A method and device for dividing a football game tactical scene

Abstract

The present application relates to the technical field of artificial intelligence, and discloses a method and device for dividing a football match tactical scene, wherein various types of match data are frame-aligned to obtain target fusion data; the match video and the target fusion data are time-aligned to obtain a target match video; and the target fusion data is input into a preset automatic division model of a football match typical scene to divide the target match video into stages and obtain a tactical scene division result. By ensuring high fusion of data, the target fusion data is input into the preset automatic division model of the football match typical scene to divide the match into stages, different tactical scenes are analyzed in detail, the defect that only single analysis is possible in the related art is overcome, the efficiency of dividing scenes is improved, and effective assistance for a more fine-grained tactical scene is provided for a football professional analyst.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and particularly relates to a method and device for dividing a football match tactical scene. BACKGROUND

[0002] Football is the world's first sport, and the analysis of football matches is of great significance for the review and evaluation of players and teams, as well as the cultivation of players and the tactical layout of teams. At present, simple statistics such as total score, pass and shot times cannot comprehensively and exhaustively reflect the technical and tactical performance of players and teams, so targeted analysis of different typical match segments or scenes plays an important role in player technical and tactical evaluation and team tactical evaluation.

[0003] In related technologies, targeted analysis of different typical match segments or scenes mainly has the following three methods:

[0004] The first method requires manual editing of different scenes and different stages by using video editing software. This method often requires analysts to spend a lot of time and effort, and cannot efficiently analyze different matches, and the analysis results are not accurate.

[0005] The second method is a semi-automatic method based on rule-based expert knowledge, which relies on predefined rules and expert knowledge. The segments divided by this method are rough, and only simple scenes such as the first half and the second half, tactical pauses, etc. can be divided, and complex tactical scenes and segments cannot be identified.

[0006] The third method is a fully data-driven automatic method that relies on data for screening, but the screened segments have deviations and may not meet the judgment standards of coaches and analysts. This method can only automatically identify a certain scene, and multiple different models are required to divide multiple tactical scenes in a match. The disadvantages of this method are obvious, and the division efficiency is low, and multiple types of scenes cannot be efficiently divided at the same time. SUMMARY

[0007] Therefore, the present application provides a method and device for dividing a football match tactical scene to solve the technical problem of how to accurately and efficiently divide a football match tactical scene.

[0008] In a first aspect, the present application provides a method for dividing a football match tactical scene, the method comprising:

[0009] obtaining a match video of the entire football match and multiple types of match data;

[0010] aligning the multiple types of match data by frame number to obtain target fusion data;

[0011] align the match video with the target fusion data in time to obtain a match video after time alignment as a target match video;

[0012] input the target fusion data into a preset automatic division model of typical scenes of a football match to obtain different scene data segments and corresponding starting frame numbers; wherein the scene data segments represent different granularities and staged football match scene segments;

[0013] perform stage division on the target match video according to the different scene data segments and corresponding starting frame numbers to obtain a tactical scene division result.

[0014] The embodiment first ensures high fusion of data by aligning multiple types of match data in frame numbers and aligning the match video in time, so that subsequent analysis is more accurate. Compared with a traditional overall analysis method, the present application can perform stage division on a match and perform detailed analysis on different tactical scenes, overcoming the defect that the prior art cannot perform detailed analysis and can only perform single analysis. The target fusion data is input into a preset automatic division model of typical scenes of a football match to obtain different scene data segments and corresponding starting frame numbers, and the target match video is divided in stages, avoiding subjective errors in manual marking and low efficiency, and quickly and accurately obtaining multiple tactical situations in a match scene, improving the efficiency of dividing scenes.

[0015] In an optional embodiment, the multiple types of match data include player data, event data, and tracking data, and the aligning multiple types of match data in frame numbers to obtain target fusion data specifically includes:

[0016] sequentially match the player data with the tracking data and event data, confirm target tracking data corresponding to the player data and multiple target event data; wherein the tracking data represents ball positions and player positions in each frame of tracking pictures within a tracking time range; and the event data represents key events occurring in each frame of event pictures within a whole match time range;

[0017] obtain time frames of the target event data and frequencies of the target tracking data; wherein the frequencies of the target tracking data represent time intervals of tracking data acquisition;

[0018] group and mark the tracking data according to the frequencies of the tracking data to obtain a group set of target tracking data group and mark, each group in the group set containing the same preset number of time frames, and each time frame of each group having a corresponding tracking data segment;

[0019] According to the time frame of the target event data and the group set marked by the target tracking data, all the target event data is inserted into the target tracking data to obtain target fusion data.

[0020] The embodiment fuses the position information tracking data of the player and the event data to form complete fusion data. By grouping and marking the frequency of the tracking data, the time sequence consistency of the data is ensured, so that each event data can be accurately matched to the corresponding tracking data time frame. After data fusion, the model can not only obtain the position and action of the player, but also combine specific game events such as passing and shooting. Compared with manual annotation or only single analysis of a certain tactical scene in related technologies, the embodiment obtains target fusion data in advance to provide accurate data basis for subsequent model identification of tactical scenes, so that the model can more comprehensively understand the tactical behavior of the player and improve the accuracy of the analysis result.

[0021] In an optional implementation, the inserting all the target event data into the target tracking data according to the time frame of the target event data and the group set marked by the target tracking data to obtain target fusion data specifically includes:

[0022] According to the time of the target event data, the group set marked by the target tracking data is sequentially searched;

[0023] When the time of the target event data is found to exist in any target group in the group set marked by the target tracking data, the tracking data segment corresponding to the position of any time frame in the target group is deleted, and the target event data is supplemented and inserted into the position of the last time frame in the target group; until all the target event data is inserted into the target tracking data to obtain target fusion data.

[0024] The embodiment sequentially searches the time of the target event data, ensures that the event data is inserted into the correct tracking data group, deletes the position data of the corresponding time frame in the target tracking data group, and inserts the event data into the last time frame, so that the target fusion data of each group has the same dimension, the tracking data and the event data on the time frame are accurately aligned in time, and the consistency is maintained, thereby improving the accuracy of data fusion.

[0025] In an optional implementation, the game data further includes game time, and the inputting the target fusion data into a preset typical scene automatic division model of football game to obtain different scene data segments and corresponding start frame numbers specifically includes:

[0026] The match time is divided into an upper half match time and a lower half match time according to frame numbers;

[0027] Extracting the ball possession state information corresponding to the target event data in the target fusion data;

[0028] According to the ball possession state information, the ball possession stream segment in the target fusion data is divided;

[0029] Based on a preset tactical scene type, the upper half match time and the lower half match time, the ball possession stream segment is defined with a corresponding scene label;

[0030] Feature engineering is performed on the ball possession stream segment with the scene label, and tracking data features and event data features are extracted;

[0031] According to the tracking data features and the event data features, high-level features are constructed for the target fusion data, wherein the high-level features are used to describe the spatial positions of players, balls and time on the playing field and tactical layouts;

[0032] The target fusion data after the high-level features are constructed is input into a preset automatic division model of typical football match scenes, and different scene data segments and corresponding start frame numbers are obtained.

[0033] In this embodiment, the match time is divided into an upper half and a lower half according to frame numbers, and the target fusion data is divided into ball possession stream segments in combination with the ball possession state information, so as to facilitate subsequent analysis and extraction of high-level features. Based on a preset tactical scene type, the ball possession stream segment is defined with a corresponding scene label, so as to ensure more accurate analysis. By extracting tracking data features and event data features and constructing high-level features, diversified feature inputs are ensured, so as to help the model better understand complex pattern relationships in the fusion data, improve recognition accuracy, and further describe the overall tactical layout on the playing field, not limited to analysis of a single player or event in related technologies.

[0034] In an optional implementation, the target match video is divided into stages according to the different scene data segments and the corresponding start frame numbers, and a tactical scene division result is obtained, specifically including:

[0035] The start frame numbers corresponding to the different scene data segments are converted into match tactical start times according to a time conversion format;

[0036] The target match video is divided into stages based on the match tactical start times, and a tactical scene division result is obtained, wherein the tactical scene division result includes a plurality of marked association information, and the marked association information includes an association relationship between a video segment and a corresponding scene data segment.

[0037] The embodiment can accurately locate a specific tactical scene in a match video by converting the frame number into time. Each tactical scene includes marked associated information, which helps football coaches and data analysts more easily understand and evaluate the tactical execution of the team.

[0038] In an optional embodiment, the automatic division model of the typical football match scene is a model obtained by pre-training, wherein the pre-training process includes: determining a basic training model according to the classification requirement of the football match tactical scene; obtaining a football match scene data set, and inputting the preprocessed data set into the basic training model for iterative training to obtain the automatic division model of the typical football match scene.

[0039] The embodiment enhances the recognition ability of the pre-trained model for different scenes by training with a large amount of football match scene data set. The trained automatic division model can complete the recognition of complex football tactical scenes with less manual operation, reduces the error of manual annotation, and facilitates continuous updating and iterative learning, and sustainable model updating.

[0040] In a second aspect, the present application provides a device for dividing a football match tactical scene, which comprises:

[0041] A data collection module is configured to obtain a match video of a whole football match and various types of match data.

[0042] A data alignment module is configured to align the frame numbers of the various types of match data to obtain target fusion data.

[0043] A time synchronization module is configured to time-align the match video with the target fusion data to obtain a time-aligned match video as a target match video.

[0044] A model output module is configured to input the target fusion data into a preset automatic division model of a typical football match scene to obtain different scene data segments and corresponding starting frame numbers. The scene data segments represent different granularities and phased football match scene segments.

[0045] A tactical scene division module is configured to divide the target match video into phases according to the different scene data segments and corresponding starting frame numbers to obtain a tactical scene division result.

[0046] In a third aspect, the present application provides a computer device, comprising: a memory and a processor, which are connected with each other in communication, the memory stores computer instructions, and the processor executes the computer instructions to perform the method for dividing the football match tactical scene according to the first aspect or any one of the corresponding embodiments thereof.

[0047] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer instructions for making a computer execute the method for dividing the football match tactical scene according to the first aspect or any one of the corresponding embodiments thereof.

[0048] In a fifth aspect, the present application provides a computer program product, which comprises computer instructions for making a computer execute the method for dividing the football match tactical scene according to the first aspect or any one of the corresponding embodiments thereof. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0050] Figure 1 is a flowchart of the method for dividing the football match tactical scene according to an embodiment of the present application;

[0051] Figure 2 is a flowchart of another method for dividing the football match tactical scene according to an embodiment of the present application;

[0052] Figure 3 is a flowchart of another method for dividing the football match tactical scene according to an embodiment of the present application;

[0053] Figure 4 is a search scene diagram of the method for dividing the football match tactical scene according to an embodiment of the present application;

[0054] Figure 5 is a structural block diagram of the device for dividing the football match tactical scene according to an embodiment of the present application;

[0055] Figure 6 is a hardware structure diagram of the computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the protection scope of the present application.

[0057] The traditional football match analysis method usually statistically analyzes the whole match as a whole, for example, the number of passes, the number of successful passes, the number of shots, the number of goals and the number of fouls. Although this method can understand the match situation of the team as a whole, it is difficult to evaluate the tactical efficiency of the team in a specific scene and the performance of the players.

[0058] In contrast, the football match is divided into stages, which can more specifically analyze the performance of different tactical scenes in the match. There are mainly two kinds of division methods at present: one completely relies on manual marking and cannot realize automatic and efficient division of the match; the other can only automatically identify and divide a certain tactical scene in a match, and needs to use multiple models or systems to identify and divide according to the data provided by different data providers. Both of these two ways are relatively inefficient.

[0059] The method for dividing football match tactical scenes provided by the embodiments of the present application is applied to the case of needing to deeply analyze different tactics in football matches. The frame numbers of multiple types of match data are aligned and input into a preset automatic division model, and different scene data segments and their starting frame numbers are obtained through the model to divide the match video. The automatic scene division of the present application avoids the errors and low efficiency division that may occur in the manual marking in the traditional technology, and can identify and divide different tactical scenes, not limited to a single tactical type, not only accurate, but also efficient in analyzing multiple tactical scenes, providing more valuable reference information for football coaches and data analysts.

[0060] According to the embodiments of the present application, a method for dividing football match tactical scenes is provided. It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described here can be executed in a different order.

[0061] In the present embodiment, a method for dividing football match tactical scenes is provided, which can be used in the above-mentioned computer, Figure 1 The flowchart of the method for dividing football match tactical scenes according to the embodiments of the present application is shown in Figure 1 The flowchart includes the following steps:

[0062] Step S101, obtaining a football match video and various types of match data.

[0063] It should be noted that the match video refers to the complete video recording of the entire football match, which can be a broadcast video or a tactical video of any perspective. For example, the full broadcast video recording of the XX Cup final; various types of match data refer to various data related to the match, such as player identification information, tracking information, event information, and ball possession time information.

[0064] It can be understood that obtaining the match video and various data is the basis of the entire match tactical scene division process, which is used for subsequent accurate analysis.

[0065] Step S102, frame number alignment of various types of match data to obtain target fusion data.

[0066] It should be noted that frame number alignment refers to aligning various information in the match data according to frames to ensure that the data and the time point of the video are consistent. Specifically, according to the timestamp information of the match data, each frame of data is matched with the corresponding time point frame. The target fusion data refers to the result of integrating all aligned data.

[0067] In this step, the fusion of various types of match data is performed, which is used as input information for the subsequent automatic division model of the football match typical scene. The fusion of multiple data makes the key features analyzed by the model more comprehensive, and the output result is more suitable for the actual match scene, thereby improving the accuracy of the tactical scene division.

[0068] Step S103, time alignment of the match video and the target fusion data to obtain a time-aligned match video as a target match video.

[0069] It should be noted that time alignment refers to synchronizing the match video and the target fusion data to ensure that each frame of target fusion data has a corresponding match video segment. Specifically, linear interpolation or manual marking can be used to achieve time alignment; the target match video refers to the video after time alignment, which contains synchronized match data. For example, each frame of a time-aligned target match video can display player data and match video segments in real time.

[0070] In this embodiment, the match video and the fusion data are time-aligned to obtain the target match video, so that each frame has corresponding match data, providing a basis for the final step S105 of dividing the tactical scene.

[0071] Step S104, inputting the target fusion data into a preset automatic division model of typical scenes of a football match to obtain different scene data segments and corresponding starting frame numbers; wherein the scene data segments represent different granularities and staged football match scene segments.

[0072] It should be noted that the automatic division model of typical scenes of a football match refers to a pre-trained model for identifying and dividing different scenes in a match, for example, a preset neural network model can be pre-trained using match video data with annotated scene types; the scene data segments and corresponding starting frame numbers refer to the match scenes identified by the model and the starting frame numbers of these scenes in the video. For example, the model identifies that an attack scene starts at frame 5000 and a defense scene starts at frame 8000.

[0073] Step S105, performing stage division on the target match recording video according to the different scene data segments and corresponding starting frame numbers to obtain a tactical scene division result.

[0074] It should be noted that stage division refers to dividing a match recording according to the identified scene data segments, for example, dividing the match recording into multiple segments, each segment representing a specific tactical scene.

[0075] It can be understood that the scene segments and starting frame numbers identified according to the above steps perform stage division on the target match recording video that has been time-aligned, generating a more detailed tactical scene division result, overcoming the problem of single division in the related art, and obtaining a comprehensive analysis result of the entire match at one time.

[0076] In this embodiment, by aligning multiple types of match data in frame numbers and time-aligning with the match recording video, the data is first highly fused, making subsequent analysis more accurate. Compared with the traditional overall analysis method, the present application can perform staged division on the match and perform detailed analysis on different tactical scenes, overcoming the defect that the prior art cannot perform detailed analysis and can only perform single analysis. By inputting the target fusion data into a preset automatic division model of typical scenes of a football match to obtain different scene data segments and corresponding starting frame numbers, performing stage division on the target match recording video, the subjective errors in manual marking and the problem of low efficiency are avoided, the conditions of multiple tactics in the match scene can be quickly and accurately obtained, and the efficiency of dividing scenes is improved.

[0077] In this embodiment, a method for dividing tactical scenes of a football match is provided, which can be used in the above-mentioned computer, and the flow includes the following steps:

[0078] Step S201, obtaining a football match video and multiple types of match data. For details, please refer to Figure 1 Step S101 of the embodiment shown, which will not be described here.

[0079] Step S202, aligning the multiple types of match data by frame number to obtain target fusion data.

[0080] Specifically, the multiple types of match data include player data, event data, and tracking data. The above step S202 includes:

[0081] Step S2021, sequentially matching the player data with the tracking data and event data, and confirming target tracking data and multiple target event data corresponding to the player data; wherein the tracking data represents the positions of the ball and the players in each frame of tracking pictures within a tracking time range; and the event data represents key events occurring in each frame of event pictures within a match time range.

[0082] It should be noted that the player data refers to the information of the participating players in the match, including the player list, jersey number, player age, and coach list of different teams, etc. The tracking data refers to the data collected by wearable devices or computer vision technology, which records the position information of the players and the ball on the court. The data frequency is 10Hz, 25Hz, etc., corresponding to 10 frames or 25 frames per second. The event data refers to the data recording the key events occurring in the match, such as kick-off, pass, interception, and shot, etc. The data frequency is not fixed, and each frame records a key event.

[0083] Examples 1 and 2 are given below to explain the tracking data and event data, respectively:

[0084] Example 1: A coordinate system is established at a certain point of the four corners of the court or the center of the court. Each frame of tracking data records the positions of the 22 players and the ball relative to the court at that moment. Therefore, the positions of the players and the ball can be represented by their coordinates on the court. The coordinates of the 11 players of the home team can be represented as representing the 11 players of the home team.

[0085] Example 2: The information it contains includes the type of event occurring at a certain time (i.e. the xth minute and xth second of the match), the players involved, the location of the event, and other related information, where each item is represented by a corresponding field variable, such as event type represented by type. If it is a pass, then type = 'pass'. The event data collected by different data vendors has different levels of detail in recording the match and different field representations, but in general, it can record the key events of a match.

[0086] It can be understood from the above two examples that the tracking data exists in every frame during the whole game, and the event data only exists in the starting frame and the ending frame of the event, so the event data needs to be fused with the tracking data to determine which time frame the key event occurs in the whole game, thereby improving the prediction ability of the model.

[0087] In step S2022, the time frame of the target event data and the frequency of the target tracking data are obtained. The frequency of the target tracking data represents the time interval of the tracking data collection.

[0088] In step S2023, the tracking data is grouped and labeled according to the frequency of the tracking data, and a group set of target tracking data grouping labels is obtained. Each group in the group set contains the same preset number of time frames, and each time frame of each group has a corresponding tracking data segment.

[0089] It can be understood that the functions of steps S2022 and S2023 are to fuse the target event data with the tracking data, and the frequency of the tracking data can determine the time frame of the tracking data. Only by corresponding the time frames of the two can the data fusion be further performed. The event data is generally related to the ball holder, so one important event is recorded in one frame of event data, and the same frame of data will not record two events. This one-to-one relationship between the time frame and the event data can make the fused data more accurate and will not have different event data in the same frame.

[0090] In step S2024, all the target event data is inserted into the target tracking data according to the time frame of the target event data and the group set of the target tracking data grouping labels, and target fusion data is obtained.

[0091] For example, if the frequency of the tracking data is 10 Hz, each 10 frames are labeled as a group, and if the frequency is 25 Hz, each 25 frames are labeled as a group. The first frame and the last frame of each group of data are converted into the time of the game.

[0092] If the collection frequency is 10 Hz, the first frame of data is 0 minutes and 0 seconds of the game, the 1-10 frame data is packaged and labeled as the 0th group, the 11th frame of data is 0 minutes and 1 second of the game, the 11-20 frame data is labeled as the 1st group, the 21st frame of data is 0 minutes and 2 seconds of the game, and the 21-30 frame data is packaged and labeled as the 2nd group. The same is true for the 21-30 frame data. After grouping, the target event data is inserted into the grouped tracking data according to the time frame of the target event data. For example, a target event data of "kick-off event" is inserted into the first group of tracking data.

[0093] It can be understood that after data fusion, the model can consider both player position tracking data and game events, thereby capturing more game dynamic information. This can help the model better understand the movement patterns and tactical intentions of players, improving prediction accuracy.

[0094] Specifically, the step S2024 includes:

[0095] A1, according to the time of the target event data, sequentially find the group set marked by the target tracking data group.

[0096] A2, when the time of the target event data is found to exist in any target group in the group set marked by the target tracking data group, delete the tracking data segment corresponding to the position of any time frame in the target group, and supplement the target event data to the position of the last time frame in the target group; until all the target event data are inserted into the target tracking data, obtaining the target fusion data.

[0097] For example, the first time frame of the event data records the kick-off at 0 minute 1 second of the game, and this event data is inserted into the first group tracking data. One frame of tracking data in the first group data is randomly deleted, and this event data is taken as the 10th data of the first group data. Finally, the target fusion data of each group of inserted time data has the same dimension, such as 10 frames or 25 frames.

[0098] In addition, if there is no event data recorded at the time point corresponding to a group of tracking data, all 10 frames of tracking data in this group are retained, thereby obtaining the fused data, and each group of fused data has the same number.

[0099] The embodiment can maintain the same dimension of the target fusion data of each group, maintain the accurate alignment of the tracking data and event data in time, and keep consistent, thereby improving the accuracy of data fusion, by sequentially finding the time of the target event data and ensuring that the event data is inserted into the correct tracking data group, deleting the position data of the corresponding time frame in the target tracking data group, and inserting the event data into the last time frame.

[0100] Step S203, time align the game video with the target fusion data to obtain the time-aligned game video as the target game video. For details, please refer to Figure 1 The step S103 of the embodiment shown in the figure will not be repeated here.

[0101] Step S204, inputting the target fusion data into a preset automatic division model of typical scenes of a football match to obtain different scene data segments and corresponding starting frame numbers; wherein the scene data segments represent different granularities and staged football match scene segments. For details, please refer to Figure 1 Step S104 of the embodiment shown in the figure will not be repeated here.

[0102] Step S205, performing stage division on the target match video according to the different scene data segments and the corresponding starting frame numbers to obtain a tactical scene division result. For details, please refer to Figure 1 Step S105 of the embodiment shown in the figure will not be repeated here.

[0103] In this embodiment, the position information tracking data and the event data of the players are fused to form complete fusion data. The frequency of the tracking data is grouped and marked to ensure the time sequence consistency of the data, so that each event data can be accurately matched to the corresponding tracking data time frame. After data fusion, the model can not only obtain the position and action of the players, but also combine specific match events such as passing and shooting. Compared with the manual annotation or the single analysis of a certain tactical scene in the related art, the target fusion data is obtained in advance in this embodiment to provide accurate data basis for subsequent model identification of the tactical scene, so that the model can more comprehensively understand the tactical behavior of the players and improve the accuracy of the analysis result.

[0104] In this embodiment, a method for dividing a tactical scene of a football match is provided, which can be used in the above-mentioned computer. The flow includes the following steps:

[0105] Step S301, obtaining a match video of a whole football match and multiple types of match data. For details, please refer to Figure 1 Step S103 of the embodiment shown in the figure will not be repeated here.

[0106] Step S302, frame number alignment of the multiple types of match data to obtain target fusion data. For details, please refer to Figure 1 Step S102 of the embodiment shown in the figure will not be repeated here.

[0107] Step S303, time alignment of the match video and the target fusion data to obtain a time-aligned match video as a target match video. For details, please refer to Figure 1 Step S103 of the embodiment shown in the figure will not be repeated here.

[0108] Step S304, inputting the target fusion data into a preset automatic division model of typical scenes of a football match to obtain different scene data segments and corresponding starting frame numbers; wherein the scene data segments represent different granularities and staged football match scene segments.

[0109] Specifically, the match data further includes match time, and step S304 includes:

[0110] Step S3041, dividing the match time into first-half match time and second-half match time according to frame numbers.

[0111] Step S3042, extracting ball possession state information corresponding to target event data in the target fusion data.

[0112] Step S3043, dividing ball possession flow segments in the target fusion data according to the ball possession state information.

[0113] Step S3044, defining corresponding scene labels for the ball possession flow segments based on preset tactical scene types, first-half match time, and second-half match time.

[0114] Step S3045, performing feature engineering on the ball possession flow segments with scene labels to extract tracking data features and event data features.

[0115] Step S3046, constructing high-level features for the target fusion data according to the tracking data features and the event data features, wherein the high-level features are used to describe spatial positions and tactical layouts of players, a ball, and time on a match field.

[0116] Step S3047, inputting the target fusion data after constructing the high-level features into a preset automatic division model of typical scenes of a football match to obtain different scene data segments and corresponding starting frame numbers.

[0117] It should be noted that dividing the match time into first-half match time and second-half match time according to frame numbers means that a match video or data is cut according to time frames (a fixed number of frames per second, for example, 25 frames) to mark specific time points in the match. The ball possession state information refers to which team is in possession of the ball at a certain time, which is determined by event data. The ball possession flow refers to a segment in which a team controls the ball for a continuous period of time. The scene label refers to labeling each ball possession flow segment with a corresponding tactical scene label, for example, a certain ball possession flow belongs to low defense.

[0118] Specifically, in step S3041, first, the end time t1 of the first half, the start time t2 of the second half, and the end time t3 of the match are extracted from the match information data. eThe three time points are used to roughly divide the whole game. In step S3042, the ball_control field in the event data in the fusion data is extracted to distinguish the home team and the away team. In step S3043, the possession flow is labeled by using expert knowledge, that is, the three fields of “cut-off”, “shot”, and “foul” are used as the end markers of the possession flow, and the possession flow of the home team and the away team is divided according to the value of ball_control, that is, the same value of ball_control in a plurality of consecutive needles, and an arbitrary marker is used to end the possession flow. In step S3044, the possession flow segment is defined as different tactical scenes such as “low defense” and “high pressure”, and the possession flow segment is marked according to the time of the first half and the second half. In step S3045, the tracking data features such as the position and speed of the player and the position and speed of the ball are extracted, and the event data features such as the event type, the position, and the related player information are extracted. In step S3046, the high-level features describing the spatial positions of the player, the ball, and the time on the game field and the tactical layout are constructed according to the extracted tracking data features and event data features. In step S3047, the automatic division model is used to automatically divide different scene data segments according to the constructed high-level features to obtain the scene data segments and the corresponding starting frame numbers.

[0119] The division of the possession flow segment in step S3043 is described in Example 3 as follows:

[0120] In Example 3, it is assumed that the value of ball_control in 15 consecutive frames is 1, and “cut-off” appears in the 16th frame, and the 15 frames are divided into active possession. If the value of ball_control jumps but does not appear for more than 5 frames, the subsequent ball_control jumps back to the original value, and more than 10 frames (1 second), it is still considered as the same team having the possession, for example, the value of ball_control in 10 consecutive frames is 1, the value of ball_control in the 11th and 12th frames is 0, the value of ball_control after the 13th frame is 1, and the shot appears in the 20th frame, and the 10-20 frames are all the possession of the home team. Thus, the possession flow of the home team and the away team in a game is automatically divided, the starting frame number of each possession flow is frame s , and the ending frame number is frame e . frame s corresponds to the group in which it is located, and frame s is replaced by the group number marker, for example, if the starting frame number frame s of a possession flow segment is 51, the group number in which it is located is the 5th group, frame s is replaced by 5; frame eSimilarly, the replacement is made. Thus, the time markers of the start and end of the ball possession actions of the home team and the away team are obtained, and the ball possession flow segments are obtained according to the ball possession segments and the time markers of the start and end.

[0121] In addition, for more detailed tactical scene segments, the embodiment of the application utilizes a knowledge and data collaborative driving manner to automatically divide. This embodiment takes the automatic screening of the “low-position defense” scene as example 4 to describe step S3044, but the specific scene division of the embodiment is not limited to the “low-position defense” scene, and also includes common tactical scenes such as “high-position pressure”, “high-position pressing”, and “counter-pressing”.

[0122] Example 4: According to the ball possession flow segments, the definition knowledge of the “low-position defense” of the professional coach and the football analyst is used to mark the ball possession flow segments according to the first half and the second half of the game time. Specifically, whether the home team and the away team ball possession flow segments screened out by the positions of the three defensive lines of the home team and the away team and the positions of the attacking players, the defensive players, and the ball relative to the three defensive lines are in the “low-position defense” scene is judged, including the segments in which the home team and the away team are in the “low-position defense” respectively, which are distinguished by the ball_control field. Finally, all the ball possession flow segments of the first half and the second half are marked.

[0123] In an optional embodiment, step S3046 constructs high-level features describing the spatial positions of the players, the ball, and the time on the game field and the tactical layout according to the extracted tracking data features and event data features, which is described in example 5:

[0124] Example 5: A coordinate system is established with the lower left corner of the field as the coordinate origin, in which the positions and speeds of the players are two-dimensional, and the position and speed of the ball are three-dimensional. The coordinates of the home team players can be expressed as The speed is expressed as The coordinates of the away team players can be expressed as The speed is expressed as The coordinates of the ball can be expressed as b(x, y, z), and the speed of the ball can be expressed as v b (x, y, z). According to the above information, the following five types of high-level features can be determined.

[0125] The first type of high-level feature: the relative distances between the players and between the players and the ball are constructed as the first type of high-level feature by using the absolute coordinates of the players and the ball;

[0126] The second type of high-level feature: the three attacking defensive lines are clustered by using the horizontal coordinates (x-axis coordinates) of the players and the ball as the second type of high-level feature;

[0127] The third type of advanced features: the player and the ball coordinates are reduced in dimension, and the positional relationship with the three defensive lines is obtained as the third type of advanced features.

[0128] The fourth type of advanced features: the event type field, the position of the event occurrence, and the related player in the event data are extracted as the fourth type of advanced features.

[0129] The fifth type of advanced features: the relative distance (vertical distance) between the position of the event occurrence and each defensive line is calculated as the fifth type of advanced features.

[0130] It is worth noting that the construction of the fourth and fifth types of advanced features is based on the fusion data. When there is a fusion data group without event data (i.e., a group in which event data is not inserted in the tracking data after grouping), the fourth and fifth types of advanced features can be aligned with other data by filling zeros. This is because the feature input dimension of the model needs to be consistent, such as a 20-dimensional vector or a 3*3 matrix. If the input dimension is inconsistent, the model cannot handle it. In related technologies, there are generally two ways to handle missing values: (1) directly discard samples with missing values, and (2) fill data with mean, median, or specified numbers. Since the problem to be solved by the embodiments of the present application is to enrich the dimensions and information quantity of football data, data filling is selected to ensure sufficient data quantity for subsequent model training.

[0131] It can be understood that the first and fourth types of features constructed in Example 5 directly extract original information from tracking data, ensuring the accuracy of the data. The second, third, and fifth types of features construct advanced features from different abstraction levels based on the basic information, ensuring feature diversity. By constructing multiple advanced features, the multi-dimensional information and potential associations in the basic data can be deeply mined, and the expression ability of the data can be improved. The construction of advanced features not only stops at the original data level, but also further abstracts and refines key tactical layouts and spatial relationships, greatly enriching the feature input of the model and improving the recognition ability of the model.

[0132] In step S305, the target match video is divided into stages according to the different scene data segments and the corresponding starting frame numbers, and a tactical scene division result is obtained.

[0133] Specifically, in step S305, the following steps are included.

[0134] In step S3051, the starting frame numbers corresponding to different scene data segments are converted into match tactical starting times according to the time conversion format.

[0135] In step S3052, the target match video is divided into stages based on the match tactic starting time, to obtain a tactic scene division result, wherein the tactic scene division result includes a plurality of marked association information, and the marked association information includes a video clip and a corresponding scene data clip.

[0136] For example, if the match video has 25 frames per second, a dribble flow clip has a starting frame number of 375 and an ending frame number of 500, and the clip is marked as a low-position defense tactic scene, the frame numbers are converted into time. The match starting time is 0 seconds, the starting frame number 375 is converted into time of 15 seconds, the ending frame number 500 is converted into time of 20 seconds, and the time range of the dribble flow clip is 15 seconds to 20 seconds. Based on the above time range, the match video is divided into stages, and in the match video, the part from 15 seconds to 20 seconds is the low-position defense tactic scene, and the match video clip corresponding to the part is the video clip of the low-position defense tactic scene.

[0137] In this embodiment, the frame numbers are converted into time, so that the specific tactic scene in the match video can be accurately located. Each tactic scene includes marked association information, which helps football coaches and data analysts to more easily understand and evaluate the tactical execution of the team.

[0138] In this embodiment, the match time is divided into the first half and the second half according to the frame number, and the target fusion data is divided into a dribble flow clip in combination with the possession state information, so as to facilitate subsequent analysis and extraction of high-level features. Based on a preset tactic scene type, the dribble flow clip is defined as a corresponding scene label, to ensure more accurate analysis. By extracting tracking data features and event data features and constructing high-level features, diversified feature inputs are ensured, to help the model better understand the complex pattern relationship in the fusion data, improve the recognition accuracy, and thus be able to describe the overall tactical layout on the match field, rather than being limited to the analysis of a single player or event in related technologies.

[0139] In an optional implementation, the football match typical scene automatic division model is a model obtained by pre-training, wherein the pre-training process includes: determining a basic training model according to the football match tactic scene classification requirement; obtaining a football match scene data set, and inputting the data set into the basic training model after preprocessing to perform iterative training, to obtain the football match typical scene automatic division model.

[0140] It should be noted that the determination process of the basic training model specifically comprises: determining the basic training model according to the classification requirement of the football match tactical scene. The basic training model can be a deep neural network or a decision tree model (such as xGboost or LightGBM). The pre-processing of the data set can be data cleaning, feature extraction, etc. The present embodiment provides the following two ways for iterative training, but is not limited thereto:

[0141] Firstly, when the basic training model is a deep neural network, the network parameter setting includes the number of network hidden layers, the number of neurons in each layer of hidden layer, the connection relationship between different layers of neurons, the activation function adopted by each neuron, the loss function of the whole neural network, the initialized input, the sample size of each batch for training, and the random seed of the training network. The training process comprises inputting the pre-processed data set into the basic training model for iterative training to obtain a football match typical scene automatic division model.

[0142] Secondly, when the basic training model is a decision tree model, the network parameter setting includes the loss function, the number of layers of the decision tree, the number of leaf nodes, the maximum number of iterations, the value of the random seed, the number of samples classified by each leaf, and the weight of each sample. Other related parameters can be set in the form of default value or random value. The training process is also to input the pre-processed data set into the basic training model for iterative training.

[0143] It is worth noting that for a deep neural network, the full connection is usually adopted between different layers, and the activation function can be selected from ReLU, LeakyReLU, sigmoid, tanh, etc. According to the data volume and the number of network layers, appropriate selection is needed, and ReLU or LeakyReLU is preferred.

[0144] The loss function is used to measure the difference between the predicted value and the true value, and the smaller the difference, the stronger the prediction ability of the model. For a classification task, cross-entropy is generally selected as the loss function; for a regression task, mean square error (MSE) is generally selected as the loss function. When the tactical scene recognition is set as a classification task, cross-entropy is preferably selected as the loss function.

[0145] For a decision tree model, the selection of related parameters such as the number of layers, the number of leaf nodes, the maximum number of iterations, and the number of samples classified by each leaf is determined according to the data volume used during model training. Specifically, the default value during interface calling is taken as the initial value, or it is set according to experience, and then adjusted according to the training effect of the model.

[0146] The embodiment enhances the recognition ability of the pre-trained model for different scenes by training with a large number of football match scene data sets, and the trained automatic division model can complete the recognition of complex football tactical scenes with less manual operation, reduces the error of manual annotation, and is convenient for continuous update and iterative learning, and can continuously update the model.

[0147] The overall process of automatic division of typical tactical scenes in football matches in the embodiment can be referred to as Figure 2 The tracking data, event data and match information data are fused into the data alignment module, the target fusion data output by the module is secondarily time-aligned with the match video, after the fusion data is divided by the automatic division module of typical tactical scenes, coarse-grained results including the first half and the second half time, the home team and the away team ball possession flow, the penalty kick and the like are obtained, fine-grained results including high position compression, high position pressing, low position defense, counter pressure and the like are obtained, the coarse-grained and fine-grained results are analyzed with the match video segments, finally the data segments and video segments with labels are obtained, and the divided data segments and video segments are related.

[0148] The overall process of constructing advanced features in the embodiment can be referred to as Figure 3 The fusion data is labeled according to expert knowledge, finally preprocessed data containing tracking data, event data and category labels are obtained, and feature engineering is performed on the preprocessed data, five types of advanced features are constructed, training sets and validation sets are obtained according to the proportion, parameters are set, and deep learning network models or decision tree models are trained, finally the model is obtained, and different tactical scenes can be divided.

[0149] In actual application, the finally divided results will store the video segments and the corresponding data segments after being packaged and labeled with IDs, when a coach or an analyst searches for or searches for corresponding tactical scene labels, corresponding data segments and video segments can be obtained, post-match review can be performed in a more targeted manner, players can analyze the technical and tactical performance and tactical effect in different tactical scenes, and the sports performance level and tactical arrangement can be further improved.

[0150] Specifically, the search of labels is performed through code commands, specific tactical division scene display information is obtained, and the specific tactical division scene display information is shown in FIG. 6. Figure 4 , Figure 4 is the related information of the "low position defense" segment automatically divided in a certain match. The ID represents the identification information of the tactical scene, the possession_team represents the ball possession team, the start_time and the end_time respectively represent the start time and the end time of the tactic, and the home team or the away team executes the "low position defense" tactic in the time period.

[0151] An apparatus for dividing a tactical scene of a football match is also provided in the embodiment. The apparatus is used to implement the above-described embodiments and preferred embodiments, and will not be described again. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.

[0152] The embodiment provides an apparatus for dividing a tactical scene of a football match, as shown in the accompanying drawings, comprising: Figure 5

[0153] a data collection module 501 configured to obtain a match video of a whole football match and a plurality of types of match data;

[0154] a data alignment module 502 configured to align the plurality of types of match data in terms of frame number to obtain target fusion data;

[0155] a time synchronization module 503 configured to time-align the match video with the target fusion data to obtain a time-aligned match video as a target match video;

[0156] a model output module 504 configured to input the target fusion data into a preset automatic division model of a typical scene of a football match to obtain different scene data segments and corresponding starting frame numbers; wherein the scene data segments represent different granularities and phased football match scene segments;

[0157] a tactical scene division module 505 configured to divide the target match video in phases according to the different scene data segments and corresponding starting frame numbers to obtain a tactical scene division result.

[0158] In an optional embodiment, the plurality of types of match data includes player data, event data, and tracking data, and the data alignment module 502 specifically includes:

[0159] a matching unit configured to sequentially match the player data with the tracking data and event data to confirm target tracking data corresponding to the player data and a plurality of target event data; wherein the tracking data represents a ball position and a player position in each frame of tracking pictures within a tracking time range; and the event data represents a key event occurring in each frame of event pictures within a whole match time range;

[0160] a time frame and frequency acquisition unit configured to acquire a time frame of the target event data and a frequency of the target tracking data; wherein the frequency of the target tracking data represents a time interval of tracking data acquisition;

[0161] ​a packet marking unit, configured to mark packets of the tracking data according to frequencies of the tracking data, to obtain a group set of packet marking of target tracking data, each group in the group set containing a same preset number of time frames, and each time frame of each group having a corresponding tracking data segment;

[0162] an insertion fusion unit, configured to insert all of the target event data into the target tracking data according to the time frames of the target event data and the group set of packet marking of the target tracking data, to obtain target fusion data.

[0163] In an optional implementation, the insertion fusion unit specifically includes:

[0164] a group set searching subunit, configured to search the group set of packet marking of the target tracking data according to the time of the target event data in sequence;

[0165] an insertion data subunit, configured to, when the time of the target event data is found to exist in any target group in the group set of packet marking of the target tracking data, delete a tracking data segment corresponding to a position of any time frame in the target group, and supplementally insert the target event data into a position of a last time frame in the target group; until all of the target event data are inserted into the target tracking data, to obtain the target fusion data.

[0166] In an optional implementation, the competition data further includes competition time, and the model output module 504 specifically includes:

[0167] a competition time dividing unit, configured to divide the competition time into first-half competition time and second-half competition time according to a frame number;

[0168] a ball possession state extraction unit, configured to extract ball possession state information corresponding to the target event data in the target fusion data;

[0169] a ball possession flow dividing unit, configured to divide a ball possession flow segment in the target fusion data according to the ball possession state information;

[0170] a scene label defining unit, configured to define a corresponding scene label for the ball possession flow segment based on a preset tactical scene type, the first-half competition time and the second-half competition time;

[0171] a feature engineering unit, configured to perform feature engineering on the ball possession flow segment having the scene label, to extract tracking data features and event data features;

[0172] The advanced feature construction unit is configured to construct advanced features of the target fusion data according to the tracking data features and the event data features, wherein the advanced features are used to describe spatial positions of players, a ball and time on a playing field and a tactical layout;

[0173] The scene data segment generation unit is configured to input the target fusion data after the advanced features are constructed into a preset automatic scene division model of a typical football game to obtain different scene data segments and corresponding start frame numbers.

[0174] In an optional embodiment, the tactical scene division module 505 specifically comprises:

[0175] The frame number conversion unit is configured to convert the start frame numbers corresponding to the different scene data segments into start times of a game tactical layout according to a time conversion format.

[0176] The stage division unit is configured to divide the target game video into stages based on the start times of the game tactical layout to obtain a tactical scene division result, wherein the tactical scene division result comprises a plurality of marked association information, and the marked association information comprises an association relationship between a video segment and a corresponding scene data segment.

[0177] In an optional embodiment, the automatic scene division model of a typical football game is a model obtained by pre-training, wherein the pre-training process comprises: determining a basic training model according to a classification requirement of a football game tactical scene; obtaining a football game scene data set, and inputting the data set into the basic training model after pre-processing to perform iterative training to obtain the automatic scene division model of a typical football game.

[0178] Further function descriptions of the above various modules and units are the same as those of the above corresponding embodiments, and will not be described here.

[0179] The device for dividing a football game tactical scene in the embodiment is presented in the form of a functional unit, wherein the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory executing one or more software or fixed programs, and / or other devices that can provide the above functions.

[0180] The embodiment of the application further provides a computer device having the device for dividing a football game tactical scene shown in the above Figure 5 .

[0181] Please refer to Figure 6 , Figure 6 is a structural schematic diagram of a computer device provided by an optional embodiment of the application, as shown inFigure 6 As shown, the computer device includes one or more processors 10, memory 20, and interfaces 30 for the various components to communicate with one another. The various components communicate through one or more buses, and can be mounted on a common motherboard or in other manners as appropriate. The processor 10 can execute instructions, for example, stored in the memory 20 or elsewhere to implement processes that the computer device implements. In some embodiments, multiple processors can be used, as appropriate, along with multiple buses. Both the memories and the processors can be used as storage and retrieval elements in processes that the computer device implements. Figure 6 The processor 10 is used in the embodiments below as an example.

[0182] The processor 10 can be a central processing unit, a network processor, or a combination thereof. The processor 10 can further include a hardware chip. The hardware chip can be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device can be a complex programmable logic device, a field programmable logic device, a general array logic, or any combination thereof.

[0183] The memory 20 stores instructions that can be executed by the at least one processor 10, to cause the at least one processor 10 to perform the methods illustrated by the above embodiments.

[0184] The memory 20 can include a program storage area and a data storage area. The program storage area can store an operating system, application programs, and the like for the computer device. The data storage area can store data that the computer device creates and / or uses while performing its functions. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 20 can optionally include a memory that is remote from the processor 10, and can be connected to the computer device through a network. Examples of the network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and a combination thereof.

[0185] The memory 20 can include a volatile memory, such as a random access memory, and can also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. The memory 20 can also include a combination of the above-mentioned types of storage.

[0186] The computer device also includes a communication interface 30 that enables the computer device to communicate with other devices or networks with other devices or communication networks.

[0187] The embodiments of the present application further provide a computer readable storage medium, and the method according to the embodiments of the present application can be implemented in hardware, firmware, or recorded in a storage medium, or stored in a remote storage medium or a non-transitory machine readable storage medium and downloaded to a local storage medium through network, so that the method described herein can be processed by such software on a storage medium using a general purpose computer, a special purpose processor, or programmable or special hardware. The storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid state disk, etc. Further, the storage medium can also include a combination of the above-mentioned memories. It can be understood that the computer, the processor, the microprocessor controller, or the 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 the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0188] Part of the present application can be applied as a computer program product, for example, computer program instructions, when executed by a computer, through the operation of the computer, the method and / or technical solutions according to the present application can be called or provided. Those skilled in the art should understand that the form of computer program instructions in computer readable medium includes but is not limited to source file, executable file, installation package file, etc. Correspondingly, the way of computer program instructions executed by computer includes but is not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Here, the computer readable medium can be any available computer readable storage medium or communication medium accessible to the computer.

[0189] Although the embodiments of the present application are described in conjunction with the accompanying drawings, various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application, and such modifications and changes fall within the scope defined by the claims.

Claims

1. A method for dividing tactical scenarios in a football match, characterized in that, The method includes: Get full match video recordings of football games and various types of match data; The target fused data is obtained by aligning the frames of various types of match data, including player data, event data, and tracking data. Specifically, this process includes: The player data is sequentially matched with the tracking data and event data to identify the target tracking data and multiple target event data corresponding to the player data; wherein, the tracking data represents the position of the ball and the position of the player in each frame of the tracking image within the tracking time range; the event data represents the key events that occur in each frame of the event image within the entire game time range; The time frame of the target event data and the frequency of the target tracking data are obtained; wherein, the frequency of the target tracking data represents the time interval of tracking data collection; The tracking data is grouped and labeled according to the frequency of the tracking data to obtain a group set of target tracking data group labels. Each group in the group set contains the same preset number of time frames, and each time frame of each group contains a corresponding tracking data segment. According to the time frame of the target event data and the group set of the target tracking data grouping markers, all the target event data is inserted into the target tracking data to obtain target fusion data, including: Based on the time of the target event data, sequentially search the group set of the target tracking data grouping markers; When the time of the target event data is found to exist in any target group in the group set of target tracking data grouping tags, the tracking data segment corresponding to the position of any time frame in the target group is deleted, and the target event data is inserted into the position of the last time frame in the target group; until all the target event data has been inserted into the target tracking data, the target fusion data is obtained. The match video recording is time-aligned with the target fused data to obtain a time-aligned match video recording, which is used as the target match video recording. The target fusion data is input into a preset automatic segmentation model for typical football match scenarios to obtain different scene data segments and corresponding starting frame numbers; wherein, the scene data segments represent football match scene segments with different granularities and stages. Based on the different scene data segments and their corresponding starting frame numbers, the target match video is divided into stages to obtain tactical scene division results; The match data also includes match time. The step of inputting the target fused data into a preset automatic segmentation model for typical football match scenes to obtain different scene data segments and corresponding starting frame numbers specifically includes: The match time is divided into the first half and the second half based on the number of frames. Extract the ball control status information corresponding to the target event data from the target fusion data; The ball control flow segments in the target fusion data are divided according to the ball control status information; Based on preset tactical scenario types, the first half time, and the second half time, corresponding scenario labels are defined for the ball possession flow segments; Feature engineering is performed on ball control flow segments with scene labels to extract tracking data features and event data features; Based on the tracking data features and the event data features, high-level features are constructed on the target fusion data, wherein the high-level features are used to describe the spatial position and tactical layout of players, ball and time on the field of play; The target fusion data after constructing the advanced features is input into a preset automatic segmentation model of typical football match scenes to obtain different scene data segments and corresponding starting frame numbers.

2. The method according to claim 1, characterized in that, The step of dividing the target match video recording into stages based on the different scene data segments and their corresponding starting frame numbers to obtain tactical scene segmentation results specifically includes: The starting frame numbers corresponding to different scene data segments are converted into the start time of the game tactics according to the time conversion format. Based on the start time of the competition tactics, the target competition video is divided into stages to obtain tactical scene division results. The tactical scene division results include multiple marked association information, which includes the relationship between video segments and corresponding scene data segments.

3. The method according to claim 1, characterized in that, The automatic segmentation model for typical football match scenarios is a model obtained through pre-training. The pre-training process includes: determining a basic training model based on the classification requirements of football match tactical scenarios; obtaining a football match scenario dataset, and inputting the dataset into the basic training model for iterative training to obtain the automatic segmentation model for typical football match scenarios.

4. A device for dividing tactical scenarios in a football match, characterized in that, Based on the method of claim 1, the apparatus comprises: The data collection module is used to acquire full match video recordings of football games and various types of match data; The data alignment module is used to align the frames of various types of match data to obtain target fused data. The time synchronization module is used to time-align the match video recording with the target fused data to obtain the time-aligned match video recording, which is used as the target match video recording. The model output module is used to input the target fused data into a preset automatic segmentation model of typical football match scenes to obtain different scene data segments and corresponding starting frame numbers; wherein, the scene data segments represent football match scene segments with different granularities and stages. The tactical scene segmentation module is used to divide the target match video recording into stages based on the different scene data segments and the corresponding starting frame number, so as to obtain the tactical scene segmentation result.

5. A computer device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the method for dividing tactical scenarios in a football match as described in any one of claims 1-3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the method for dividing a football match tactical scenario according to any one of claims 1-3.

7. A computer program product, characterized in that, Includes computer instructions for causing a computer to perform the method for dividing a football match tactical scenario according to any one of claims 1-3.

Images