Method, system, medium for identifying team formation during positioning attack

Abstract

Disclosed herein are methods, systems, media for identifying a team formation during a positional attack. A computing system receives one or more tracking data streams. The computing system identifies a positional attack contained in the one or more tracking data streams. The computing system identifies a defensive formation of a first team and an offensive formation of a second team. The computing system extracts, via a convolutional neural network, one or more features corresponding to a type of defensive formation implemented by the first team by passing the positional attack through the convolutional neural network. The computing system scans, via a machine learning algorithm, the positional attack to identify one or more features indicative of a type of offensive formation implemented by the second team. The computing system derives the type of defensive formation implemented by the first team.

Description

[0001] This application is a divisional application of the parent application for the invention patent application with application number 201980018888.2 (International application number: PCT / US2019 / 014614, filing date: January 22, 2019, invention name: Method, system, medium for identifying team alignment during set-piece attacks).

[0002] Cross Reference to Related Applications

[0003] This application claims the benefit of U.S. Provisional Application Serial No. 62 / 619,896, filed January 21, 2018, which is incorporated by reference herein in its entirety. TECHNICAL FIELD

[0004] The present disclosure relates generally to systems and methods for classifying attacking team alignment versus defending team alignment. BACKGROUND

[0005] In the 2016-2017 Premier League season, approximately 16% of all goals scored came from set-pieces (e.g., corners and free kicks). However, there is also a large variance between teams in these numbers, such as West Bromwich Albion who had 16 of their 43 goals come from set-pieces (> 35% of their goals), while other teams had less than 7% of their goals come from set-pieces (e.g., Sunderland had 2 of their 29 goals come from set-pieces). The resource gap between the richest and poorest teams in world football is growing season by season, as evidenced by Paris Saint-Germain’s record-breaking 220 million Euro spend on Neymar who scored 15 goals. The ability of smaller market teams to replicate the same goal output at the price of an effective set-piece strategy is a market inefficiency that can be exploited. Therefore, a new method that can help teams exploit this inefficiency would be a key advantage. SUMMARY

[0006] Embodiments disclosed herein generally relate to systems and methods for classifying offensive team formations and defensive team formations. In some embodiments, a method of identifying defensive formations and offensive formations in a set play is disclosed herein. A computing system receives one or more tracking data streams associated with one or more games. The computing system identifies a set play included in the one or more tracking data streams. The computing system identifies a defensive formation of a first team and an offensive formation of a second team. The computing system extracts, via a convolutional neural network, one or more features corresponding to a type of defensive formation implemented by the first team by passing the set play through the convolutional neural network. The computing system scans, via a machine learning algorithm, the set play to identify one or more features indicative of a type of offensive formation implemented by the second team. The computing system infers, via the machine learning algorithm, the type of defensive formation implemented by the first team based at least on the one or more identified features and the one or more extracted features.

[0007] In some embodiments, a system of identifying defensive formations and offensive formations in a set play is disclosed herein. The system includes a processor and a memory. The memory has programming instructions stored thereon that, when executed by the processor, perform one or more operations. The one or more operations include one or more tracking data streams associated with one or more games. The one or more operations also include identifying a set play included in the one or more tracking data streams. The one or more operations also include identifying a defensive formation of a first team and an offensive formation of a second team. The one or more operations also include extracting, via a convolutional neural network, one or more features corresponding to a type of defensive formation implemented by the first team by passing the set play through the convolutional neural network. The one or more operations also include scanning, via a machine learning algorithm, the set play to identify one or more features indicative of a type of offensive formation implemented by the second team. The one or more operations also include inferring, via the machine learning algorithm, the type of defensive formation implemented by the first team based at least on the one or more identified features and the one or more extracted features.

[0008] In some embodiments, a non-transitory computer-readable medium is disclosed. The non-transitory computer-readable medium includes one or more sequences of instructions that, when executed by one or more processors, cause performance of one or more operations. A computing system receives one or more streams of tracking data associated with one or more games. The computing system identifies a positioned attack contained in the one or more streams of tracking data. The computing system identifies a defensive formation of a first team and an offensive formation of a second team. The computing system extracts, via a convolutional neural network, one or more features corresponding to a type of defensive formation implemented by the first team by passing the positioned attack through the convolutional neural network. The computing system scans, via a machine learning algorithm, the positioned attack to identify one or more features indicative of a type of offensive formation implemented by the second team. The computing system infers, via the machine learning algorithm, the type of defensive formation implemented by the first team based at least on the one or more identified features and the one or more extracted features. BRIEF DESCRIPTION OF DRAWINGS

[0009] In order that the manner in which the above-recited features of the present disclosure can be understood in detail, a more particular description of the disclosure, briefly summarized above, can be had by reference to embodiments, some of which are illustrated in the appended drawings. It is to be noted, however, that the appended drawings illustrate only typical embodiments of this disclosure and are therefore not to be considered limiting of its scope, for the disclosure can admit to other equally effective embodiments.

[0010] Figure 1 is a block diagram illustrating a computing environment in accordance with example embodiments.

[0011] Figure 2 is a block diagram illustrating a positioned attack syntax model that can define one or more hand-crafted features of a machine learning module in accordance with example embodiments.

[0012] Figure 3A is a block diagram illustrating a graphical representation of a defensive formation of a positioned attack in accordance with example embodiments.

[0013] Figure 3B is a block diagram illustrating a graphical representation of a defensive formation of a positioned attack in accordance with example embodiments.

[0014] Figure 3C is a block diagram illustrating a graphical representation of a defensive formation of a positioned attack in accordance with example embodiments.

[0015] Figure 4 is a flow diagram illustrating a method of training a convolutional neural network in accordance with example embodiments.

[0016] Figure 5 is a flowchart illustrating a method of classifying defensive formations in a positional attack according to example embodiments.

[0017] Figure 6 is a flowchart illustrating a method of classifying defensive formations and offensive formations in a positional attack according to example embodiments.

[0018] Figure 7A is a block diagram illustrating a Hinton diagram that illustrates offensive and defensive positional attack styles of a team according to example embodiments.

[0019] Figure 7B is a block diagram illustrating one or more graphical elements for offensive and defensive styles according to example embodiments.

[0020] Figure 7C is a block diagram illustrating one or more graphical elements for offensive and defensive styles according to example embodiments.

[0021] Figure 8A is a block diagram of a computing device according to example embodiments.

[0022] Figure 8B is a block diagram of a computing device according to example embodiments.

[0023] For ease of understanding, the same reference numbers will be used in different drawings to designate the same elements where possible. It is contemplated that elements disclosed in one embodiment can be beneficially utilized on other embodiments without specific recitation. DETAILED DESCRIPTION

[0024] One or more techniques disclosed herein relate generally to a system and method for classifying and identifying defensive formations and offensive formations in a positional attack during a game. For example, one or more techniques described herein propose an attribute-driven positional attack analysis method that utilizes a hybrid deep learning approach to detect complex attributes, such as defensive marking schemes and manually created features, to enable interpretability.

[0025] Such techniques differ from conventional operations for analyzing positional attacks during a match. In particular, when analyzing the defensive formation of a positional attack, conventional techniques are insufficient to capture the nuances of corner behavior. For example, previous approaches have used a template method of player position versus role. Such an approach can be useful for improving prediction tasks in open play, such as predicting expected goal scoring values on shots or modeling defensive behavior when player movement paths are relatively linear and role swapping is infrequent. However, such approaches assume that all players are important and present in the match, which is not typically the case for positional attacks. For example, some attacking teams can reserve three defenders in the back, meaning there can be four attackers versus eight defenders.

[0026] Another conventional approach involves taking a bottom-up / feature creation approach, where distances and angles between players are computed. However, this approach is limited by the large and complex number of features needed to capture the nuances of defensive formations. Moreover, since the area around the goal can be small and distances close to the goal are most important, subtle movements can be missed by large movements of some players.

[0027] Even though current approaches are insufficient to capture such behavior, the fine-grained space, variable number of players, and different movement patterns (short and long passes) lend themselves to image-based representations. One or more techniques disclosed herein take this approach by implementing deep learning methods.

[0028] Figure 1 is a block diagram illustrating a computing environment 100 in accordance with example embodiments. The computing environment 100 can include a tracking system 102, an organization computing system 104, and one or more client devices 108 in communication via a network 105.

[0029] The network 105 can be of any suitable type, including various connections via the Internet, such as a cellular network or a Wi-Fi network. In some embodiments, the network 105 can use a direct connection, such as radio frequency identification (RFID), near field communication (NFC), near field communication (NFC), Bluetooth TM , Bluetooth TM Low Energy (BLE), Wi-Fi TM , ZigBee TMThe terminal, service, and mobile device can be connected using any of a variety of connection types (e.g., Bluetooth®, Wi-Fi®, cellular, ABC, USB, WAN, or LAN). Because the information being transmitted can be personal or confidential, security considerations can require that one or more of these types of connections be encrypted or otherwise secured. However, in some implementations, the information being transmitted can not be as personal, and network connections can be selected for convenience rather than security.

[0030] The network 105 can include any type of computer networking arrangement used to exchange data or information. For example, the network 105 can be the Internet, a private data network, a virtual private network using public networks and / or other suitable connections that enable the components in the computing environment 100 to send and receive information between the components of the environment 100.

[0031] The tracking system 102 can be located in a playing venue 106. For example, the playing venue 106 can be configured to host a sporting event that includes one or more players 112. The tracking system 102 can be configured to record the motion of all players (i.e., athletes) on the playing field, as well as the motion of one or more other relevant objects (e.g., a ball, referees, etc.). In some implementations, the tracking system 102 can be an optical-based system, for example, using multiple fixed cameras. For example, a system with six fixed, calibrated cameras can be used that can project the three-dimensional positions of the athletes and the ball onto a two-dimensional overhead view of the playing field. In some implementations, the tracking system 102 can be a radio-based system, for example, using radio frequency identification (RFID) tags worn by the players or embedded in objects to be tracked. In general, the tracking system 102 can be configured to sample and record at a high frame rate (e.g., 25 Hz). The tracking system 102 can be configured to store, for each frame in the game file 110, at least the player identity and position information (e.g., (x, y) position) of all players and objects on the playing field.

[0032] The game file 110 can be augmented with other game information corresponding to the captured one or more frames, such as but not limited to game event information (passes, shots, turnovers, etc.) and context information (current score, time remaining, etc.). In some implementations, the game file 110 can include one or more set piece plays during a game (e.g., in a soccer game). A set piece play can be defined as a portion of a game where play begins (or resumes) after a stoppage. Exemplary set piece plays can include a corner or free kick in a soccer game, a throw-in in a basketball game, a face-off in a lacrosse game, etc.

[0033] The tracking system 102 can be configured to communicate with the organization computing system 104 via the network 105. The organization computing system 104 can be configured to manage and analyze data captured by the tracking system 102. The organization computing system 104 can include at least a web client application server 114, a pre-processing engine 116, a data store 118, and a classification agent 120. Each of the pre-processing engine 116 and the classification agent 120 can include one or more software modules. The one or more software modules can be a set of code or instructions stored on a medium (e.g., a memory of the organization computing system 104) representing a series of machine instructions (e.g., program code) that implement one or more algorithmic steps. Such machine instructions can be actual computer code that is interpreted by a processor of the organization computing system 104 to implement the instruction, or alternatively, can be a higher-level encoding of instructions that are interpreted to obtain the actual computer code. The one or more software modules can also include one or more hardware components. One or more aspects of the example algorithms can be performed by the hardware components (e.g., circuitry) themselves, rather than as a result of instructions.

[0034] The classification agent 120 can be configured to analyze one or more frames of a game and classify defensive formations of a particular team in the game. For example, the classification agent 120 can utilize an attribute-driven positioning attack analysis method that can utilize a hybrid deep learning approach to detect complex attributes, such as defensive marking schemes and manually created features, to enable interpretability. The classification agent 120 can include a machine learning module 128. The machine learning module 128 can include a CNN 126 and one or more pre-defined or manually created features 130.

[0035] The CNN 126 can be configured to analyze one or more tracking datasets corresponding to positioning attacks during a game and extract both offensive and defensive structures of a team. In some implementations, the one or more tracking datasets provided to the CNN 126 can have an image-based representation. For example, the CNN 126 can be trained to extract one or more defensive structures of a given team based on a training dataset.

[0036] The machine learning module 128 can be configured to classify one or more extracted features from one or more tracking datasets based in part on one or more predefined or manually created features 130. For example, the machine learning module 128 can be trained to classify offensive structures from one or more extracted features of a positioned offense based on one or more manually created features 130 defined by an end user. The machine learning module 128 can also be trained to classify defensive structures based on one or more extracted features of a positioned offense. For example, the machine learning module 128 can include one or more instructions for training a predictive model used by the classification agent 120. To train the predictive model, the machine learning module 128 can receive one or more data streams from the data store 118 as input. The one or more streams of user activity can include one or more extracted positioned offense features from one or more games. In some implementations, the machine learning module 128 can also receive the one or more manually created features 130 as input. The machine learning module 128 can implement one or more machine learning algorithms to train the predictive model. For example, the machine learning module 128 can use one or more of the following: a decision tree learning model, an association rule learning model, an artificial neural network model, a deep learning model, an inductive logic programming model, a support vector machine model, a clustering pattern, a Bayesian network model, a reinforcement learning model, a prototypical learning model, a similarity and metric learning model, a rule-based machine learning model, and the like.

[0037] Accordingly, the classification agent 120 can implement a combination of the CNN 126 that learns one or more features to extract and the machine learning module 128 that learns to perform a classification task from relevant features. In this way, the classification agent 120 is able to provide a comprehensive overview of the methods taken by an offensive team in a particular positioned offense and the methods taken by a defensive team.

[0038] The preprocessing agent 116 can be configured to process data retrieved from the data store 118 prior to input to the classification agent 120. For example, the preprocessing agent 116 can be configured to parse the one or more streams of tracking data to identify one or more positioned offenses contained therein. The preprocessing agent 116 can then extract one or more portions of the tracking data from the one or more streams of data that include one or more positioned offenses and normalize the one or more portions containing the positioned offenses for processing.

[0039] The data store 118 can be configured to store one or more match files 122. Each match file 122 can be captured and generated by the tracking system 102. In some embodiments, each match file of the one or more match files 122 can include all raw data captured from a particular match or event. For example, the raw data captured from a particular match or event can include each localization attack event in each match.

[0040] The client device 108 can communicate with the organization computing system 104 via the network 105. The client device 108 can be operated by a user. For example, the client device 108 can be a mobile device, a tablet computer, a desktop computer, or any computing system having the capabilities described herein. The user can include, but is not limited to, an individual, such as a subscriber, a client, a prospective client, or a customer associated with the organization computing system 104, such as an individual who has obtained, will obtain, or is likely to obtain a product, service, or consultation from an entity associated with the organization computing system 104.

[0041] The client device 108 can include at least an application 132. The application 132 can represent a web browser that allows access to a website or a standalone application. The client device 108 can access the application 132 to access one or more functions of the organization computing system 104. The client device 108 can communicate over the network 105 to request a web page, for example, from the web client application server 114 of the organization computing system 104. For example, the client device 108 can be configured to execute the application 132 to access content managed by the web client application server 114. The content displayed to the client device 108 can be sent from the web client application server 114 to the client device 108 and subsequently processed by the application 132 for display through a graphical user interface (GUI) of the client device 108.

[0042] Figure 2 is a block diagram illustrating a localization attack grammar model 200 that can define one or more manually created features 130 of the machine learning module 128 in accordance with example embodiments. Generally, the information used for analysis can include at least the x, y coordinates of the players and the ball. In some embodiments, the information used for analysis can also include one or more event labels, such as but not limited to, a header, a shot, a clearance, and the like.

[0043] To capture the evolving situation of a positional attack, an end user can define a positional attack grammar model 200 (hereinafter "model 200") based on domain expertise. The model 200 can define a temporal ordering of events during a given positional attack, which can allow for robust recognition in positional attacks with a variable number of phases. For example, a corner kick should start with an in-swinging cross that can be flicked-on at the near post to a second attacker to score. This example has a clear beginning, middle, and end. However, in another example, an arbitrary short pass of the ball by an attacker can be cross into the box, then cleared by a defender, then regained by an attacker, then back into the box to be claimed by the goalkeeper and fall to an attacker's feet for a score. The second example is much more complex, with three potential points at which the positional attack can be considered complete (e.g., initial clearance, goalkeeper claim, or goal). The model 200 attempts to define a grammar that enables multiple phases to occur.

[0044] The model 200 can be configured to capture the start and end of a positional attack, as well as specific types of events that can occur between the start and end of the positional attack. Such a model can free the machine learning module 128 from looking only at the first touch after a positional attack delivery, and capture which teams can produce a goal via a flick-on (e.g., a single pass) and a second touch that can be considered a rebound. The machine learning module 128 can also be able to measure which defensive teams can turn into counterattacks.

[0045] As shown, each set play 202 can start with some type of delivery 204. Example delivery types can be an in-swing, an out-swing, and a short pass (e.g., one or more passes). One possible path from 204 can be a shot 206. The shot 206 can result in a goal 208. Another possible path from 204 can be a flick-on 210. In some embodiments, the flick-on 210 can result in a goal 212. In some embodiments, the flick-on can result in a turnover 214. For example, the turnover can result in a counterattack (e.g., the attacking team becomes the defending team, and the defending team becomes the attacking team) or a build-up.

[0046] Another possible path from 204 can be that the serve type 204 is covered 220. One possible path from covered 220 can be that the ball goes out of bounds / foul 222 (e.g., a defensive player is fouled or the ball goes out of bounds). Another possible path from covered 220 can be a second touch 222. For example, an offensive player can shoot the ball at the goalkeeper; the goalkeeper successfully blocks the shot; and that offensive player (or another offensive player) regains possession via a rebound. In some implementations, after the second touch 224, the model 200 can resume as the serve type 204 (i.e., another pass between offensive players).

[0047] The machine learning module 128 can be trained to classify offensive formations using one or more manually created features 130 defined by the model 200. For example, the machine learning module 128 can be trained to perform predictions using manually created features when analyzing tracking data corresponding to a positional attack. Fine-grained information (e.g., the location of a serve) can also be incorporated into the machine learning process once the grammar is laid out for the machine learning module 128.

[0048] Figures 3A to 3C One or more graphical representations of a defensive set-up for a positional attack are illustrated in accordance with example implementations. Figure 3A A graphical representation 300 of man-to-man coverage during a positional attack. Figure 3B A graphical representation of a mixed method of coverage (e.g., a mix of man-to-man and zonal coverage) during a positional attack. Figure 3C A graphical representation of a zonal marking method of coverage during a positional attack.

[0049] As discussed earlier, even though conventional methods can be insufficient to capture such defensive behavior, the fine-grained spatial, variable number of players, and different patterns of play (short and long passes) are amenable to image-based representations. As such, the classification agent 120 can leverage the CNN 126 to learn one or more predictive features from one or more sets of tracking data corresponding to one or more positional attacks, e.g., by identifying local edge patterns and / or interactions and combining them in a hierarchical manner to better represent the tracking data. Thus, the CNN 126 can be trained to extract such one or more predictive features from positional attack information.

[0050] Figure 4 is a flowchart illustrating a method 400 of training a CNN 126 to classify defensive set-ups for positional attacks in accordance with example implementations. The method 400 can begin at step 402.

[0051] At step 402, the pre-processing engine 116 can receive one or more match data sets from the data store 118. For example, the pre-processing engine 116 can retrieve one or more tracking data sets to generate a training set for the CNN 126. Such match data can include spatial event data capturing each touch of the ball, as well as x, y coordinates and a timestamp.

[0052] At step 404, the pre-processing engine 116 can parse the one or more match data sets for one or more positioning attacking actions contained in the one or more match data sets. In other words, the pre-processing engine 116 can be configured to filter the one or more match data sets to identify those portions of the match data that include positioning attacking actions. Such positioning attacking actions can include, for example, a corner kick or a free kick.

[0053] At step 406, the pre-processing engine 116 can extract one or more portions from the one or more match data sets that contain player movement before and after a positioning attacking action. The pre-processing engine 116 can extract one or more portions from the one or more match data sets that contain player movement two seconds before a corner kick and player movement two seconds after a corner kick.

[0054] At step 408, the CNN 126 can learn one or more unique features that define various particular defensive arrangements (e.g., man-to-man, mixed man-to-man, zonal). In some embodiments, the CNN 126 can also learn one or more unique features that define particular defensive player positions, such as whether the defending team has a player near the post, the post / area, the near post, the front area, the near post area, both posts, etc. For example, the CNN 126 can undergo a backpropagation process using the tracking data such that the CNN 126 can learn one or more unique features of various defensive arrangements. In some embodiments, the architecture of the CNN 126 includes two hidden layers. For example, the CNN 126 can have a first hidden layer with 80 neurons and a second hidden layer with 50 neurons. Although the foregoing example discusses hidden layers with 80 neurons and 50 neurons, respectively, one of skill in the art will readily appreciate that this discussion is not to be limited to the above example. For example, a linear rectifier function (“ReLu”) activation can be applied at each layer with a softmax logit function used in conjunction with the final label prediction. Although ReLu activation can be used in the foregoing example, one of skill in the art can readily appreciate that other forms of activation can be used. By utilizing tracking data, the CNN 126 is able to process a variable number of players and can also capture interactions of players in proximity to one another without the need to manually construct distances and angles for each frame.

[0055] Figure 5 FIG. 13 is a flowchart illustrating a method 500 of classifying defensive formations for a set piece according to example embodiments. The method 500 can begin at step 502.

[0056] At step 502, the classification agent 120 can receive tracking data related to a match. In some embodiments, the tracking data can be captured by the tracking system 102. In some embodiments, the image stream can be live (or near live) data or historical data. In some embodiments, the tracking data can have an image-like representation. In some embodiments, the tracking data can comprise an image stream, or can be a video feed stored in the data store 118.

[0057] At step 504, the pre-processing engine 116 can parse the tracking data for one or more set piece actions contained in the tracking data. In other words, the pre-processing engine 116 can be configured to filter the tracking data to identify those portions of the match data that include set piece actions. Such set piece actions can include, for example, a corner kick or a free kick, a throw-in, a penalty kick, a snap, a faceoff, etc.

[0058] At step 506, the pre-processing engine 116 can extract one or more portions of the tracking data corresponding to player movements before and after the set piece occurs. For example, the pre-processing engine 116 can extract one or more portions of the tracking data that include player movements two seconds before and two seconds after a corner kick.

[0059] At step 508, the CNN 126 can extract one or more features of the attacking and defending teams in the set piece based on unique identifiers learned during training. For example, the CNN 126 can identify the defending team in the tracking data and extract one or more unique identifiers contained therein. Such unique identifiers can correspond to man-to-man, mixed man marking, and zonal marking. The CNN 126 can also identify whether the defending team has a defensive player near the back post, back post / zone, front post, front zone, front post zone, both posts, etc. In another example, the CNN 126 can identify the attacking team in the normalized tracking data and extract one or more unique identifiers contained therein corresponding to the attacking team.

[0060] Figure 6 FIG. 14 is a flowchart illustrating a method 600 of classifying attacking and defensive formations for a set piece according to example embodiments. The method 600 can begin at step 602.

[0061] At step 602, the classification agent 120 can receive tracking data related to a match. In some embodiments, the tracking data can have an image-like representation. In some embodiments, the tracking data can comprise a stream of images, or can be a video feed taken by the tracking system 102. For example, the stream of images can be a video feed stored in the data store 118. Typically, the tracking data can be live (or near live) data or historical data.

[0062] At step 604, the pre-processing engine 116 can parse the tracking data to identify one or more set-piece actions contained therein. In other words, the pre-processing engine 116 can be configured to filter the tracking data to identify those parts of the tracking data that include a set-piece action. Such set-piece actions can include, for example, a corner kick or a free kick.

[0063] At step 606, the classification module 120 can extract one or more features associated with the defensive formation and the attacking formation from one or more parts of the tracking data that contain a set-piece. For example, the pre-processing engine 116 can extract one or more parts from the tracking data that include the movement of players before and after the set-piece occurs. The CNN 126 can extract one or more features corresponding to the defensive side and one or more features corresponding to the attacking side from the one or more set-piece actions based on the unique identifiers learned during the training process. For example, the CNN 126 can identify the defensive team in the tracking data, locate one or more unique identifiers contained therein, and extract one or more features from the tracking data corresponding to the one or more unique identifiers.

[0064] At step 608, the classification module 120 can identify the attacking formation and the defensive formation based on the one or more extracted features. For example, the machine learning module 128 can identify the attacking formation based on the manually created features 130 defined in the Figure 2 machine learning module 128 can also identify the defensive lineup based on the training performed in conjunction with the Figure 4 above.

[0065] At step 610, the classification module 120 can determine whether each set-piece action was successful. In other words, the classification module 120 can determine whether the attacking side scored (i.e., the attacking side was successful) or whether the defensive side prevented the attacking side from scoring (i.e., the defensive side was successful). For example, the classification module 120 can analyze one or more image frames after the set-piece to determine whether a goal has been scored.

[0066] At step 612, the classification module 120 can generate a graphical representation that illustrates the effectiveness of various defensive and offensive styles across positional attacking actions. For example, the classification module 120 can generate the following Hinton diagram in connection with Figure 7A the Hinton diagram discussed above.

[0067] Figure 7A is a block diagram illustrating a Hinton diagram that illustrates offensive and defensive positional attacking styles for an example team in accordance with an example embodiment. To illustrate the benefits of the methods described above, the offensive and defensive behavior of all teams in the 2016 / 17 Premier League season was analyzed. Table 1A below illustrates the ranking of all teams in terms of their offensive efficiency. Table IB illustrates the ranking of all teams in terms of their defensive efficiency.

[0068] Table 1

[0069]

[0070]

[0071] Table 2

[0072]

[0073]

[0074] While it is useful to measure which teams can score or prevent positional attacking goals effectively, this information alone is not enough for a coach to take action when preparing for the next match.

[0075] The Hinton diagram illustrates the average offensive and defensive positional attacking styles of each team in the 2016-17 Premier League season and visualizes their effectiveness at both scoring and preventing goals. The Hinton diagram can illustrate that different teams have different strategies. For example, Chelsea and West Bromwich Albion both scored 16 goals from positional attacking (excluding direct free kicks). However, the methods they used to achieve this goal were very different. West Bromwich Albion's corner style was highly predictable, with 73% of their corners using an inswinger, whereas Chelsea used a mix of inswingers, outswingers and short corners to make it more difficult to predict.

[0076] Figure 7B and Figure 7C are block diagrams illustrating one or more graphical elements for offensive style 730 and defensive style 735 in accordance with example embodiments.

[0077] One of the problems with scouting an opponent for an upcoming game is selecting the most relevant games to analyze. The last game played can be a single example, but is not necessarily the most relevant, as injuries, personnel changes, home versus away status, relegation / promotion, and time since the last game can affect the "relevance" of a given game in complex ways. Thus, finding enough examples can be very time consuming.

[0078] The classification agent 120 can enable a team to scout an opponent to more effectively analyze their opponent. For example, using the methods discussed above to classify attacking formations and defensive formations, the classification agent 120 can also determine whether a team has a unique attacking or defensive style by a method called affinity propagation that clusters average team behavior.

[0079] For example, as shown in FIGS. 730 and 735, the classification agent 120 can identify clusters of teams that have similar attacking or defensive styles. Figure 7B and Figure 7C As shown in FIG. 730, six attacking style clusters are found in graph 730, and four defensive style clusters are found in graph 735. For graph 730, the classification agent 120 finds three unique teams (West Bromwich Albion (squares), Manchester City (circles), and Bournemouth (triangles). The classification agent 120 also identifies a cluster of two teams (Southampton (diamonds) and Watford), as well as two larger clusters that represent teams that differ in their service (stars) and teams that focus on inside and outside corner kicks (crosses). For graph 735, the classification agent 120 can identify four cluster types that indicate that teams' defensive arrangements are more predictable than their attacking styles.

[0080] Affinity propagation is an unsupervised clustering technique that works by finding one or more exemplars (which can be thought of as centroids). Then, Euclidean squared distances can be computed between the exemplars and all other teams to determine how similar teams are to each other. The team that best represents the cluster can be identified as the final exemplar. Using this technique, an analyst can quickly see whether an opponent has played a team within their cluster and see how that team has performed against that style in both attack and defense.

[0081] Figure 8AA system bus computing system architecture 800 is illustrated in accordance with an example embodiment. System 800 can represent at least a portion of organization computing system 104. One or more components of system 800 can be in electrical communication with each other using bus 805. System 800 can include a processing unit (CPU or processor) 810 and a system bus 805 that couples various system components including the system memory 815, such as read-only memory (ROM) 820 and random access memory (RAM) 825, to the processor 810. System 800 can include a cache of high-speed memory directly coupled to, in close proximity to, or integrated as part of the processor 810. System 800 can copy data from memory 815 and / or the storage device 830 to cache 812 for quick access by the processor 810. In this way, the high-speed cache 812 can provide a performance boost that avoids the latency

[0082] To enable user interaction with the computing device 800, an input device 845 can represent any number of input mechanisms, such as a microphone for speech, a touch-sensitive screen for gesture or graphical input, keyboard, mouse, motion input, speech and the like. An output device 835 can also be one or more of a number of output mechanisms known to those of skill in the art. In some instances, multimodal systems can enable a user to provide multiple types of input to communicate with the computing device 800. The communications interface 840 can generally govern and manage the

[0083] Storage 830 can be a non-transitory memory and can be a hard disk or other types of computer readable media which can store data which is accessible by a computer, such as magnetic cassettes, FLASH cards, solid state memory devices, digital versatile disks, cartridges, random access memories (RAMs) 825, read only memories (ROMs) 820, and hybrids thereof.

[0084] Storage 830 can include services 832, 834, and 836 which control the processor 810. Other hardware or software modules are contemplated. Storage 830 can be connected to system bus 805. In one aspect, a hardware module that performs a particular function can include the software component, such as a piece of

[0085] Figure 8B A computing system 850 having a chipset architecture is illustrated and can represent at least a portion of the computing system 104. The computing system 850 can be an example of computer hardware, software, and firmware that can be used to implement the disclosed technology. The system 850 can include a processor 855, which is representative of any number of physically and / or logically distinct resources that are capable of executing software, firmware, and hardware configured to perform identified computing. The processor 855 can be in communication with a chipset 860, which can control input to and output from the processor 855. In this example, the chipset 860 outputs information to an output device 865, such as a display, and can read information from and write information to a storage device 870, which can include magnetic media and solid state media, for example. The chipset 860 can also read data from and write data to a RAM 875. A bridge 880 for interfacing with a variety of user interface components 885 can be provided to interface with the chipset 860. Such user interface components 885 can include a keyboard, a microphone, touch detection and processing circuitry, a pointing device, such as a mouse, and the like. In general, the input to the system 850 can come from any of a number of machine- generated and / or human-generated sources.

[0086] Chipset 860 can also interface with one or more communication interfaces 890 that can have different physical interfaces. Such communication interfaces can include an interface to a wired or wireless local area network, an interface to a wide area network such as the Internet, and an interface to a personal area network. Some applications of the methods disclosed herein include the receipt of sorted datasets over a physical interface, or the generation of the datasets by the machine itself from analysis of stored data in memory 870 or 875. Further, the machine can receive inputs from a user through user interface components 885 and execute appropriate functions, such as browsing functions, by interpreting these inputs using processor 855.

[0087] It is understood that the example systems 800 and 850 can have more than one processor 810, or be part of a group or cluster of computing devices networked together to provide greater processing capability.

[0088] While the foregoing is directed to implementations of the present disclosure, other and further implementations can be devised without departing from the basic scope of the disclosure. For example, aspects of the present disclosure can be implemented in hardware or software or a combination of hardware and software. One implementation described herein can be implemented as a program product stored on a computer readable storage medium that can direct a computing system to facilitate implementing aspects described herein. The program product defining functionally the present implementations (including aspects described herein) can be stored in a variety of computer readable storage media. Exemplary computer readable storage media include, but are not limited to: (i) non-writable storage media (e.g., read-only memory devices within a computer such as CD-ROM disks readable by a CD-ROM drive, flash memory, ROM chips or any type of solid-state non-volatile memory) on which information is permanently stored; and (ii) writable storage media (e.g., floppy disks within a diskette drive, hard-disk drive, or any type of solid-state random access memory) on which alterable information is stored. Such computer readable storage media when carrying computer readable instructions are embodiments of the present disclosure.

[0089] Those skilled in the art will realize that the foregoing examples are non-limiting and are illustrative of the many aspects of the present disclosure. It is intended that all such modifications, permutations, equivalents, and alternatives fall within the true spirit and scope of the present disclosure. It is intended that the scope of the present disclosure should include all changes, substitutions, variations, alternatives, and modifications that the skilled artisan might conceive of following review of the present specification and drawings. Accordingly, the appended claims are intended to embrace all such alterations, permutations, and equivalents.

Claims

1. A method for identifying a team's formation during a set piece attack, the method comprising the steps of: receiving, by a computing system, tracking data for a team, the tracking data associated with a plurality of games during a season; identifying, by the computing system, a plurality of located plays contained in the tracking data; For each of the plurality of positional plays, extracting, by the computing system via a convolutional neural network, one or more unique identifiers indicative of a defensive formation implemented by the first team and one or more unique identifiers indicative of an offensive formation implemented by the second team, and extracting, by the computing system, one or more features corresponding to the first team and one or more features corresponding to the second team from one or more set plays based on the unique identifier; identifying, by the computing system, an offensive formation and a defensive formation based on one or more characteristics corresponding to the first team and one or more characteristics corresponding to the second team; The computing system identifies, for each positioning attack, whether the positioning attack is successful; and The effectiveness of each defensive formation used by the team during the season is determined by the computing system.

2. The method according to claim 1, wherein The step of determining, by the computing system, the effectiveness of various defensive formations used by the team throughout the season comprises: Determine the success rate of each defensive formation.

3. The method according to claim 2, wherein: The steps of determining the success rate of each defensive formation include: The success rate of each defensive formation against a given offensive formation is determined.

4. The method according to claim 1, wherein The step of determining, by the computing system, the effectiveness of various defensive formations used by the team throughout the season comprises: Determine the defensive efficiency of each defensive formation.

5. A non-transitory computer-readable medium comprising one or more sequences of instructions that, when executed by a processor, cause a computing system to perform operations comprising: receiving, by the computing system, tracking data for a team, the tracking data associated with a plurality of games during a season; identifying, by the computing system, a plurality of located plays contained in the tracking data; For each of the plurality of positional plays, extracting, by the computing system via a convolutional neural network, one or more unique identifiers indicative of a defensive formation implemented by the first team and one or more unique identifiers indicative of an offensive formation implemented by the second team, and extracting, by the computing system, one or more features corresponding to the first team and one or more features corresponding to the second team from one or more set plays based on the unique identifier; identifying, by the computing system, an offensive formation and a defensive formation based on one or more characteristics corresponding to the first team and one or more characteristics corresponding to the second team; The computing system identifies, for each positioning attack, whether the positioning attack is successful; and The effectiveness of each defensive formation used by the team during the season is determined by the computing system.

6. The non-transitory computer-readable medium of claim 5, wherein: Determining, by the computing system, the effectiveness of various defensive formations used by the team across the season includes: Determine the success rate of each defensive formation.

7. The non-transitory computer-readable medium of claim 6, wherein: Determining the success rate of each defensive formation includes: The success rate of each defensive formation against a given offensive formation is determined.

8. The non-transitory computer-readable medium of claim 5, wherein: Determining, by the computing system, the effectiveness of various defensive formations used by the team across the season includes: Determine the defensive efficiency of each defensive formation.

9. A system for identifying a defensive formation and an offensive formation in a positioning attack, the system comprising: processor; as well as a memory having stored thereon programming instructions, the programming instructions causing the system to perform operations when executed by the processor, the operations comprising: receiving tracking data for a team, the tracking data associated with a plurality of games during a season; identifying a plurality of positioning offenses contained in the tracking data; For each of the plurality of positional plays, extracting, via a convolutional neural network, one or more unique identifiers indicative of a defensive formation implemented by the first team and one or more unique identifiers indicative of an offensive formation implemented by the second team; extracting one or more features corresponding to the first team and one or more features corresponding to the second team from one or more set plays based on the unique identifier; identifying an offensive formation and a defensive formation based on one or more characteristics corresponding to the first team and one or more characteristics corresponding to the second team; For each positioning attack, identifying whether the positioning attack is successful; as well as The effectiveness of various defensive formations used by the team across the season is determined.

10. The system according to claim 9, wherein: The operation of determining the effectiveness of various defensive formations used by the team across the season includes: Determine the success rate of each defensive formation.

11. The system according to claim 10, wherein: The operation of determining the success rate of each defensive formation includes: The success rate of each defensive formation against a given offensive formation is determined.

Images