CN107019901B - A method of establishing an automatic game robot for chess and card games based on image recognition and automatic control - Google Patents

Abstract

The invention discloses a method for establishing an automatic gaming robot for chess and card games based on image recognition and automatic control. The method includes: automatically recognizing game scenes based on image recognition technology; performing corresponding operations on each game scene to enter the game versus game scene; recognizing the action sequence and current game information of current game participants; and transmitting the recognized information to artificial intelligence computing system, the coping strategy is given by the artificial intelligence computing system; the automatic operation technology is used to automate the operation according to the coping strategy given by the artificial intelligence computing system. The invention makes it possible to test the machine game system on a large scale with human players, and provides a new battle testing method, large-scale experimental method and database data accumulation method for the research on game problems with human players in the field of artificial intelligence.

Description

Automatic game robot for chess and card games based on image recognition and automatic control build method

technical field

The invention belongs to the technical field of artificial intelligence and machine gaming, in particular to a method for establishing an automatic gaming robot for chess and card games based on image recognition technology and automatic control technology combined with machine gaming technology. The present invention can provide a large-scale online test platform for testing with human players for researchers and scientific research institutions studying machine gaming technology; the platform can be used for machine gaming by establishing automated gaming robots to conduct large-scale tests with human players. Domain research provides more convincing verification of intelligence levels. At the same time, the present invention provides a method for establishing a large-scale human-machine game information database, which provides a data basis for the research represented by the deep learning method in the current artificial intelligence field. The invention is realized through the design of a game scene recognition method, a game information recognition method and an automatic mouse operation method.

Background technique

Artificial intelligence is an important branch of the computer field. Its central task is to study how to make computers do the work that can only be done by human intelligence. As a research field of artificial intelligence, machine game is one of the means to test the development level of artificial intelligence. For more than half a century, machine games have been a hotbed of AI development and innovation, and the resulting success is an important milestone in the history of AI development. From Deep Blue (chess) to Cepheus (Texas Hold'em) to AlphaGo (Go), machine gaming systems have challenged the highest human intelligence in one area after another. That is to say, the machine gaming system can already compete with human players on the same stage. However, the current machine game system does not have the ability to operate on the common Internet platform of chess and card games. In order to let the machine game system compete with human players on the same stage, the present invention adopts the method based on image recognition, so that the computer can understand each scene in the game platform and obtain relevant information therefrom.

Image recognition refers to the technology that uses computers to process, analyze and understand images to identify targets and objects in various patterns. At present, image recognition technology is improving day by day, and new records are constantly being created in various image recognition competitions. Although the current technology cannot achieve 100% accuracy in some relatively complex recognition tasks, in the task of the present invention, image recognition is relatively simple. The main reasons are: (1) the original pictures obtained on the game platform are relatively stable and have less noise; (2) the characteristics of the categories to be identified are relatively simple, and there is a big difference between the game scene and the game information.

SUMMARY OF THE INVENTION

In view of the technical problems existing in the prior art, the present invention proposes a method for establishing an automatic gaming robot for chess and card games based on image recognition technology and automatic control technology. The system of the present invention is actually an automated gaming robot based on an Internet gaming platform for human players, including the robot's "eye": a game information recognition system based on image recognition technology; "hand": an automated operating system; and "brain": Artificial intelligence machine game system. Through the recognition system, the human-recognizable visual information is transformed into the computer-recognizable data structure, the strategy calculation is performed through the machine game system, and the operation interaction with the Internet game platform is finally realized through the automated operating system.

The method for establishing an automatic game robot for chess and card games based on image recognition and automatic control of the present invention, the steps of which include:

A method for establishing an automatic game robot for chess and card games based on image recognition and automatic control, the steps of which include:

1) Automatically identify game scenes based on image recognition technology;

2) Perform corresponding operations on each game scene and enter the game versus game scene;

3) Identify the current game participant's action sequence and current game information;

4) Introduce the information identified in step 3) into the artificial intelligence computing system, and the artificial intelligence computing system provides a coping strategy;

5) Using automated operation technology to perform automated operations according to the coping strategies given by the artificial intelligence computing system.

Further, after the game is decided by the authorities, the result of the game is recorded, and then the game scene is automatically recognized, and the next game is entered.

Further, step 1) utilizes the characteristics of each scene of the game platform to distinguish different game scenes; the characteristics include RGB characteristics of multiple reference positions.

Further, in step 3) the action sequence of the game participant includes whether it is his turn to play the card, and the current game information includes the hand card and the cards revealed by each player.

Further, step 3) when carrying out the described identification, first utilize the method of windows system screen shot to transform the game process into a picture frame, and carry out feature extraction to this picture frame, and then adopt the classifier based on k nearest neighbors to classify the picture. .

Further, the image feature extraction method adopted in step 3) includes binarization, edge detection and horizontal threading feature extraction, and then an appropriate window is selected according to the size of the image, and the information in the window is counted as a feature.

Further, in step 3) when classifying pictures based on the k-nearest neighbor method, a cross-validation method is adopted to select the optimal k value; the picture classification feature adopts the Euclidean distance as the metric distance.

Further, the given method of the corresponding strategy in step 4) depends on the specific artificial intelligence computing system realized by the prior art, such as: chess artificial intelligence system etc.; the corresponding strategy should meet the specific chess and card game rules of the current application, but the strategy itself Good or bad does not affect the operation of the system claimed in the present invention.

Further, step 5) utilizes C++ to call the automatic control function of the windows system to operate on the game platform, so as to realize a complete computer automatic game.

An automatic game robot for chess and card games based on image recognition and automatic control, comprising an image recognition system, an automated operating system and an artificial intelligence computing system, the image recognition system uses image recognition technology to automatically recognize game scenes, and when entering a game versus game After the scene, the action sequence and current game information of the current game participants are identified; the artificial intelligence computing system provides a response strategy to the information identified by the image recognition system; the automatic operating system adopts automatic operation technology and calculates according to artificial intelligence The response strategy given by the system is automatically operated.

The main significance and beneficial effect of the present invention are:

In the prior art, there is no automatic game robot for automatic human-machine game. Even in AlphaGo's competition against human Lee Sedol, the AlphaGo system still needs a person to help it recognize scenes and implement operations. This situation has led to the training and testing of previous machine game system research, mainly from the battle between itself and other machine game systems, because only in this way can large-scale battle testing be achieved. Occasionally, research institutions invite professional human players to conduct small-scale battle tests. Therefore, the present invention enables the large-scale testing of the machine game system and human players, and provides a new battle testing method, a large-scale experimental method and a database data accumulation method for the research on game problems with human players in the field of artificial intelligence. At the same time, the invention has the function of automatic error correction, and the accuracy rate of identifying various valid information of the game platform reaches 99.9%.

Description of drawings

Figure 1. General framework diagram of the method of the present invention.

Figure 2. Determine the swing flow chart.

Figure 3. Feature extraction flow in image recognition.

Figure 4. Example of binarization in feature extraction.

Figure 5. Horizontal threading feature extraction process for figure 5.

Figure 6. The horizontal threading feature extraction results of playing cards.

Figure 7. Example of statistical feature extraction results.

Figure 8. Image recognition flow.

Detailed ways

The present invention will be further described below through example figures and accompanying drawings.

Fig. 1 is the overall framework of the method of the present invention, and the specific steps include: automatically identify game scenes based on image recognition technology; perform corresponding operations on each scene to enter the game battle scene; identify whether it is your turn to play cards and current card table information (including The hand cards, the cards shown by each player) are classified by the classifier based on k nearest neighbors; the current table information is passed into the AI function, and the AI will give the countermeasures; the automatic operation technology is used to respond according to the strategies given by the AI Lay the cards; record the score of the game after the game is won or lost. A specific description is given below.

1. Identify the game scene

The application scenario of the present invention is to use a mobile phone virtual machine in a computer to run a game platform, and the game is a chess and card game. In order to reduce the operation process, we concentrate the main functions in the operations related to the battle, so it is assumed that the game platform has been opened and logged in, and the logged-in page is called the main page. In scene recognition, there are three scenes: unrecognized scene, normal battle, and special battle. Each scenario has its corresponding operation. Here, the specific implementation of the operation is realized by relying on the API (application programming interface) based on the windows system, which will be described in detail later.

1) Unrecognized scene: refers to other scenes other than the battle scene, such as the main page or the pop-up window is not closed, etc. Collectively referred to as exceptions. At this time, the computer will determine what kind of anomaly it belongs to, and select the appropriate action for the anomaly. If it stops on the main page, click to match players and enter the game battle page; if there is an unclosed small window, click the close button.

2) Common battle: the operation on the common battle page is also the main content of the present invention. Here, it is necessary to let the computer understand the situation of the card table at this time and make corresponding operations. Divide the ordinary battle into three situations: (1) In the first round, at this time, we will first identify the five cards dealt in the first round and the placement of the other two cards, and pass this information into the artificial intelligence computing system. (that is, the AI in Figure 1, which can also be called a machine game system), and then place the cards according to the obtained five-card placement method, and determine whether the placement of the cards is correct. Then check the game record of the previous game, and record the result of the previous game to count the performance of the machine game system. (2) Waiting for other players, this does not need to do anything, simply let the process sleep for a period of time to reduce computer overhead. (3) For other rounds, at this time, the three hand cards issued, the placement of the other two cards and the remaining cards will be identified, and these information will be transmitted to the artificial intelligence computing system, and then placed according to the obtained swing method. cards, and determine whether the cards are placed correctly.

3) Special battles: In the Big Pineapple game, there will be special battles in some cases, and some platforms call it Dreamland. The scene at this time will be somewhat different from the layout of the normal battle, but the corresponding operations are also somewhat different. First, identify all the hand cards dealt and the placement of the other two cards, and transmit this information to the artificial intelligence computing system, and then place the cards according to the obtained placement method, and determine whether the placement of the cards is correct. Then check the game record of the previous game, and record the result of the previous game to count the performance of the machine game system.

In laying the cards, it is necessary to determine whether the laying cards are correct. In the actual operation process, there will be some exceptions that the game platform will handle by itself. For example, when the network connection is unstable, the situation will occur when the card is placed: During the operation of the mouse, the network connection is disconnected. At this time, in fact, any card placement operation cannot be successful, that is to say, the mouse operation does not have any effect there. In order to solve this situation, we have formulated a plan to determine the placement of cards. The specific process is shown in Figure 2. The specific method is: after the program operation finishes placing a card, identify whether there is a card at the destination location, and if there is no card, then Re-place the card; if there is a card, identify whether the card is the card that should be placed in this operation, if not, remove the card and place the card again.

2. Image feature extraction

In the image recognition task, the extraction of image features plays a very important role, and a good feature expression has a great impact on the accuracy of subsequent recognition. There are many image feature extraction methods, such as Histogram of Oriented Gradient (HOG) feature, Local Binary Pattern (LBP) feature, Scale Invariant Feature Transform (SIFT) feature, etc. In the task of the present invention, the content of the picture to be recognized is relatively simple, but in order to keep up with the game flow, the number of times of recognition is relatively large, so it is necessary to reduce the amount of calculation and speed up the picture recognition. Therefore, without affecting the subsequent recognition accuracy, the present invention adopts a simpler statistical information as a feature.

The main feature extraction process of the present invention for a picture I is shown in Figure 3, and specifically includes the following content:

1) Binarization: Binarization is the first step in all operations, mainly used to eliminate the influence of animation and light and shadow masks in the game on image recognition. The function of binarization is to convert the image from the grayscale RGB value-limited mode to 01 for display. Figure 4 is an example of a binarized image, in which the right side is an image, and the left side is the result of binarization processing.

2) Edge detection: The purpose of edge detection is to align the recognition area to improve the recognition efficiency and accuracy. The object of edge detection in this embodiment is to align the upper left coordinate point of the recognition area. This is achieved by identifying the intersection of pixels where a 1 appears in the first column and a 1 appears in the first row.

3) Design and color recognition: Design and color recognition is a relatively simple identification step in the system, but it plays a key role in improving the recognition speed of the system. The specific method is: by taking the pixel RGB value of a specific area of the playing card, the suit of the playing card is determined. The color characteristics determined by this system are shown in Table 1.

Table 1: Feature interval for suit recognition

playing card suit R G B square piece 0～25 0～25 200～255 grass flower 0～25 200～255 0～25 Hearts 200～255 0～25 0～25 Spades 200～255 200～255 200～255 other -- -- --

The system confirms the suit of playing cards through the above table. If it does not fall within the identification range of any suit, it will be recognized as a non-playing card or a card back pattern, and the system will jump out of the identification process of this card. Otherwise, the suit recognition result will be combined with the card content recognized by other processes to form the final result. For example: Spades + K = Spades K.

4) Horizontal threading feature extraction: The horizontal threading feature is one of the keys to identify playing cards in this system. The method is to obtain the pixel features in the horizontal direction by horizontally identifying the 01 distribution in the playing cards. In this embodiment, a total of 6 features of playing cards are collected, namely: horizontal midline feature, upper 1/4 line feature, lower 1/4 line feature, and the first 1 feature corresponding to the three features. The method to obtain features is to traverse horizontally, and record that the horizontal line intersects with several pixel points consisting of 1 in total. The first 1 feature refers to: the first intersection, whether it is on the left or right side of the recognition area, the left side is 1, and the right side is 0.

Figure 5 shows the digital threading characteristics of the number 5 in playing cards. It can be seen that the number of 1s crossing the three threads is 1, 1, and 1, respectively, and the positions where 1 is encountered for the first time are on the right, left, and right sides, respectively. , so the digit stringing feature of 5 is 111010.

Figure 6 shows the extraction results of horizontal threading features of all playing cards.

5) Feature extraction of sub-regional number column: This module divides the recognition region of playing cards into 9 regions of 3*3, and performs horizontal threading feature extraction again to supplement the global threading feature. The specific process is similar to step 4), No more Aoshu here.

6) Feature comparison: select a suitable window w according to the size of the picture I, and count the information in w as a feature. Here, since the difference between the original picture background and the content is obvious, the proportion of the content pixels is calculated, that is, The proportion of pixels with a statistical value of 1 in the window; the statistical feature vector X of image I is represented by the statistical information of all windows. which is:

X=(f(w ₁ ),...,f(w _n ))

where f( _wi ) represents the proportion of pixels with a value of 1 in the window _wi .

Figure 7 is an example of the feature extraction result. As shown in Figure 7, first we get the original image (playing card K) obtained by intercepting, and then perform feature extraction on it to obtain the picture as shown in the figure, and use an 8*10 window w to obtain the final feature vector X. Here this window is adjustable, mainly to make the dimension of the final feature vector fixed, which is convenient for the next step.

3. Image recognition

Figure 8 shows the process of image recognition: first, the image is preprocessed, mainly cropping and other operations to intercept the key area image. Then feature extraction is performed, and then the features are input into the k-nearest neighbor classifier, and finally the class label of the image is obtained.

For image recognition, enough training datasets need to be prepared for the k-nearest neighbor classifier, that is, dataset collection. The method taken here is to intercept the required data from the game platform, including pictures of different playing cards, winning and losing scores, etc. The specific method is: count the position and size of various data in the game platform, and then use the program to take screenshots to capture the corresponding position and size.

After extracting the features of the image, a classifier is needed to classify these features to obtain the required information. In the task of the present invention, it is hoped that the same method can be used to identify game information of different platforms, and additional operations can be reduced when adapting to different platforms. Through investigation, it is found that the k-means neighbor (k-NN) method meets the above requirements. First of all, it works well on simple tasks, and this is a classifier that does not require explicit learning, which means that it does not need to be trained on the training dataset in advance. This means that when adapting a platform, you only need to collect the game information data set of the platform to be adapted.

The k-nearest neighbor method is a basic classification and regression method. Its input is the feature vector of the instance, corresponding to the point in the feature space; the output is the class of the instance. The k-nearest neighbor method assumes that given a training data set, the class of instances in it has been determined. During classification, new instances are predicted by majority voting according to the categories of their k nearest neighbors of training instances. The main problems are: the choice of k value, distance measure and classification decision rules.

k-nearest neighbor algorithm:

1) Input training set:

T={(x ₁ , y ₁ ), (x ₂ , y ₂ ), ... (x _N , y _N )},

为特征向量，y_i∈Y＝{c₁,c₂…c_k}为类别；in,

is the feature vector, y _i ∈ Y={c ₁ ,c ₂ ...c _k } is the category;

2) Neighborhood N _k (x): the neighborhood of x covering the above k points;

3) Determine the category y of x:

The following describes the specific selection of the three elements in the task:

a) k value selection:

The k points in the training set that are adjacent to x. If the value of k is too small, it means that the overall model becomes complicated and is prone to overfitting, that is, if the adjacent instance points happen to be noise, the prediction will be wrong. The increase of the value of k means that the overall model becomes simpler. In the extreme case, k=N, where N represents the number of instances in the training set, then no matter what the input instance is, it is simply predicted that it belongs to the most training set. class, such a model is too simplistic. The k value generally takes a relatively small value, and a cross-validation method is usually used to select the optimal k value. That is to say, the selection of the value of k has to pass many trials to ensure a better result.

After multiple trials, different values of k were determined for different recognition tasks. For example, in the recognition of playing cards, the k value is selected as 1, that is, the nearest neighbor method is used here.

b) Distance metric:

Among them, L _p ( _xi , x _j ) represents the distance between the feature vectors x _i and x _j , n represents the number of features, when p=1, it is the Manhattan distance, and when p=2, it is the Euclidean distance. In our task we choose p=2.

c) Classification decision rule: It is generally a majority voting rule, that is, the majority class in the k training instances in the result is used as the class of x.

4. Automated operating system

The automatic operating system mainly allows the program to operate on the game platform to realize a complete computer automatic game without human intervention in the battle process. You can use C++ to call the automatic control function of the windows system to operate on the game platform. The specific method is that after opening the program and the corresponding game platform (the game platform runs on the mobile phone virtual machine), the program obtains the handle of the virtual machine with the help of the windows API, and obtains the display window of the chess and card game that needs to be recognized in the windows system position and activate the window. In order to ensure that each operation can be performed in the game platform, such as screenshot or mouse operation, an operation of obtaining a handle and activating it is performed before each operation. There are two main reasons:

(a) If the game platform is covered by other programs in whole or in part, it will result in that when the covered part of the information needs to be covered, the captured picture cannot reflect the content that actually happened in the game platform, resulting in the subsequent information identification will not be effective. , adversely affect the game system.

(b) If the game platform is covered by other programs in whole or in part, it will cause that the mouse cannot act on the platform when the mouse is required to operate on this position of the game platform, so that it cannot handle exceptions or place cards in real time.

After activating the window, the program will perform an automatic game process. During this process, you need to take screenshots to identify relevant information, and operate the mouse to complete the necessary actions. The specific methods are as follows:

(1) Screenshot

When taking a screenshot, you can cut off the entire interface, and then perform region matching to obtain the required information. However, the computer overhead of such an approach is very high. As a real-time battle, it is necessary to control the consumption of time. Considering that the layout of the game platform will not be easily changed, the following methods are adopted: First, for a game platform, it is necessary to collect useful information in each scene and its position relative to the virtual machine window, as well as its size, such as the position and size of the hand cards. Size, location and size of game scores, etc. Then, when the information of a certain size needs to be actually identified, the position of the information on the screen can be located by using the position of the window and the position of the information relative to the window. Finally, take a screenshot of that size of the screen at that location.

(2) Operate the mouse

Windows provides a very complete API to control the movement or click of the mouse, which facilitates the automated operation of the present invention. Similar to the screenshot, you first need to collect the relevant information of each operation, such as the starting position and destination position of the card required for the card placement operation, and the action of clicking the card needs to be maintained during the dragging process. Then, depending on the collected information, the corresponding API interface is used to implement the operation when the operation is required. It is worth noting that in the two consecutive mouse operations, the program needs to be suspended for a while, so that the game platform can respond to the operation. In this example, 0.3 seconds is used.

Table 2 shows the test data of a single PC computer from February 24 to March 14, 2017 on the domestic game platform Lianzhong. A total of 7,691 sets were tested in 10 days, with a total score of 10,702 points and an average score of 1.39 points per game. In addition to the contribution of the AI system, the present invention also makes key contributions: 1. The recognition accuracy rate is 98.87%, which basically guarantees the correct calculation of the AI system. 2. The accuracy rate of automated operation is 95.65%, which ensures the normal operation of the system under normal and short-term disconnection, so that the average test time per day exceeds 20 hours, and the total number of test rounds per day exceeds 700. It can be seen that if the present invention is applied to more platforms and loaded on more computers, the accumulated quantity of test data will be further greatly increased.

Table 2: The test data of the system in the Lianzhong platform Pineapple game

The above embodiments are only used to illustrate the technical solutions of the present invention rather than limit them. Those of ordinary skill in the art can modify or equivalently replace the technical solutions of the present invention without departing from the spirit and scope of the present invention. The scope of protection shall be subject to what is stated in the claims.

Claims (8)

1. a method for establishing an automatic game robot for chess and card games based on image recognition and automatic control, the steps of which comprise: 1) automatically identify game scenes based on image recognition technology; described game scenes include unrecognized scenes, common battles, and special battles, and the unrecognized scenes are other scenes than the battle scenes; 2) Perform corresponding operations on each game scene and enter the game versus game scene; 3) Identify the action sequence of the current game participant and the current game information; when performing the identification, first use the method of the windows system screenshot to convert the game process into a picture frame, and perform feature extraction on the picture frame, and then adopt the method based on the The classifier of k-nearest neighbors classifies the picture; the feature extraction includes: 3.1) Binarization: used to eliminate the influence of animation and light and shadow masks in the game on image recognition; 3.2) Edge detection: the purpose is to align the recognition area to improve the recognition efficiency and accuracy; 3.3) Suit identification: determine the suit through the RGB value of the pixel in a specific area of chess and cards; 3.4) Horizontal threading feature extraction: Obtain the pixel features in the horizontal direction by identifying the 01 distribution in the chess and cards horizontally; 3.5) Sub-regional digital column feature extraction: divide the recognition area of chess and cards into multiple regions, and perform horizontal threading feature extraction again as a supplement to the global threading feature; 3.6) Feature comparison: select a suitable window according to the size of the picture, and count the information in the window as a feature; 4) Introduce the information identified in step 3) to the artificial intelligence computing system, and the artificial intelligence computing system provides a coping strategy; 5) Using automated operation technology to perform automated operations according to the coping strategies given by the artificial intelligence computing system. 2 . The method according to claim 1 , wherein the result of the game is recorded after the official game is decided, and then the game scene is automatically recognized, and the next game is entered. 3 . 3. The method according to claim 1, characterized in that, in step 1), the characteristics of each scene of the game platform are used to distinguish different game scenes; the characteristics include RGB characteristics of a plurality of reference positions. 4. The method of claim 1, wherein in step 3) the action sequence of the game participants includes whether it is their turn to play cards, and the current game information includes hand cards and cards revealed by each player. 5. method as claimed in claim 1, is characterized in that, when step 3) is based on k nearest neighbor method when picture is classified, take the method of cross-validation to choose optimal k value; The distance measure method of picture classification feature is:

The Euclidean distance is used as the metric distance. 6. The method according to claim 1, wherein step 5) utilizes C++ to call the automatic control function of the windows system to operate on the game platform to realize complete computer automatic game. 7. method as claimed in claim 1 is characterized in that, adopt the following method to determine whether the card is placed correctly in the process of placing the card: after the program operation finishes placing a card, identify whether there is a card at the destination position, if there is no card, Then re-place the card; if there is a card, identify whether the card is the card that should be placed in this operation, if not, remove the card and re-place the card. 8. an automatic game robot for chess and card games based on image recognition and automatic control using the method according to any one of claims 1 to 7, it is characterized in that, comprises image recognition system, automatic operating system and artificial intelligence computing system , the image recognition system automatically recognizes the game scene by using image recognition technology, and recognizes the current game participant's action sequence and current game information after entering the game versus game scene; the artificial intelligence computing system recognizes the image recognition system The information provided by the artificial intelligence computing system provides a coping strategy; the automated operating system adopts an automated operation technology and performs automated operations according to the coping strategy provided by the artificial intelligence computing system.

Images