KR102045567B1 - Apparatus and method for recommendation of curling game strategy using deep learning - Google Patents

Abstract

It provides a device and method for recommending curling game strategy, the curling field strategy recommending device includes a memory for storing a curling game strategy analysis and recommendation program applying deep learning and Monte Carlo tree navigation techniques, and a processor for executing a program stored in the memory, The processor recognizes preset game situation information by recognizing an image of an ongoing curling game according to the execution of the curling game strategy analysis and recommendation program, and collects the recorded data of the plurality of curling events. Learning data and the game situation information through deep learning and Monte Carlo tree search techniques to build a strategy recommendation model, and recommending information predicting one or more strategies corresponding to the recognized game situation information through the strategy recommendation model. .

Description

Apparatus and method for recommending curling game strategy based on deep learning {APPARATUS AND METHOD FOR RECOMMENDATION OF CURLING GAME STRATEGY USING DEEP LEARNING}

The present invention relates to an apparatus and method for recommending a strategy in a curling race, and more particularly, to an apparatus and method for recommending an optimal curling pitching strategy based on artificial intelligence technology.

Curling is a game where a player scores a goal by sliding a flat stone (ie curling stone) on the ice using a Broom. Curling consists of two teams of four, with each team alternating, sliding the stone towards a circular target called the House within a 42.1m long and 4.3m wide link. At this time, the two sweepers use the timing mechanism and the judgment according to the experience, and move along the path of the stone to adjust the course of the stone using the brom so that the stone stops as close as possible to the target point. Wipe the center line. Based on this, the score is calculated based on how close the stone is to the house at the end. The stone weighs 20kg or less, circumference 91.4cm or less and 11.4cm in height. 16 games are played in one end, and 10 or 12 games are played in one game.

In general, curling is sometimes referred to as `` chess on ice, '' because curling events require a player to take very complex strategic thoughts to locate and route stones. In a curling race, a strategy is to decide what shot to use, and a pitching strategy that takes into account the speed, direction of rotation, and direction of the stone is required. In order to reduce the friction between the stone and the floor, and the amount of rotation of the stone, A sweeping strategy is also needed to clean the ice in front of the path. The slower the stone, the more it bends, so sweeping can increase the distance the stone travels and straighten the path. In addition, it is important that the maneuver be made in accordance with the ever-changing arena conditions throughout the game, such as the ice conditions of the curling seat (ie the ice face of the game) and the location of the opponent stone. That is, in curling, it is important to accurately recognize the state of each stone and to determine an optimal pitch position. However, in curling, the pitching position is almost infinite, so it is difficult to select the optimal pitching position and method.

Conventionally, although a person analyzes a curling game and establishes a game strategy, a means of providing an interface for recording curling game information in real time and allowing a person to analyze or presenting statistical data has been used. The conventional curling game analysis method has a high possibility of a strategy error due to human factors.

In addition, the curling competition strategy algorithm has to consider the limited time of the curling competition in addition to the uncertainty of curling due to the friction of the ice, the stone position of the opponent, etc., it is difficult to determine the optimal strategy in a short time.

Korean Patent No. 10-1478097 (name of the invention: a mobile terminal for performing a curling simulation, a curling simulation method and a recording medium) Korean Patent No. 10-1499570 (name of the invention: a mobile terminal for curling analysis and a curling match analysis system using the same) Korea Patent Registration No. 10-1611431 (name of the invention: curling analysis method, recording medium and apparatus for performing the same)

One embodiment of the present invention is to provide an apparatus and method for recommending an optimal strategy for a curling game based on deep learning and Monte Carlo tree search techniques.

However, the technical problem to be achieved by the present embodiment is not limited to the technical problem as described above, and other technical problems may exist.

According to an aspect of the present invention, there is provided a curling match strategy recommending apparatus comprising: a memory storing a curling match strategy analysis and recommendation program applied to a deep learning and Monte Carlo tree search technique; And a processor for executing a program stored in the memory, wherein the processor recognizes preset game situation information by recognizing an image of an image of an ongoing curling game according to the execution of the curling game strategy analysis and recommendation program. Learning the data collected from the plurality of recorded curling matches and the game situation information through deep learning and Monte Carlo tree search techniques to construct a strategy recommendation model, and recognizing the recognition through the strategy recommendation model. Recommend information for predicting one or more strategies corresponding to the matched game situation information, wherein the recorded data of the plurality of curling matches performed includes at least one of image information and text information.

In addition, the curling competition strategy recommendation method according to another aspect of the present invention, extracting the curling data including the location information and the strategy information of the stone from the data collected from the recorded data for the plurality of curling competitions carried out, and proceeds Extracting curling data including the location information of the stone from the game situation information of image recognition of the image of the current curling game; Extracting, from the curling data, multiple features including a feature according to a curling rule, a feature in an existing race, and an error feature of the ongoing race; Learning the multiple features through deep learning and Monte Carlo tree search techniques to build a strategy recommendation model; And recommending information predicting one or more strategies corresponding to the game situation information through the strategy recommendation model.

According to an embodiment of the present invention, while detecting the culling stone pitching strategy through deriving the optimal pitching candidate group based on deep learning, by applying a fast tree search algorithm through the game analysis technology, it is possible to quickly and accurately determine the curling game situation It is possible to establish an optimal game strategy.

In other words, by using deep learning based culling game strategy recommendation technique and Monte Carlo tree search algorithm that enables fast 'number of cases' search, it is possible to analyze the probability of winning the culling game and the optimal stone (pitting strategy) position prediction analysis. It is possible to analyze and predict the state of the game at high speed within the curling race time.

In addition, according to an embodiment of the present invention, by using a deep learning structure that can consider multiple features (feature) can avoid overfitting and reflect various features such as important rules or elements in the curling competition analysis and strategy The accuracy of the establishment can be improved.

1 is a block diagram of an apparatus for recommending a curling match strategy according to an embodiment of the present invention.
2 is a view showing a curling stone location feature map according to an embodiment of the present invention.
3 is a view showing a score-based curling stone location feature map according to an embodiment of the present invention.
4 is a diagram for describing a method of generating a dynamic split coordinate system according to an embodiment of the present invention.
FIG. 5 is a diagram for describing a deep learning structure based on multiple feature fusion according to an embodiment of the present invention.
6 is a diagram illustrating a concept of a deep learning design method for recommending a strategy for a curling game according to an embodiment of the present invention.
7 is a view for explaining an overview of the optimal pitching strategy recommendation method using deep learning according to an embodiment of the present invention.
FIG. 8 is a diagram for explaining a Monte Carlo tree search method in a culling simulation based on a coordinate space according to an embodiment of the present invention.
FIG. 9 illustrates a method of processing a Monte Carlo tree search using a deep learning based pitching candidate region according to an embodiment of the present invention.
FIG. 10 is a diagram illustrating a pitching strategy recommending method applying reinforcement learning reflecting uncertainty according to an embodiment of the present invention.
11 is a diagram illustrating an example of a pitching strategy candidate derivation and recommendation result according to an exemplary embodiment of the present invention.
12 is a flowchart illustrating a curling match strategy recommendation method according to an embodiment of the present invention.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and like reference numerals designate like parts throughout the specification.

Throughout the specification, when a part is "connected" to another part, this includes not only "directly connected" but also "electrically connected" with another element in between. . In addition, when a part is said to "include" a certain component, which means that it may further include other components, except to exclude other components, unless specifically stated otherwise, one or more other features It is to be understood that the present disclosure does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, parts, or combinations thereof.

1 is a block diagram of an apparatus for recommending a curling match strategy according to an embodiment of the present invention.

As shown in FIG. 1, the curling game strategy recommendation apparatus 100 includes a communication module 110, a memory 120, a processor 130, and a database (DB) 140. Meanwhile, the curling match strategy recommendation apparatus 100 may further include a display (not shown) means by itself, and various information processed through the processor 130 may be output through a display.

The communication module 110 receives image data continuously photographing the curling competition situation from one or more camera devices (not shown) installed in the curling stadium, and through the communication network 400, through the camera device (or the camera device) of the curling stadium. An external device for transmitting the captured image data).

In addition, when the curling match strategy recommendation apparatus 100 transmits information, such as a result of analyzing the playing strength and strategy for the curling match, to an external information requesting terminal (not shown), the communication module 110 transmits the communication network 400. It can communicate with the corresponding information request terminal (eg, user terminal).

In addition, the communication module 110 is connected to the Internet network through the communication network 400, under the control of the processor 130 collects data (ie learning data) for the existing curling competitions through the Internet network. To the processor 130.

The communication module 110 may be a device including hardware and software necessary for transmitting and receiving a signal such as a control signal or a data signal through a wired or wireless connection with another network device.

The memory 120 stores a curling race strategy analysis and recommendation program for predicting and recommending an optimal pitching strategy for a currently running curling race based on deep learning and Monte Carlo distribution techniques.

Such a curling match strategy analysis and recommendation program allows to build a curling match strategy recommendation model based on data data collected for an existing curling match, and to analyze an ongoing curling match. Recognizing the game situation through the image recognition and information analysis of the data related to the and includes a algorithm for predicting and recommending a curling game strategy suitable for the game situation recognition information through the curling game strategy recommendation model.

The memory 120 may be referred to as a nonvolatile storage device that maintains stored information even when power is not supplied, or a volatile storage device that requires power to maintain stored information.

On the other hand, the curling competition strategy recommendation apparatus 100, various information related to the analysis and recommendation of the curling competition strategy (that is, learning data and feature data collected and extracted in relation to the curling competition, image data of the analysis target curling competition, etc.) It may further include a database 140 for storing the culling game strategy recommendation model modeled by applying the deep learning and Monte Carlo distribution techniques. The database 140 may be built as a separate component on the curling game strategy recommendation device 100, or may be included in the memory 120.

The processor 130 executes a program stored in the memory 120, and performs the following processing according to the curling game strategy analysis and execution of the recommendation program. For reference, the processor 130 may refer to a data processing apparatus embedded in hardware, for example, having a physically structured circuit for performing a function represented by a code or an instruction included in a program. As an example of a data processing device embedded in hardware, a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, and an application-specific integrated device (ASIC) It may include a processing device such as a circuit, a field programmable gate array (FPGA), etc., but the scope of the present invention is not limited thereto.

The processor 130 recognizes preset game situation information by recognizing an image of an image of an ongoing curling game, and learns data collected from the record data of a plurality of previously performed curling games based on deep learning. Construct a recommendation model, predict one or more strategy candidates corresponding to the recognized game situation through the strategy recommendation model, and output recommendation information corresponding to the predicted strategy on a display. At this time, the recorded data for the plurality of curling competitions carried out includes at least one of image information and text information.

First, the processor 130 will be described with respect to the process of collecting the data on the curling game carried out, and extracting the curling data from the collected data data.

The processor 130 may collect big data about the curling game required for the machine learning, for example, may collect the curling game result accumulated for several years to several decades from the recent.

In this case, the processor 130 may process data using an image recognition technique and a text mining technique to secure big data ranging from hundreds of thousands to millions. The processor 130 may extract location information of stones from each piece of data using image recognition, and may acquire strategy information using text mining. In addition, a curling rule based curling strategy may also be used as curling data, and pitching error related data may be extracted according to the degree of friction of the ice of the curling stadium.

In addition, since deep learning-based learning requires a large amount of data, in order to expand the insufficient data, the processor 130 may expand the data using left-right symmetry and random error insertion used in image recognition.

The processor 130 collects existing curling game records (eg, data provided by Curling Information Technology (CURLIT), etc.) through web crawling to obtain data necessary for learning a curling strategy. can do. For reference, international curling notation (curling notation) is a document that records the content of the international curling competitions hosted by the World Curling Federation is published in a pdf format data file, a document that the culling experts observe the actual game recorded. The information included in the curling game notation includes game image information continuously recording the arrangements of stones, and text information such as end number, shot number, strategy performed, stone rotation direction, and the like. More specifically, the curling match notation includes end number, shot number, name of country, name of player, performed strategy, direction of rotation ( Direction of curl, Score of current stategy, Current state of stones, and the like.

The processor 130 extracts and stores the collected curling game records in the database 140 to extract information necessary when the strategy generation model learns through template matching. Template matching is a pattern matching technique using a predetermined template (template), and is a method of extracting an object matching a given figure from an image.

In this case, the processor 130 converts the document file (eg, pdf) of the collected curling game record data into an image file, extracts a predetermined required area (ie, a stadium area) through outline recognition, and extracts the extracted image file. Image recognition is performed to extract image information such as coordinates of a curling stone and collision position coordinates existing on the image. The processor 130 stores the extracted image information and the curling match record image file in the curling match database.

In addition, the processor 130 changes the document file (eg, pdf) of the collected curling game record data into a text file format (txt), and then texts such as an end number, a shot number, a strategy performed, and a stone's rotation direction. Extract the information. The processor 130 stores the extracted text information and the curling match record text file in a curling match database.

As another method for securing data, the processor 130 may reconstruct a game image taken at an actual curling stadium to recover a screen viewed from the vertical direction of the stadium. An image reconstruction process is an operation method which synthesize | combines an object or its cross-sectional image by the image conversion technique from the combination of the imaging data lower than that by one dimension. The processor 130 may perform image recognition processing of the restored screen (ie, image) to acquire data (ie, game situation information) necessary for culling strategy learning.

At this time, the processor 130 restores the image of two or more cameras (eg, a camera next to the stadium, a stadium top (ceiling) camera, etc.) installed in the curling stadium to an image taken in the vertical upper direction through image reconstruction. The processor 130 matches the recovered image with the position coordinates of a preset curling simulator, extracts information (ie, game situation information) necessary for the strategy generation model to learn through template matching, and then extracts the database 140. ).

The processor 130 reproduces the movement of the curling stone and the collision between the curling stones using a physics engine-based curling simulator, and sets the curling sheet, which is an infinite space, as the coordinate space to indicate the location of the curling stones as coordinates. . For reference, the physics engine may be implemented with computer software capable of maximizing simulation of simple specific physics systems such as rigid body dynamics (including collision detection), soft body dynamics, and flow dynamics.

As described above, the processor 130 databases the existing curling game record data collected through web crawling and the curling game situation data recognized based on image recognition, so as to learn by using a machine learning technique.

In addition, the processor 130 may process data augmentation of the collected curling data to obtain a large amount of data required for strategy learning. The data expansion process arbitrarily expands the input data. For example, image data is a method of increasing the number of data by artificially making small changes such as rotating, cropping, tilting, and adjusting brightness of an image. For example, the processor 130 may perform left and right inversion processing, scale reduction or enlargement processing, left and right rotation processing, and curling stone on the image data extracted from the recorded data of the existing curling race and the image data of the ongoing curling race. At least one data expansion process of the noise addition process for the position coordinate of is used as the learning data of the machine learning.

In addition, the processor 130 may further secure data by extracting predetermined types of information required when the strategy recommendation model is trained using self-matching competition information between curling simulators learning different strategies based on machine learning. have.

Next, a process of extracting a feature of a predetermined kind from the extracted curling data will be described with reference to FIGS. 2 to 4.

The processor 130 extracts unique features of curling for curling race situation recognition and learning, but extracts multiple features to handle fusion of various features. At this time, the unique feature of curling is data related to important variables in the curling game, such as stone position, curling rules, the maximum score of the current game situation, existing game content, pitch error in the actual game, and the opponent's game strategy. Can be entered.

The processor 130 applies temporal / spatial / statistical methods to extract features from the acquired curling data, so that the characteristics according to the curling rule, the features in the existing game, and the error feature in the actual game (ie, the current game) Extract the back. In this case, the processor 130 extracts a feature vector by using variance, standard deviation, correlation coefficient, auto-regressive technique, and the like when extracting each feature.

A feature based on the culling rule is a feature related to a predefined culling strategy.

For example, curling strategies include slow stones, including draw, front, guard, raise, wick, and freeze, and take out. And quick stones including hot and roll, clearing, double take out and promotion take out. In this case, the draw refers to a shot that accurately sends the stone to a desired position inside the house without touching another stone. The front means a shot that sends the stone in front of the house, and the guard means placing the stone in front of the house to protect another team's stone. Wick means a shot that places a stone different from the stone in a desired position, and freeze means a shot that sends the stone as close as possible to another stone. Take out means to keep the stone in the playing field and send the other team's stone off the playing field, and hit and roll means to send the stone to the desired position and to send the other team's stone out. Also, clearing is a shot that sends out both the stone and the stones of other teams, and double take-out is a shot that sends out the stones of two other teams. Promotional take-out means pushing your team's stone out to get another team's stone out.

 In addition, the characteristics of the existing game, based on the result of the game when the victory and defeat for each existing game to extract the optimal feature through different reinforcement learning. By learning the error of pitching due to uncertainty, it is possible to secure stable performance by adjusting the weight of dangerous strategy.

The processor 130 may extract features in the curling data based on the location coordinates of the stone.

2 is a view showing a curling stone location feature map according to an embodiment of the present invention, Figure 3 is a view showing a score-based curling stone location feature map according to an embodiment of the present invention, Figure 4 is a present invention A diagram for describing a method of generating a dynamic split coordinate system, according to an exemplary embodiment.

Referring to FIG. 2, the processor 130 configures a feature map by reflecting a location of a stone included in each curling data in a map (that is, a map corresponding to the location coordinates of a curling simulator), but the first team (eg, Coordinate the stone position of the user's team to be analyzed, the stone position of the second team (eg, the opposing team), and the empty space, respectively. In other words, the processor 130 allocates a coordinate system to search for continuous space.

Referring to FIG. 3, the processor 130 simulates the pitching of the first team's stone positions from all the coordinates based on the game situation at the time of analysis, calculates the expected scores, and calculates the scores for each score after the pitching. In this case, a stone position feature map is constructed.

In addition, referring to FIG. 4, the processor 130 may configure a feature map using a coordinate system having the same interval of all position coordinates. In addition, the processor 130 finely divides the interval of the position coordinates within a certain range from the center of the house in order to increase the accuracy of the curling pitching strategy. The feature map may be constructed by applying the divided dynamic segmentation coordinate system. In other words, the processor 130 may apply the dynamic coordinate system allocation method in order to maximize the reflection of the game situation and strategy.

Next, a process of designing and learning a deep learning structure for learning features extracted from the curling data by the processor 130 will be described.

The processor 130 learns based on deep learning using the extracted curling data features as learning data. In this case, the processor 130 designs a multiple feature fusion deep learning structure that can simultaneously learn about features having different dimensions in order to use the extracted various features. This avoids errors such as overfitting and enables faster and more appropriate learning and updating while reducing the number of pitches (ie, strategies) candidates on the Monte Carlo tree.

FIG. 5 is a diagram for describing a deep learning structure based on multiple feature fusion according to an embodiment of the present invention.

Referring to FIG. 5, the processor 130 simultaneously learns multiple features including rule-based features, learning-based features, and features of uncertainty among the features of the curling data to predict the location coordinates of the stones.

The processor 130 generates a game strategy based on the Monte Carlo tree search using deep learning.

6 is a diagram illustrating a concept of a deep learning design method for recommending a strategy for a curling game according to an embodiment of the present invention.

The processor 130 predicts the location of the curling stone based on deep learning, but uses a convolutional neural network (CNN) -based deep learning that can accurately identify the game situation and establish a pitching strategy through a curling game strategy and situation identification algorithm. Construct the structure (shown in Figure 6).

FIG. 7 is a view for explaining an outline of an optimal pitching strategy recommendation method using deep learning according to an embodiment of the present invention, and FIG. 8 is a Monte Carlo simulation in a curling simulation based on a coordinate space according to an embodiment of the present invention. It is a figure for explaining a tree search method. FIG. 9 is a diagram illustrating a method of processing a Monte Carlo tree search using a deep candidate based pitching candidate region according to an embodiment of the present invention.

Referring to FIG. 7, the processor 130 learns through a strategy recommendation model based on deep learning for a given game situation (ie, current game situation) to generate a candidate group for the next pitch. In this case, the processor 130 may learn through a strategy recommendation model applying the Monte Carlo tree search technique as shown in FIG. 8, but may process the Monte Carlo tree search in a coordinate space-based curling simulator.

Monte Carlo tree traversal is a heuristic search algorithm for decision making, especially when playing games. Analyzing how moving is the most promising, we focus on expanding the search tree based on randomization in the search space. Applying Monte Carlo tree navigation to a game is based on many 'playouts'. At each playout, you'll end up with a random selection. Nodes are weighted as the final game result for each playout to increase the likelihood of selecting them in future playouts. The most basic way to use the playout is to apply the same number of playouts after each participant that matches the rules, and choose the move that led to the greatest number of wins. This method is often more efficient as time goes on, as more playouts are imposed on the movement that led the player to victory in the previous playout. Each time the Monte Carlo tree traversal consists of four steps: selection, expansion, simulation and back propagation. Selection starts at the root and selects successive child nodes. When the game is not won at the end of the game and expansion ends, one or more child nodes are created or the other node is selected. Simulation performs a random playout from the node. Backpropagation updates the information of nodes along all paths as a result of playout. By using this method, many candidate groups can be excluded except for candidates having a higher probability than the existing game tree. In this way, a quick strategy recommendation is possible.

Monte Carlo tree navigation alone finishes the game through random selection in each playout. The processor 130 according to an embodiment of the present invention selects based on the weights learned in consideration of a culling strategy suitable for a game situation, and finally, The result of the game is to update the node weights to find the optimal strategy and to perform fast search.

The processor 130 builds a strategy recommendation model for establishing an optimal strategy by reducing a pitching candidate through a deep learning based fast search algorithm in order to establish a limited time strategy given to each team in a curling game.

Referring to FIG. 9, the strategy recommendation model receives a current game situation using a CNN and outputs a candidate group of a next pitching strategy through a Monte Carlo tree search.

In addition, the processor 130 may select the priority of the candidates of the Monte Carlo tree using deep learning weights to reduce the candidate width of the wide tree based on the coordinates.

In this case, in order to overcome the problem of interfering with the outcome of the uncertainty and to prevent the optimal learning, the processor 130 performs the learning in consideration of the uncertainty through the reinforcement learning technique that can normalize the outcome of the uncertainty. The processor 130 may reduce the candidate width of the Monte Carlo tree by utilizing the error of the learned pitch as the deep learning weight factor. As such, we design a reinforcement learning structure that reflects the reliability by evaluating the outcome of the outcome. For reference, reinforcement learning is a method in which a device defined in an environment recognizes the current state and selects an action or a sequence of actions that maximizes reward among selectable actions.

FIG. 10 is a diagram illustrating a pitching strategy recommendation method using reinforcement learning reflecting uncertainty according to an embodiment of the present invention.

Referring to FIG. 10, the processor 130 may process reinforcement learning that learns a strategy having a high winning rate through self-confrontation. As shown in FIG. 10, the processor 130 processes the reinforcement learning technique of updating the deep learning variable to make a throwing pitch and also updating the deep learning variable so as not to throw the throwing pitch. As the game progresses through this reinforcement learning technique, technology is gradually advanced, and it is possible to recommend a culling strategy through real-time additional learning according to the game situation, and also to recommend a more stable strategy by learning about uncertainty.

Next, the processor 130 recommends a curling match strategy based on the results of the deep learning learning.

The processor 130 performs optimal strategy prediction based on deep learning to which rule-based recommendation and learning-based recommendation are simultaneously applied. The processor 130 learns the curling data (for example, the game situation data input in real time) according to the current game situation through the strategy recommendation model based on the Monte Carlo tree search technique based on deep learning, and the result of the learning. As the next pitching candidate group (ie, at least one strategy recommendation candidate) on the display.

In this case, the processor 130 recommends an optimal curling match strategy that recommends 'force', 'rotation', and 'path' of the stones, which are the basic elements of the curling pitch, through deep learning learning through the strategy recommendation model. Depending on the opponent's strategy, the optimal curling race strategy may be updated.

That is, the processor 130 outputs stone force, rotation, and path information according to an optimal curling strategy recommended by learning through a strategy recommendation model rescued based on deep learning and Monte Carlo tree search techniques. In this case, the processor 130 may generate a table of estimated position coordinates of the stone through a plurality of simulations and process coordinate matching accordingly in order to convert the recommended curling strategy into the optimal stone force, rotation, and path information. .

11 is a diagram illustrating an example of a pitching strategy candidate derivation and recommendation result according to an exemplary embodiment of the present invention.

In FIG. 11A, an example of a result of deriving a stable pitching strategy candidate as a result of learning an error due to uncertainty is shown. In addition, in FIG. 11B, the force, rotation, and path information of the stone are matched with the heat map coordinates based on the position coordinate table as a strategy recommendation result. The processor 130 may provide the curling match strategy recommendation result in real time for the user to check.

Hereinafter, a curling match strategy recommendation method according to an embodiment of the present invention will be described with reference to FIG. 12.

12 is a flowchart illustrating a curling match strategy recommendation method according to an embodiment of the present invention.

First, big data about pre-set curling races for performing deep learning learning are collected to extract curling data of a predetermined type and form (S121).

In this case, the curling data may be extracted from the competition data that have been previously performed using image recognition and text mining techniques. In addition, the collected curling data may be subjected to data augmentation, thereby obtaining a large amount of data required for strategy learning.

Next, various features required for learning through the curling competition strategy recommendation model are extracted from the collected curling data (S122).

In this case, data related to a significant variable in a curling game, such as a stone position, a curling rule, a maximum score of a current game situation, an existing game content, an error of a pitch in an actual game, and an opponent's game strategy, can be extracted as a feature. For example, using variances, standard deviations, correlation coefficients, auto-regressive techniques, etc., features based on curling rules, features from existing races, and errors from actual races (ie, races in progress). Extract multiple features such as

Then, learning based on deep learning using the extracted curling data features as learning data to build a strategy recommendation model (S123).

At this time, we can design a multi-feature convergence deep learning structure that can learn about different features at the same time, but build a strategy recommendation model using convolutional neural network-based deep learning and Monte Carlo tree search. Through this, the extracted culling data features are learned through a strategy recommendation model based on deep learning to generate candidate groups for the next pitch.

In addition, a strategy recommendation model of the reinforcement learning structure that reduces the candidate width of the Monte Carlo tree may be generated by utilizing the error of the learned pitch as a deep learning weighting factor.

Next, using the established strategy recommendation model to learn the current curling game situation, and outputs the result of the curling strategy recommendation accordingly (S124).

In this case, the pitching strategy recommended through the optimal strategy prediction may be output by converting the stone's force, rotation, and path information, and a corresponding position coordinate of the stone may be output together.

One embodiment of the present invention can also be implemented in the form of a recording medium containing instructions executable by a computer, such as a program module executed by the computer. Computer readable recording media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. In addition, the computer readable recording medium may include a computer storage medium. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data.

Although the methods and systems of the present invention have been described in connection with specific embodiments, some or all of their components or operations may be implemented using a computer system having a general purpose hardware architecture.

The foregoing description of the present invention is intended for illustration, and it will be understood by those skilled in the art that the present invention may be easily modified in other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are exemplary in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined form.

The scope of the present invention is shown by the following claims rather than the detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present invention. do.

100: curling match strategy recommendation device 400: communication network
110: communication module 120: memory
130: Processor 140: Database

Claims (10)

In the curling match strategy recommendation device,
A memory storing a curling game strategy analysis and recommendation program using deep learning and Monte Carlo tree traversal techniques; And
A processor for executing a program stored in the memory;
The processor according to the execution of the curling match strategy analysis and recommendation program,
Recognize preset game situation information by recognizing the image of the image of the ongoing curling game, and deep learning and Monte Carlo tree search on the game situation information and the data collected from the recorded data of the plurality of curling events. Learn through a technique to build a strategy recommendation model, and through the strategy recommendation model recommends information predicting one or more strategies corresponding to the recognized game situation information,
The recorded data of the plurality of curling competitions carried out include at least one of image information and text information,
Coordinates the position of the stone among the features extracted from the recorded data and the curling data according to the ongoing curling game, wherein the interval between the position coordinates within the preset range corresponding to the house in the curling game is outside the preset range. Curling match strategy recommendation apparatus, which uses a dynamic split coordinate system set narrower than the interval between.
The method of claim 1,
The processor,
Extracting the location information of the stone from the data collected from the recorded data and the game situation information using image recognition technique, and extracting the strategy information from the collected data using text mining technique,
And the curling data includes location information and strategy information of the stone.
The method of claim 2,
The processor,
Image data according to the collected data and the game situation information is subjected to at least one data augmentation process among left and right inversion processing, scale reduction or enlargement processing, left and right rotation processing, and noise addition processing for the position coordinates of the curling stone. A curling match strategy recommendation device for extracting curling data.
The method of claim 2,
The processor,
From the curling data, a multi-feature including a feature according to a curling rule, a feature in an existing game, and an error feature of the ongoing game is extracted,
The multiple features,
Curling match strategy recommendation device, including at least one kind of the position of the stone, the curling rules, the maximum score of the current game situation, the existing game content, the error of the pitch in the actual match, the opponent's match strategy.
The method of claim 4, wherein
The processor,
And constructing the strategy recommendation model through deep learning learning based on a convolutional neural network (CNN) that simultaneously learns about the multiple features having different dimensions.
The method of claim 4, wherein
The processor,
Curling game strategy recommendation apparatus for constructing the strategy recommendation model by setting the priority of the pitching candidate on the Monte Carlo tree by applying the error of the pitch as a weighting factor in deep learning.
The method of claim 6,
The processor,
Curling game strategy recommendation apparatus for evaluating the win or lose according to the results of the simulation simulation for each of the curling data through a preset curling simulator, and performing the reinforcement learning by updating the result of the evaluation with the variables of deep learning.
delete The method of claim 1,
The processor,
Learning the ongoing game situation information using the established strategy recommendation model, and outputs the predicted pitching strategy through the learning,
And outputting at least one of a result of changing the pitching strategy by the force, rotation, and path information of the stone and the estimated position coordinates of the stone.
In the curling match strategy recommendation method using a curling match strategy recommendation device,
Curling data including stone location information and strategy information is extracted from data collected from a plurality of recorded culling races, and stone from race state information obtained by image recognition of an image of an ongoing culling race. Extracting curling data including location information of the;
Extracting, from the curling data, multiple features including a feature according to a curling rule, a feature in an existing game, and an error feature of the ongoing game;
Learning the multiple features through deep learning and Monte Carlo tree search techniques to build a strategy recommendation model; And
Recommending information predicting one or more strategies corresponding to the game situation information through the strategy recommendation model;
Coordinate the position of the stone among the features extracted from the recorded data and the curling data according to the ongoing curling game, wherein the interval between the position coordinates within the preset range corresponding to the house in the curling game is outside the preset range. The method of recommending a curling match strategy, wherein the dynamic split coordinate system is set to be narrower than the interval between.

Images