the image feature sequences, the defensive pressure prediction results output by the defensive pressure prediction model, and the defensive interception prediction results output by the defensive interception prediction model to obtain a new image feature sequence, and use the new image feature sequence and the defensive scene labels to train the preset scene classification model to obtain the defensive scene classification model. The classification result of the defensive scene classification model is used to reflect the probability that the ball holding segment belongs to the preset defensive scene category.

[0109] Specifically, the training process of the defensive interception prediction model and the defensive scenario classification model is the same as the training process of the defensive pressure prediction model in step S2031, and will not be repeated here.

[0110] Step S2034, inputting the new image feature sequence into the defense scene classification model to obtain a defense scene classification result.

[0111] Step S2035: divide the new image feature sequence into a plurality of scene sample data according to the scene classification labels.

[0112] Specifically, the new image feature sequence is divided into a plurality of scene sample data representing different defense styles and defense scenes according to scene classification labels.

[0113] Step S2036, use the scene sample data to train the preset running generation model to obtain a defensive running generation model, which is used to reflect the movement trajectory of the defensive player in the continuous ball-holding segment. The defensive pressure prediction model, the defensive interception prediction model, the defensive scene classification model, and the defensive running generation model together constitute a defensive decision analysis model.

[0114] Specifically, the training input of the defensive running generation model is a sequence of classified new image features, whose categories are determined by multiple scene sample data divided by the defensive scene classification model. The latent variable of the defensive running generation model is a multi-dimensional Gaussian distribution. The output of the defensive running generation model is to generate a sequence of positions and speeds of defensive players in continuous ball-holding segments, that is, the running trajectory of defensive players. The generated results are used to assist analysts in tactical analysis and can also directly guide the running of defensive players.

[0115] In a specific embodiment, the architecture of the defensive running generation model consists of a graph feature encoding module, a temporal feature extraction module, and a generator module. The graph feature encoding module and the temporal feature extraction module of the defensive running generation model have the same structure as the graph feature encoding module and the temporal feature extraction module of the preset pressure prediction model, the preset interception prediction model, and the preset scene classification model, and the spatiotemporal encoding is also obtained to help better understand the temporal and spatial structure. The generator module maps the spatiotemporal encoding of each defensive player to multiple two-dimensional Gaussian distributions, and each player samples a new position and speed from them. The graph feature data sets divided into different scenes can generate different types of defensive habits or defensive styles, and the positions and speeds sampled and generated by the defensive running generation model can guide the defensive running positions corresponding to different types of defensive scenes.

[0116] The football defense decision analysis method provided in this embodiment takes multiple groups of stadium data as samples, trains the defense decision analysis model according to preset rules, obtains multiple sub-models, and uses the multiple sub-models to analyze the stadium data from different angles. The analysis process is more comprehensive and facilitates analyzing the defense situation from a global perspective.

[0117] Step S204, obtaining current court data, and using a defense decision analysis model according to the current court data to obtain a defense decision analysis result.

[0118] Specifically, Figure 5 As shown, it is a process of obtaining a defensive decision analysis result by using a defensive decision analysis model according to current court data. The defensive decision analysis result includes: a defensive pressure prediction result, a defensive interception prediction result, a defensive scenario classification result, and a defensive running generation result. The above step S204 includes:

[0119] Step S2041, fuse the current court data to obtain current fused data, and perform feature engineering on the current fused data to obtain a current graph feature sequence of the current fused data. For details, please refer to the description in step S2031, which will not be repeated here.

[0120] Step S2042, inputting the current graph feature sequence into the defensive pressure prediction model to obtain the current defensive pressure prediction result.

[0121] Specifically, the defensive pressure prediction model is used to determine the defensive pressure prediction results under the current court status, which can effectively evaluate the defensive team's pressure ability.

[0122] Step S2043, inputting the current graph feature sequence into the defensive interception prediction model to obtain the current defensive interception prediction result.

[0123] Specifically, the defensive interception prediction model is used to determine the defensive interception prediction results under the current stadium status, which can effectively evaluate the interception ability of the defensive team.

[0124] Step S2044, merging the current graph feature sequence, the current defensive pressure prediction result, and the current defensive interception prediction result to obtain a current new graph feature sequence, which is input into a defensive scene classification model to obtain a current defensive scene classification result.

[0125] Specifically, the defensive scene classification model is used to determine the defensive scene classification result under the current court state, so that the defensive style and defensive scene can be quickly classified.

[0126] Step S2045, using the defensive movement generation model according to the current defensive scene classification result, obtains the current defensive movement generation result.

[0127] Specifically, the defensive run generation model is used to determine the defensive run generation results under the current court status, which can provide decision-making suggestions for defensive runs, assign responsibilities for mistakes and lost balls, explore new defensive styles and modes, plan team training plans, and provide the team with real-time defensive strategy adjustments and tactical optimization suggestions, thereby improving the team's defensive efficiency and competitiveness in the game.

[0128] It should be noted that according to the actual stadium analysis and decision-making needs, you can choose between fully automatic reasoning analysis and semi-automatic reasoning analysis. In the case of fully automatic reasoning analysis, the defensive pressure prediction model, defensive interception prediction model, and defensive scenario classification model are used to automatically obtain the defensive pressure prediction results, defensive interception prediction results, and defensive scenario classification results. The graph features processed by feature engineering are directly used by the model to obtain the corresponding scenes of the fragments and generate defensive running positions; in the case of semi-automatic reasoning analysis, the coach or analyst of the defensive team can customize the defensive pressure prediction results, defensive interception prediction results, and defensive scenario classification results. The analyst of the defensive team can modify the defensive scenario according to needs to obtain the running trajectories under different defensive styles.

[0129] The football defensive decision analysis method provided in this embodiment uses a defensive decision model to comprehensively analyze the current stadium data, obtain analysis results from multiple different angles, automatically evaluate the pressure intensity and interception situation of the defensive players, further judge the defensive scenario and defensive style, and then generate defensive trajectories under different defensive styles, providing defensive decision suggestions for the current scenario.

[0130] In this embodiment, a football defense decision analysis system is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the system described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0131] This embodiment provides a football defense decision analysis system. Figure 6 As shown, including:

[0132] The data processing module 601 is used to obtain multiple sets of court data and fuse each set of court data to obtain fused data corresponding to each set of court data.

[0133] The label processing module 602 is used to perform feature engineering processing on multiple groups of fused data to obtain graph feature sequences of each fused data, and to label each graph feature sequence with a preset feature label to obtain a label corresponding to each graph feature sequence.

[0134] The defense analysis module 603 is used to train a preset defense analysis model according to the graph feature sequences of each fusion data and their corresponding labels to obtain a defense decision analysis model.

[0135] The defense decision module 604 is used to obtain current court data and obtain a defense result using a defense decision analysis model based on the current court data.

[0136] In some optional implementations, the data processing module 601 includes:

[0137] The structural unified unit is used to obtain player data, game information data, stadium event data, and tracking data according to a preset data structure.

[0138] The numbering unification unit is used to unify the player data in each stadium data.

[0139] The timestamp alignment unit is used to align the timestamps of the course event data and the tracking data, and to determine the tracking time range based on the timestamps of the course event data.

[0140] The data tracking unit is used to determine the tracking data according to the tracking time range, and the tracking data is the ball position and the player position in each frame of the tracking picture within the tracking time range.

[0141] The data fusion unit is used to fuse the event data of each court and the tracking data to obtain fused data corresponding to the data of each court.

[0142] In some optional implementations, the tag processing module 602 includes:

[0143] The time determination unit is used to obtain the time range of the target graph feature sequence.

[0144] The graph feature sequence division unit is used to divide the time range of the target graph feature sequence into multiple time periods according to the timestamps of the fused data and the ball-controlling teams corresponding to different timestamps.

[0145] The ball holding segment determination unit is used to divide the target image feature sequence into multiple ball holding segments according to multiple time periods.

[0146] The label labeling unit is used to label each ball holding segment using a preset feature label to obtain a label of each ball holding segment as a label corresponding to the target graph feature sequence.

[0147] In some optional implementations, the defense analysis module 603 includes:

[0148] The pressure model training unit is used to train the preset pressure prediction model using the feature sequences of each graph and the corresponding defensive player pressure intensity labels to obtain a defensive pressure prediction model. The prediction results of the defensive pressure prediction model are used to reflect the pressure contribution of the players of the defensive team to the offensive team in continuous ball-holding segments.

[0149] The interception model training unit is used to train the preset interception prediction model using the feature sequences of each graph and the corresponding defensive player interception strength labels to obtain a defensive interception prediction model. The prediction results of the defensive interception prediction model are used to reflect the interception contribution of the players of the defensive team to the offensive team in continuous ball-holding segments.

[0150] The scene classification model training unit is used to merge the image feature sequences, the defensive pressure prediction results output by the defensive pressure prediction model, and the defensive interception prediction results output by the defensive interception prediction model to obtain a new image feature sequence, and use the new image feature sequence and the defensive scene label to train the preset scene classification model to obtain a defensive scene classification model. The classification result of the defensive scene classification model is used to reflect the probability that the ball holding segment belongs to the preset defensive scene category.

[0151] The scene classification result determination unit is used to input the new image feature sequence into the defense scene classification model to obtain the defense scene classification result.

[0152] The sample data division unit is used to divide the new image feature sequence into multiple scene sample data according to the scene classification label.

[0153] The decision model training unit is used to train the preset running generation model using scene sample data to obtain a defensive running generation model, which is used to reflect the movement trajectory of the defensive player in the continuous ball-holding segment. The defensive pressure prediction model, the defensive interception prediction model, the defensive scene classification model, and the defensive running generation model together constitute the defensive decision analysis model.

[0154] In some optional implementations, the defense decision module 604 includes:

[0155] The current court data fusion unit is used to fuse the current court data to obtain current fused data, and perform feature engineering processing on the current fused data to obtain a current graph feature sequence of the current fused data.

[0156] The current pressure result prediction unit is used to input the current graph feature sequence into the defensive pressure prediction model to obtain the current defensive pressure prediction result.

[0157] The current interception result prediction unit is used to input the current graph feature sequence into the defensive interception prediction model to obtain the current defensive interception prediction result.

[0158] The current scene determination unit is used to merge the current image feature sequence, the current defensive pressure prediction result, and the current defensive interception prediction result to obtain the current new image feature sequence, and input it into the defensive scene classification model to obtain the current defensive scene classification result.

[0159] The current running generation unit is used to obtain the current defensive running generation result by using the defensive running generation model according to the current defensive scene classification result.

[0160] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0161] The football defense decision analysis system in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0162] The embodiment of the present invention also provides a computer device having the above Figure 6 The football defensive decision analysis system shown.

[0163] See also Figure 7 , Figure 7 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 7As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 7 A processor 10 is taken as an example.

[0164] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0165] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0166] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0167] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0168] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0169] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0170] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0171] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A football defense decision analysis method, characterized in that: The method comprises: Acquire multiple sets of court data, and fuse each set of court data separately to obtain fused data corresponding to each court data; Perform feature engineering on multiple groups of fused data to obtain graph feature sequences of each fused data, and use preset feature labels to label each graph feature sequence to obtain labels corresponding to each graph feature sequence; According to the graph feature sequences of each fusion data and their corresponding labels, a preset defense analysis model is trained to obtain a defense decision analysis model; Current court data is obtained, and the defense decision analysis model is used according to the current court data to obtain a defense decision analysis result.

2. The method according to claim 1, characterized in that The court data includes: player data, game information data, court event data, and tracking data. Each set of court data is fused separately to obtain fused data corresponding to each court data, including: Obtain player data, game information data, stadium event data, and tracking data according to a preset data structure; Uniformly number the player data in each stadium data; Align the timestamps of the course event data and the tracking data, and determine the tracking time range based on the timestamps of the course event data; Determine tracking data according to the tracking time range, wherein the tracking data is the ball position and the player position in each frame of the tracking picture within the tracking time range; The event data and tracking data of each court are fused to obtain the fused data corresponding to the data of each court.

3. The method according to claim 2, characterized in that The method of labeling each graph feature sequence by using a preset feature label to obtain a label corresponding to each graph feature sequence includes: Get the time range of the target graph feature sequence; Dividing the time range of the target graph feature sequence into multiple time periods according to the timestamps of the fused data and the ball-controlling teams corresponding to different timestamps; Dividing the target graph feature sequence into a plurality of ball holding segments according to the plurality of time periods; Each ball holding segment is annotated using a preset feature label to obtain a label for each ball holding segment as a label corresponding to the target graph feature sequence.

4. The method according to claim 1, characterized in that The labels include: defensive player pressure intensity labels, defensive player interception intensity labels, and defensive scenario labels. The preset defensive analysis models include: preset pressure prediction models, preset interception prediction models, preset scenario classification models, and preset running generation models.

5. The method according to claim 4, characterized in that According to the graph feature sequences of the fused data and their corresponding labels, a preset defense analysis model is trained to obtain a defense decision analysis model, including: Using the feature sequences of each graph and the corresponding defensive player pressure intensity labels, a pre-set pressure prediction model is trained to obtain a defensive pressure prediction model, wherein the prediction results of the defensive pressure prediction model are used to reflect the pressure contribution of the players of the defensive team to the offensive team in the continuous ball holding segments; Using the feature sequences of each graph and the corresponding defensive player interception strength labels, a preset interception prediction model is trained to obtain a defensive interception prediction model, wherein the prediction results of the defensive interception prediction model are used to reflect the interception contribution of the players of the defensive team to the offensive team in the continuous ball holding segments; The image feature sequences, the defensive pressure prediction results output by the defensive pressure prediction model, and the defensive interception prediction results output by the defensive interception prediction model are combined to obtain a new image feature sequence, and a preset scene classification model is trained using the new image feature sequence and the defensive scene label to obtain a defensive scene classification model, wherein the classification result of the defensive scene classification model is used to reflect the probability that the ball holding segment belongs to a preset defensive scene category; Inputting the new image feature sequence into the defense scene classification model to obtain a defense scene classification result; Divide the new image feature sequence into multiple scene sample data according to scene classification labels; The preset running generation model is trained using the scene sample data to obtain a defensive running generation model, which is used to reflect the movement trajectory of the defensive player in the continuous ball-holding segment. The defensive pressure prediction model, the defensive interception prediction model, the defensive scene classification model, and the defensive running generation model together constitute a defensive decision analysis model.

6. The method according to claim 4, characterized in that The defensive decision analysis results include: defensive pressure prediction results, defensive interception prediction results, defensive scenario classification results, and defensive running generation results; The method of obtaining a defense decision analysis result by using the defense decision analysis model according to the current court data includes: The current court data is fused to obtain current fused data, and the current fused data is subjected to feature engineering processing to obtain a current graph feature sequence of the current fused data; Inputting the current graph feature sequence into the defensive pressure prediction model to obtain a current defensive pressure prediction result; Inputting the current graph feature sequence into the defensive interception prediction model to obtain a current defensive interception prediction result; The current graph feature sequence, the current defensive pressure prediction result, and the current defensive interception prediction result are combined to obtain a current new graph feature sequence, and the new sequence is input into the defensive scene classification model to obtain a current defensive scene classification result; A current defensive running generation result is obtained by utilizing a defensive running generation model according to the current defensive scenario classification result.

7. A football defense decision analysis system, characterized in that: The system comprises: A data processing module is used to obtain multiple sets of court data and fuse each set of court data to obtain fused data corresponding to each court data; A label processing module is used to perform feature engineering on multiple groups of fused data to obtain a graph feature sequence of each fused data, and to label each graph feature sequence using a preset feature label to obtain a label corresponding to each graph feature sequence; A defense analysis module, used to train a preset defense analysis model according to the graph feature sequences of each fusion data and their corresponding labels to obtain a defense decision analysis model; The defense decision module is used to obtain current court data and obtain a defense result by using the defense decision analysis model according to the current court data.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the football defense decision analysis method according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the football defense decision analysis method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer instructions, wherein the computer instructions are used to cause a computer to execute the football defense decision analysis method according to any one of claims 1 to 6.

Images