JP2023044410A - Tactical analyzer and method for controlling the same, and control program - Google Patents

Abstract

To achieve a tactical analyzer that can predict a player's next tactics during a game.SOLUTION: A tactical analyzer (10) for analyzing a player's tactics in a game includes a tactics acquisition unit (22) that acquires the player's tactics history in games from images of the games, and a strategy estimation unit (23) that estimates the player's game strategy using a learning model including a machine-learned strategy, which is a chronological pattern of tactics, from the player's tactics history in the past games, and the tactics history acquired by the tactics acquisition unit (22).SELECTED DRAWING: Figure 1

Description

The present invention relates to a tactical analysis device that analyzes a player's tactics in a game, a control method thereof, and a control program.

2. Description of the Related Art In recent years, computer-based analysis has been actively performed even in the field of competitive competitions such as sports. For example, Patent Literature 1 discloses an attack pattern extraction device that analyzes attack patterns based on player position information and play histories in team sports competitions.

JP 2017-209237 A JP 2020-185370 A

However, Patent Document 1 merely identifies past tactics (attack patterns), and does not predict future tactics from the player's tactical history during the game. Also, in baseball broadcasts, commentators sometimes predict future pitching based on the history of pitching during the game. It is not done intentionally.

An object of one aspect of the present invention is to realize a tactical analysis device or the like that can predict future tactics of a player during a game.

In order to solve the above-described problems, a tactical analysis device according to one aspect of the present invention is a tactical analysis device that analyzes a player's tactics in a game, and analyzes the player's tactical history in the game from a video of the game. a tactics acquisition unit to acquire; a learning model of the player in which a strategy, which is a time-series pattern of tactics, is machine-learned using the tactics history of the player in the past competition; and the tactics acquired by the tactics acquisition unit and a strategy estimator for estimating the player's strategy for the game using the history.

Here, the tactics refer to characteristic action contents of the player in a certain action unit in the game. For example, in the case of baseball, the action unit may be one pitching, and the characteristic action may be at least one of the pitching course, type of pitch, and speed of the pitch. In the case of table tennis, the action unit may be one rally, and the characteristic actions include the rally trajectory in the rally, the style of the serve, the position and rotation of the ball in the serve, and the player's behavior. It may be at least one of position and pose.

According to the above configuration, the player's tactical history in the game is acquired from the video of the game, and the player's strategy in the game is estimated using the acquired tactical history and the player's learning model. Therefore, the user can refer to the strategy estimated from the video during the game, that is, the time-series pattern of the tactics, and thereby predict the future tactics of the player during the game. The user may be the player's opponent or the opponent's coach.

In the tactical analysis device according to this aspect, the tactical acquisition unit extracts time-series data of representative points related to tactics from the video, and performs singular value decomposition on the extracted time-series data to acquire the tactics of the player. You may In this case, even if the time lengths of the time-series data of the representative points are different, the player's tactics can be acquired.

In the tactical analysis device according to this aspect, the tactical acquisition unit extracts the trajectory of the moving body from the video of the game, acquires the tactics of the player based on the extracted trajectory of the moving body, and repeats this process. Thus, the player's tactical history may be obtained.

Here, the moving object is an object that moves in the game, and includes, for example, a player in the game, a ball used in the game, and the like. In addition, as preprocessing, the tactic acquisition unit may remove scenes unrelated to acquisition of the tactics (for example, close-up video of the player) from the video of the game as unnecessary scenes.

In the tactical analysis device according to this aspect, the moving object includes the player, and the tactical acquisition unit estimates the trajectory of the player's skeleton by performing machine learning using the video of the game. At least part of the tactics of the player may be obtained based on the estimated skeletal trajectory. In this case, since the behavior of the player can be estimated with high accuracy, the tactics of the player can be estimated with high accuracy.

In the tactical analysis device according to this aspect, the learning model may be created by performing machine learning by integrally combining a plurality of models. In this case, the estimation accuracy of the strategy can be improved compared to other learning models.

In the tactical analysis device according to this aspect, the plurality of models may be fuzzy clustering models. In this case, the estimated strategy will be provided by fuzzy rules (IF THEN rules). Therefore, the user can easily predict future tactics.

In the tactical analysis device according to this aspect, the learning model generates virtual data based on estimated data estimated by one of the plurality of models subjected to machine learning, and combines the virtual data and learning data. Another model of the plurality of models may be created by being machine-learned using. In this case, the data for machine learning can be increased, so the strategy estimation accuracy can be further improved.

A control method for a tactical analysis device according to another aspect of the present invention is a control method for a tactical analysis device that analyzes the tactics of a player in a game, and acquires the tactical history of the player in the game from video of the game. A tactic acquisition step, a learning model in which a strategy, which is a time-series pattern of tactics, is machine-learned using the tactic history of the player in the past competition, the tactic history acquired in the tactic acquisition step, and a strategy estimation step of estimating the player's strategy for the game using .

According to the control method described above, the same effects as those of the above-described tactical analysis device can be obtained.

The tactical analysis device according to each aspect of the present invention may be realized by a computer. A control program for a tactical analysis device that is realized by a computer, and a computer-readable recording medium recording it are also included in the scope of the present invention.

According to one aspect of the present invention, future tactics of the players can be predicted during the competition.

1 is a block diagram showing a schematic configuration of a tactical analysis device according to one embodiment of the present invention; FIG. It is a flow chart which shows the flow of processing in the learning mode in the above-mentioned tactics analysis device. It is a flow chart which shows the flow of processing in competition mode in the above-mentioned tactics analysis device. FIG. 4 is a diagram showing superimposed black-and-white video converted from broadcast color video; It is a figure which shows an example of the image containing the tactics information which the tactics acquisition part in the said tactics analysis apparatus acquires. 3 is a block diagram showing a schematic configuration of a strategy estimation section and a tactical history storage section in the tactical analysis device; FIG. 4 is a flow chart showing a flow of learning processing in the strategy estimation unit; It is a figure which shows an example of the tactics log|history which the said tactics acquisition part acquired in tabular form. 10 is a flow chart showing the flow of processing for extracting a ball from video in a tactical analysis device according to another embodiment of the present invention. FIG. 10 is a diagram showing an example of a frame image processed by a white blog extraction process in the above process; FIG. 10 is a diagram showing an example of a frame image processed by the inter-frame difference detection process in the above process;

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail. For convenience of explanation, members having the same functions as members shown in each embodiment are denoted by the same reference numerals, and descriptions thereof are omitted as appropriate.

[Embodiment 1]
An embodiment of the present invention will be described with reference to FIGS. 1-8.

(Overview of tactical analysis equipment)
FIG. 1 is a block diagram showing a schematic configuration of a tactical analysis device according to this embodiment. The tactical analysis device 10 of this embodiment obtains and analyzes the tactics of the opponent (player) from the video, and estimates the strategy of the opponent. In addition, below, although the case where the said game is table tennis is demonstrated, it is not limited to this. Further, the details of the table tennis rules and the like are well known, so the explanation thereof will be omitted.

As shown in FIG. 1 , the tactical analysis device 10 has a control section 11 , a storage section 12 and a communication section 13 . The control unit 11 comprehensively controls operations of various components of the tactical analysis device 10, and is configured by a computer including a CPU (Central Processing Unit) and memory, for example. Operation control of various configurations is performed by causing a computer to execute a control program. The storage unit 12 records information, and is configured by a storage device such as a hard disk or flash memory. The communication unit 13 is configured by a communication device that transmits and receives information via a communication network such as the Internet.

The control unit 11 has a tactic acquisition unit 22 and a strategy estimation unit 23 . The storage unit 12 also includes a tactical history storage unit 31 that stores tactical history data. The present embodiment includes a learning mode TM in which the learning model of the strategy estimating section 23 undergoes machine learning, and a competition mode PM in which the machine-learned learning model is used during the competition.

The tactic acquisition unit 22 acquires tactic history data (time-series data of tactics) of the opponent from video data acquired via the communication unit 13 . The tactic acquisition unit 22 stores the acquired tactic history data in the tactic history storage unit 31 .

Specifically, the tactical acquisition unit 22 processes to remove unnecessary scenes from the video, and obtains the position and rotation of the ball (moving body), the position and rotation of the player (moving body), and the Posture, etc. are identified and used as tactical information. Next, the tactical acquisition unit 22 acquires this tactical information for each rally, and sets a plurality of pieces of tactical information in a plurality of rallies as a tactical history. Then, the tactic acquisition unit 22 stores the acquired tactic history data of the opponent in the tactic history storage unit 31 .

Note that the above video can also be a video for broadcasting. In this case, there is no need to use a special camera system. Further, the video may be a video captured by one camera, or may be a plurality of videos captured by a plurality of cameras.

The strategy estimating section 23 uses the tactical history data stored in the tactical history storage section 31 to estimate the opponent's strategy. Specifically, the strategy estimation unit 23 has a learning model, and in the learning mode TM, machine learning is performed on the learning model using the opponent's past tactical history. On the other hand, in the competition mode PM, the strategy estimating section 23 estimates the opponent's strategy using the opponent's tactical history during the competition and the machine-learned learning model. The strategy estimation unit 23 outputs the estimated strategy data to an external device such as a user's terminal via the communication unit 13 .

(processing in learning mode)
FIG. 2 is a flow chart showing the flow of processing in the learning mode TM in the tactical analysis device 10 configured as described above. The processing shown in FIG. 2 is executed for each player.

As shown in FIG. 2, first, the tactic acquisition unit 22 collects videos of past games of a certain player (S11). Next, the tactic acquisition unit 22 collects the player's past tactic history from the collected video and stores it in the tactic history storage unit 31 (S12).

Next, the strategy estimating unit 23 performs machine learning on the learning model using the past tactical history stored in the tactical history storage unit 31 (S13). The strategy estimating unit 23 estimates the player's strategy using a part of the past tactical history and a machine-learned learning model, and transmits data of the estimated strategy via the communication unit 13. You may output to an external device. After that, the processing in the learning mode TM ends.

FIG. 3 is a flow chart showing the flow of processing in the competition mode PM in the tactical analysis device 10 having the above configuration. The processing shown in FIG. 3 is executed during the competition.

As shown in FIG. 3, first, the tactic acquisition unit 22 acquires the image of the opponent (player) during the competition (S21). Next, the tactic acquisition unit 22 acquires the tactic history of the opponent during the game from the acquired video, and stores it in the tactic history storage unit 31 (S22, tactic acquisition step).

Next, the strategy estimating unit 23 uses the tactical history during the game stored in the tactical history storage unit 31 and the learning model of the opponent machine-learned in step S13 of FIG. (S23, strategy estimation step). Then, the strategy estimating unit 23 outputs data of the estimated strategy to the terminal (external device) of the opponent or the coach via the communication unit 13 (S24). After that, the processing in the competition mode PM ends.

Therefore, the tactical analysis device 10 of the present embodiment acquires the tactical history of the opponent in the competition from the video during the competition, and uses the acquired tactical history and the machine-learned learning model of the opponent. , to estimate the strategy of the opponent during the competition. By referring to the estimated strategy, that is, the time-series pattern of tactics, the opponent or his coach can predict future tactics of the opponent.

(Additional notes)
The strategy estimating unit 23 may have the learning model previously machine-learned. In this case, the learning mode can be omitted.

(Details of the tactical acquisition part)
Next, details of the tactic acquisition unit 22 and the strategy estimation unit 23 will be described.

First, the tactic acquisition unit 22 removes unnecessary scenes such as commercials, close-up images of competitors (players), slow-motion playback images, etc. from the 30 fps broadcast video, and divides the video into videos for each rally. Next, the tactical acquisition unit 22 converts the color video into black-and-white video.

FIG. 4 is a diagram showing the above black-and-white video superimposed. As can be understood from FIG. 4, the tactical acquisition unit 22 can identify the positions of moving objects, that is, the ball and the player, from the black-and-white video, and acquire the trajectories of the ball and the player. . In addition, the tactic acquisition unit 22 can acquire the player's posture from the black-and-white video. Also, the tactic acquisition unit 22 can acquire the coordinates of the four corners of the table tennis table from the black-and-white video. Also, the tactic acquisition unit 22 can specify the game number, the rally number, and the score immediately before the current rally from the upper left part of the black-and-white video.

Next, the tactic acquisition unit 22 performs affine transformation using the coordinates of the four corners of the ping-pong table, so that the trajectory of the ball and the player is converted from the coordinates on the image to the trajectories on the plane of the ping-pong table viewed from above. to the coordinates of Next, the tactic acquisition unit 22 identifies the position (landing point) where the ball lands on the table tennis table from the trajectory of the ball. A rally trajectory is acquired by connecting the identified positions with straight lines in chronological order.

Also, the tactic acquisition unit 22 identifies the style of serving and the rotation of the ball from the trajectory of the ball and the player, and the posture of the player. Examples of the above serve styles include forehands, backhands, hooks, and the like. Examples of ball rotations in the above serve include right lateral rotation and left lateral rotation, upward rotation and downward rotation, combinations thereof, knuckles (no rotation), and the like. The above-mentioned hook refers to a technique of throwing up a ball, standing a racket against the falling ball, and rubbing the ball with either side of the racket to serve.

Then, the tactical acquisition unit 22 acquires the rally trajectory, the style of the serve, and the ball rotation of the serve as tactical information for each rally, and sets a plurality of pieces of tactical information in a plurality of rallies as a tactical history. In the learning mode TM, the tactic acquisition unit 22 stores a plurality of tactic history data in a plurality of past matches in the tactic history storage unit 31 as learning data described later. On the other hand, in the competition mode PM, the data of the tactical history during the competition (for example, the tactical history of one game) is stored in the tactical history storage unit 31 as evaluation data, which will be described later. In the examples, these processes could be realized using Anaconda Python 3.5.2, OpenCV 3.1.0, easygui, and tqdm programs.

FIG. 5 is a diagram showing an example of an image containing the tactical information. As shown in FIG. 5, the left portion of the image includes a frame indicating the table tennis table and the rally trajectory, and the right center portion of the image includes a legend of the rally trajectory. In the above image, the rally trajectory of the serve is indicated by a solid line, the final rally trajectory is indicated by a two-dot chain line, and the rally trajectory between the rally trajectory of the serve and the last rally trajectory is indicated by a one-dot chain line. there is The ball rotation of the serve is included below the rally trajectory. In the example of FIG. 5, the ball spin of the serve is a Clockwise side back spin.

The upper right part of the above image contains the name of the server, the name of the receiver, the game number, and the rally number in order from the top. In the lower right part of the above image, the score before the rally, serve style, and ball rotation of serve are included in order from the top. there is

(Details of Strategy Estimation Department)
FIG. 6 is a block diagram showing a schematic configuration of the strategy estimation section 23 and the tactical history storage section 31. As shown in FIG. As shown in FIG. 6, the strategy estimation unit 23 includes a learning device group 41, a result acquisition unit 42, a virtual data generation unit 43, and a final determination unit 44. The tactic history storage unit 31 also includes a learning data storage unit 51 , an evaluation data storage unit 52 , and a virtual data storage unit 53 .

The learning data storage unit 51 stores past strategy history data as learning data TRD. The evaluation data storage unit 52 stores tactic history data during the game as evaluation data CHD.

In this embodiment, an ensemble learning fuzzy clustering model is used as the learning model. Therefore, the learning device group 41 includes n learning devices (n is an integer equal to or greater than 2) that perform machine learning. Each learner uses a fuzzy clustering model as a learning model (model). Fuzzy clustering classifies an input data set into classes using if-then rules, and estimates boundaries of the classes.

In the learning mode TM, the first learning device 41a performs learning using the set of learning data TRD from the learning data storage unit 51, and outputs an estimated class set as a result. The other learning device 41b performs learning using a part of the set of learning data TRD from the learning data storage unit 51 and the set of virtual data VD from the virtual data storage unit 53, and determines the estimated class. Output the set as the result.

In the competition mode PM, all learned learners 41a and 41b estimate classes from the set of evaluation data CHD from the evaluation data storage unit 52, and output the estimated class set as a result.

The result acquisition unit 42 acquires a set of classes estimated by the learners 41 a and 41 b of the learner group 41 . The result acquisition unit 42 sends the estimated set of classes to the virtual data generation unit 43 in the learning mode, and to the final determination unit 44 in the competition mode.

The virtual data generation unit 43 generates a set of virtual data VD using the set of learning data TRD from the learning data storage unit 51 and the set of estimated classes from the result acquisition unit 42. . The virtual data generation unit 43 stores the generated virtual data in the virtual data storage unit 53 .

Specifically, the virtual data generator 43 first compares the actual class included in the learning data TRD with the estimated class for each of the learning data TRD to identify whether the class is correct or not. do. Next, generating a set of virtual data VD based on at least part of the erroneously estimated learning data TRD (misestimated data) and the correctly estimated learning data TRD (positive estimated data); The generated virtual data is stored in the virtual data storage unit 53 .

For each of the evaluation data CHD, the final determination unit 44 finally determines, as the estimated class, the class with the largest number among the sets of classes estimated by the learning devices 41a and 41b. The final determination unit 44 transmits the set of classes estimated for the set of evaluation data CHD to an external device via the communication unit 13 .

~ (learning process)
FIG. 7 is a flow chart showing the flow of learning processing in the strategy estimating section 23 configured as described above. As shown in FIG. 7, first, a variable i is initialized to 1 (S31), and the i-th learning device (first learning device 41a) uses the set of learning data TRD from the learning data storage unit 51 to Learning is performed (S32).

Next, the virtual data generator 43 compares the class estimated by the i-th learner with the actual class for each of the learning data TRD, and determines whether the class is correct or not (S33). Next, if the correctly estimated ratio is equal to or greater than the threshold (YES in S34), the learning process ends.

On the other hand, if the ratio of correctly estimated data is less than the threshold (NO in S34), the virtual data generator 43 estimates the (i+1)-th low-estimated area on the multidimensional coordinate (space) containing the erroneously estimated data. Set as an area (S35). Next, the virtual data generation unit 43 generates a set of virtual data VD based on the set of erroneous estimation data and correct estimation data located in the (i+1)th low estimation region, and stores the set in the virtual data storage unit 53. (S36).

Next, the (i+1)-th learner (another learner 41b) uses the set of erroneously estimated data and correct estimated data located in the (i+1)-th low estimation area and the set of the virtual data VD. Learning is performed (S37).

Next, the variable i is incremented by 1 (S38). If the variable i is less than or equal to the predetermined number of times (NO in S39), the process returns to step S12 and repeats the above operations. Then, when the variable i exceeds the predetermined number of times (YES in S39), the learning process ends.

(Example)
FIG. 8 is a diagram showing an example of the tactical history acquired by the tactical acquisition unit 22 in tabular form. The tactical history shown in FIG. 8 includes tactical information when the opponent is the server and points are obtained in each of the first to fourth games. That is, the part where the tactical information is not described in FIG. 8 corresponds to the case where the opponent is the server and the opponent has obtained points. In addition, in the example of FIG. 8, the tactical information includes the hitting method of the serve (forehand, backhand, etc.), the rotation of the ball, and the type of attack.

The strategy estimation unit 23 performs learning processing using the tactical history shown in FIG. 8 as learning data TRD, and the following is an example of a set of classes estimated.

If Game Number is Small and Rally Number is Small, then (Forehand + Right Side Down) Serve is Large.
If Game Number is Small and Rally Number is Small, then (N or V) Attack is Large.
If Rally Number is Small, then (N or V) Attack is Large.
If Rally Number is Large, then (A or V) Attack is Large.
If Game Number is Large, then Serve Type is Large.
Next, it is assumed that the tactical acquisition unit 22 acquires the tactical history of the first game and the second game in FIG. 8 from the images of the first game and the second game. Then, it is assumed that the strategy estimation unit 23 has estimated the set of classes by using the tactical histories of the first game and the second game in FIG. 8 as a set of the evaluation data CHD.

In this case, at the end of the second game, if the opponent or his coach refers to the set of the above class, the opponent's strategy from the third game onwards will be predicted as follows (a) and (b). be able to.

(a) In the first half of the rally, they tend to use the N attack or V attack, and in the latter half of the rally, they tend to use the A attack or V attack.

(b) As the game progresses, the types of serves increase.

The above (a) is an intra-game strategy, and the above (b) is an inter-game strategy. In this way, since the intra-game strategy and the inter-game strategy can be predicted, the future strategy of the opponent in the main match can be predicted well.

Therefore, in the learning device group 41 of the tactical analysis device 10 of the present embodiment, machine learning is performed by integrally combining a plurality of learning devices 41a and 41b. Estimation accuracy can be improved. Also, since each of the plurality of learners 41a and 41b uses fuzzy clustering as a learning model, the estimated strategy is provided by fuzzy rules (IF THEN rules). Therefore, opponents or their coaches can easily anticipate future tactics.

Further, virtual data VD is generated based on estimated data estimated by the (i−1)th learner (i is an integer of 2 or more and n or less) subjected to machine learning, and the virtual data VD and the learning data TRD are generated. is used to machine-learn the i-th learner 41b. As a result, it is possible to increase the data for performing machine learning, so that the strategy estimation accuracy can be further improved.

[Embodiment 2]
Another embodiment of the invention is described with reference to FIGS. 9-11. The tactical analysis device 10 of this embodiment differs from the tactical analysis device 10 shown in FIGS. be.

FIG. 9 is a flow chart showing the flow of processing for extracting a ball from video in the tactical acquisition unit 22 of the tactical analysis device 10 of this embodiment. In this embodiment, the trajectory of the ball is estimated from the broadcast video automatically divided for each rally by the white video detection process (S40) for the white ball and the inter-frame difference detection process (S45).

As shown in FIG. 9, in the white image detection process (S40), first, a white portion image is extracted by performing threshold processing on each color scale of RGB (S41).

Next, the skeletal position of the athlete is estimated by CenterNet learning (S42). The input to CenterNet is a moving image in MP4 format, and the output from CenterNet is 17 points of two-dimensional skeleton position coordinates and two points of two-dimensional coordinates that specify the entire skeleton region. Therefore, for two players, the number of output variables is (17×2+2×2)×2=76. Since step S42 can accurately estimate the behavior of the player, it is possible to accurately estimate the tactics of the player.

Then, the white blog of the white ball is detected by deleting the white blog of the player's racket, taping, bib characters, shoes, etc. other than the white ball (S43).

FIG. 10 is a diagram showing an example of a frame image processed by the white blog extraction process (S40). Referring to FIG. 10, it can be understood that not only the skeleton of the player and the white ball, but also several white blog parts such as the white frame of the ping-pong table, the white part of the top of the net, and the white score frame are detected. .

On the other hand, in the inter-frame difference detection process (S45), as shown in FIG. 9, first, the difference in RGB color scale between the image of the current frame and the image seven frames before is calculated (S46). Next, the same process as step S42 of the white image detection process (S40) is performed (S47). Next, white blogs other than the skeleton region of the player are detected by deleting the white blogs of the player other than the white ball from the position coordinates of the player's skeleton estimated in step S47 (S48).

FIG. 11 is a diagram showing an example of a frame image processed by the inter-frame difference detection process (S45). Referring to FIG. 11, it can be understood that only the skeleton of the player and the white ball are detected, and other white blog parts such as the white frame of the ping-pong table, the white part of the top of the net, the white part of the scoring frame, etc. are deleted. . Therefore, the position of the white ball can be detected by the white image detection process (S40) and the inter-frame difference detection process (S45).

After the white image detection process (S40) and the inter-frame difference detection process (S45), the ball position of the next frame is estimated (S51). This estimation is limited to a white blog portion (ball candidate group) within a circular area with a radius of 80 pixels from the ball position in the current frame and within an inner range reduced by 10 pixels from the range of the ping-pong table. This is done by extracting the white blogs of If detection is not possible, the radius is expanded by 20 pixels for each frame, and the above estimation is performed. Then, using the Kalman filter, the bi-cube interpolation method, etc., the ball position is interpolated when the ball is hidden behind the player's body (occlusion), and the position coordinates of the ball trajectory are estimated (S52). Subsequent processing is the same as that of the tactical analysis device 10 shown in FIGS.

(Additional notes)
By the way, the trajectories of the ball and the players are indicated by time-series data of positions (representative points), but the length of the time-series data varies depending on the number of rallies, elapsed time, and the like. Therefore, the strategy acquisition unit 22 may acquire singular values calculated by singular value decomposition of the time-series data as strategy information for each rally. In this case, even if the time lengths of the time-series data are different, the tactics of the opponent can be obtained.

Also, the competition mode PM shown in FIG. 3 may be executed after the competition. In this case, the opponent's strategy in the current game can be estimated and analyzed, so that it can be utilized in the next game with the above opponent.

Also, in the above embodiment, the present invention is applied to a table tennis competition, but the present invention is not limited to this. The above embodiments can be applied to various competitions using balls, such as tennis, baseball, and soccer. In the case of baseball, the contents of pitching correspond to tactics. Also, in the case of soccer, tactics are determined not by each player but by the team to which the player belongs. Moreover, the above embodiments can also be applied to martial arts such as wrestling and judo. Also, in the above embodiment, the strategy of the opponent (enemy) is estimated, but the strategy of the opponent (ally) may be estimated.

[Example of realization by software]
The function of the tactical analysis device 10 (hereinafter referred to as "device") is a program for causing a computer to function as the device, and the computer is used as each control block (especially each part included in the control unit 11) of the device. It can be realized by a program for functioning.

In this case, the apparatus comprises a computer having at least one control device (eg processor) and at least one storage device (eg memory) as hardware for executing the program. Each function described in each of the above embodiments is realized by executing the above program using the control device and the storage device.

The program may be recorded on one or more computer-readable recording media, not temporary. The recording medium may or may not be included in the device. In the latter case, the program may be supplied to the device via any transmission medium, wired or wireless.

Also, part or all of the functions of the above control blocks can be realized by logic circuits. For example, integrated circuits in which logic circuits functioning as the control blocks described above are formed are also included in the scope of the present invention. In addition, it is also possible to implement the functions of the control blocks described above by, for example, a quantum computer.

The present invention is not limited to the above-described embodiments, but can be modified in various ways within the scope of the claims, and can be obtained by appropriately combining technical means disclosed in different embodiments. is also included in the technical scope of the present invention.

According to aspects of the present invention, for example, the following becomes possible. (1) From the broadcast video, it is possible to analyze the opponent's past data and predict the opponent's next strategy, so the opponent can recognize what kind of attack and defense should be done. (2) the results of the first game can be analyzed immediately; Therefore, the opponent's coach can give advice to the opponent during the second game as to what kind of attacks should be made in the second game. (3) Since the strategy of the top players can be referred to, it can also be used for player development.

10 Tactical analysis device 11 Control unit 12 Storage unit 13 Communication unit 22 Tactics acquisition unit 23 Strategy estimation unit 31 Tactics history storage unit 41 Learning device group 41a, 41b Learning device 42 Result acquisition unit 43 Virtual data generation unit 44 Final decision unit 51 Learning evaluation data storage unit 52 evaluation data storage unit 53 virtual data storage unit

Claims (9)

A tactical analysis device for analyzing a player's tactics in a competition,
a tactic acquisition unit that acquires the tactical history of the player in the competition from the video of the competition;
Using the player's learning model in which a strategy, which is a time-series pattern of tactics, is machine-learned using the player's tactical history in the past competition, and the tactical history acquired by the tactical acquisition unit, and a strategy estimator for estimating the player's strategy for the game. 2. The tactic according to claim 1, wherein said tactic acquisition unit acquires said player's tactic by extracting time-series data of representative points related to tactics from said video and performing singular value decomposition on said extracted time-series data. Analysis equipment. The tactic acquisition unit extracts the trajectory of the moving object from the video of the game, acquires the tactic of the player based on the extracted trajectory of the moving object, and repeats this to obtain the tactic history of the player. 3. A tactical analysis device according to claim 1 or 2 for obtaining. the moving object includes the player,
The tactic acquisition unit estimates a trajectory of the player's skeleton by performing machine learning using the video of the game, and acquires at least part of the tactic of the player based on the estimated trajectory of the skeleton. 4. The tactical analyzer of claim 3, wherein the tactical analyzer does. The tactical analysis device according to any one of claims 1 to 4, wherein the learning model is created by performing machine learning by integrally combining a plurality of models. 6. The tactical analyzer of claim 5, wherein said plurality of models are fuzzy clustering models. In the learning model, virtual data is generated based on estimated data estimated by a certain model of the plurality of models subjected to machine learning, and the virtual data and the learning data are used to generate another model of the plurality of models. 7. The tactical analysis device according to claim 5 or 6, wherein the model of is created by machine learning. A control method for a tactical analysis device that analyzes a player's tactics in a game, comprising:
a tactical acquisition step of acquiring the tactical history of the player in the competition from the video of the competition;
Using a learning model in which a strategy, which is a time-series pattern of tactics, is machine-learned using the player's tactical history in the past competition, and the tactical history acquired in the tactical acquisition step, and a strategy estimation step of estimating a strategy of said player in a game. A control program for causing a computer to function as the tactical analysis device according to claim 1, the control program for causing the computer to function as the tactical acquisition unit and the strategy estimation unit.

Images