Offline chess and card activity cheating detection method, system and equipment

Abstract

According to the offline chess and card activity cheating detection method, system and equipment, analysis is carried out based on facial expressions and hand actions of players and card surface conditions on a card table during chess and card activities, whether the players cheat or not can be comprehensively judged, and compared with a single-factor judgment method, the judgment result is more accurate, and the detection efficiency is improved. And the fairness of offline chess and card activities is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of anti-cheating in chess and card games, and in particular, relates to a method, system and device for detecting cheating in offline chess and card games. Background Art

[0002] Cheating problems occur frequently in offline chess and card games. The reason is that offline chess and card games are basically judged by on-site personnel whether the players are cheating, which has great limitations and it is difficult to supervise every player. Secondly, the cheating methods of players in offline chess and card games are more diverse and more covert. For example, players can communicate with others through specific hand movements, facial expressions, etc. to achieve cheating, and it is difficult for on-site referees to detect these more covert cheating behaviors.

[0003] In response to the cheating problem in offline chess and card activities, ZL202022845739.6, "Anti-cheating detection components, mahjong machines and chess and card entertainment equipment" and ZL202311154739.3, "Unattended intelligent chess and card room management method and system" proposed their own solutions. Among them, ZL202022845739.6, "Anti-cheating detection components, mahjong machines and chess and card entertainment equipment" mainly focuses on technical improvements to the props used in chess and card activities, increasing the difficulty of cheating for players at the hardware level, while ZL202311154739.3, "Unattended intelligent chess and card room management method and system" mainly determines whether the player is cheating by judging whether the player's card-playing method complies with the rules.

[0004] It is not difficult to see from the above-mentioned existing technologies that the current anti-cheating technology for offline chess and card games does not involve the players' hand movements, facial expressions, etc., which means that players can still secretly transmit information to others and cheat, resulting in the loss of fairness in chess and card activities. Summary of the invention

[0005] The purpose of the present invention is to provide a method, system and device for detecting cheating in offline chess and card games, so as to solve the problem that the existing technology lacks detection of hand movements, facial expressions and card tables of chess and card game players, which leads to cheating of players. To achieve this purpose, the specific technical scheme is as follows:

[0006] The offline chess and card activity cheating detection method includes the following steps:

[0007] 1) Obtaining a moving image including a player's face, hands and a card table, processing the moving image, extracting the player's face, hands and card table, and obtaining face detection key points, hand detection key points and card table detection key points;

[0008] 2) Process the three key points to obtain the key point mutual trust value;

[0009] 3) Based on the facial detection key points and the mutual trust value of the key points, the cheating probability is obtained, and the cheating probability is compared with the preset cheating threshold to determine whether cheating occurs.

[0010] Furthermore, step 1) is completed by an image acquisition module and a key point detection module. Specifically, the image acquisition module uses a camera to capture the image frame of the player in real time. For a certain player Pi, the active image frame obtained in real time at a certain moment is Xi, and Xi also includes the three targets of the player Pi's face, hand and card table. Xi is input into the key point detection module (KPD) and the detection result is output. The detection result includes the face detection key point F, the hand detection key point H and the card table detection key point T.

[0011] Furthermore, the specific process of the key point detection module (KPD) is as follows:

[0012] F,H,T=Relu(ConV(Relu(ConV(Relu(ConV(X))))))

[0013] Furthermore, step 2) is completed by the key point attention fusion module. Specifically, the attention fusion module KAT first obtains the corresponding low-dimensional features f, h and t of the three key detection points F, H and T through the convolution module respectively, and then inputs f, h and t into the attention module to calculate the attention of f and h and the attention of h and t. Then, the two calculated attention vectors are concatenated and the key point mutual trust value A is obtained through a convolution module.

[0014] Furthermore, the convolution module consists of two convolution layers with a convolution kernel of 3 and a stride of 1 and a ReLU activation function.

[0015] Furthermore, the attention calculation formula is

[0016]

[0017] Further, step 3) is completed by the cheating probability prediction module. Specifically, the facial expression detection key points F and the key point mutual trust value A are input into the cheating probability prediction module (SC). The cheating probability prediction module (SC) first performs convolution processing on the facial detection key points F and the key point mutual trust value A respectively to obtain the convolution results ConV(F) and ConV(A). Then, it calculates the attention value A1 = AT(F, ConV(A)) of F and ConV(A), and inputs the result into the probability adjustment module for probability adjustment of the result. Subsequently, it calculates the attention value between the output of the probability adjustment module and the convolution result of F, A2 = AT(P, ConV(F)), and obtains the final cheating probability V = ConV(A2) through convolution of the result. Then, it compares V with the preset cheating threshold M. If V < M, player Pi does not cheat when playing cards. If V >= M, player Pi is determined to cheat when playing cards.

[0018] The offline chess and card activity cheating detection system includes:

[0019] An image acquisition module for taking images during the activity;

[0020] A key point detection module for extracting the key point information of the face, hands, and card table in the captured image frame;

[0021] A key point attention fusion module for processing the three key points to obtain the key point mutual trust value;

[0022] A cheating probability prediction module for analyzing the cheating probability based on the non-linear correlation between the facial expressions of the card player and the key point mutual trust value.

[0023] An electronic device includes a processor, a memory, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, it implements the offline chess and card activity cheating detection method.

[0024] The beneficial effects of the present invention are:

[0025] Different from the existing offline chess and card activity cheating detection methods, the present invention shifts the focus of cheating detection from chess and card props to players. By detecting the hands, faces, and card tables of players and strengthening the analysis of players' facial expressions, the cheating judgment is made more accurate, effectively avoiding the possibility of players cheating secretly and ensuring the fairness of chess and card activities. Description of the Drawings

[0026] Figure 1 is the overall flowchart of the present invention;

[0027] Figure 2 is the flowchart of the KAT module of the present invention;

[0028] Figure 3 It is the flow chart of the SC module of the present invention. DETAILED DESCRIPTION

[0029] Take the game of throwing eggs as an example. When the game is being played, each of the four players is monitored by a camera, and the game image frames of the players are obtained in real time. Specifically, the camera is used to capture the image frames of the card players in real time. For a certain player Pi, the game image frame obtained in real time at a certain moment is Xi, and Xi also includes the three targets of the player Pi's face, hands and card table.

[0030] The real-time card-playing image frame Xi of player Pi is obtained through the image acquisition module. Xi is input into the key point detection module (KPD) to output the detection results. The detection results include face detection key points F, hand detection key points H and table detection key points T. The KPD process is as follows:

[0031] F,H,T=Relu(ConV(Relu(ConV(Relu(ConV(X))))))

[0032] ConV is the convolution operation, the convolution kernel is 3, and the step size is 1. Relu is the activation function.

[0033] The key point detection module (KPD) consists of three trained object detection algorithms, each of which detects the face, hand, and card table in the image. Each detection result is expressed as four coordinate points in the figure. The four coordinate points can be connected to form an object detection box. The detection results (i.e., coordinate points) of the three algorithms are marked together in the image Xi, which is the three key point information detected by the key point detection module (KPD) at the same time.

[0034] In particular, the three object detection algorithms are obtained through supervised training of the YoLo_V5 model on three image datasets: face, hand, and card table.

[0035] Through the key point detection module, the detection results F, H, and T of an image frame X can be obtained. These three detected key points respectively correspond to the detection and recognition of the face of the card player, the hand of the card player, and the card table in the image. In order to determine whether the card player is cheating through the three detected key points, this application proposes to calculate the correlation of these three key points. Therefore, this application inputs F, H, and T into the key point attention fusion module KAT simultaneously, and outputs the key point attention value A. Specifically, the attention fusion module KAT first passes the three key detection points F, H, and T through the convolution module respectively to obtain the corresponding low-dimensional features f, h, and t. The convolution module consists of two convolutional layers with a convolutional kernel of 3 and a stride of 1, as well as the Relu activation function. Subsequently, f, h, and t are input into the attention module to calculate the attention between f and h and the attention between h and t. The attention calculation formula is as follows:

[0036]

[0037] Subsequently, the two calculated attention vectors are concatenated and passed through a convolution module to obtain the key point mutual trust value A.

[0038] After passing through the key point attention fusion module, the mutual trust value A of the three key points in the image is calculated. In order to effectively determine whether the card player is cheating, it is necessary to fully consider the non-linear correlation between the facial expression of the card player and the key point mutual trust value A. Therefore, this application proposes to input the facial expression detection key point and the key point mutual trust value A into the cheating probability prediction module (SC) for cheating probability calculation. Specifically, the cheating probability prediction module (SC) first performs convolution processing on the facial detection key point F and the key point mutual trust value A respectively, with a convolutional kernel of 4 and a stride of 1, to obtain the convolution results ConV(F) and ConV(A) respectively. Then calculate the attention value A1 = AT(F, ConV(A)) between F and ConV(A). And input the result into the probability adjustment module for result probability adjustment. The formula of the probability adjustment module is as follows:

[0039]

[0040] Subsequently, calculate the attention value between the output of the probability adjustment module and the convolution result of F, A2 = AT(P, ConV(F)). And pass the result through convolution to obtain the final cheating probability V = ConV(A2).

[0041] Set the cheating threshold M, compare the result of the SC module with M. If V < M, there is no cheating. If V >= M, it is determined that cheating has occurred.

[0042] An offline chess and card activity cheating detection system, including:

[0043] An image acquisition module for taking images during the activity;

[0044] Key point detection module, used to extract key point information of face, hand and card table in the captured image frame;

[0045] The key point attention fusion module is used to process the three key points and obtain the key point mutual trust value;

[0046] The cheating probability prediction module analyzes the cheating probability based on the nonlinear correlation between the player's facial expression and the mutual trust value of key points.

[0047] An electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, a method for detecting cheating in offline chess and card games is implemented.

Claims

1. A method for detecting cheating in offline chess and card games, characterized in that: It includes the following steps: 1) Obtain an active image including the player's face, hand, and the table part, process the active image, extract the player's face, hand, and the table part, and obtain facial detection key points, hand detection key points, and table detection key points; 2) Process the three key points to obtain the key point mutual trust value; 3) Based on the facial detection key points and the key point mutual trust value, obtain the cheating probability, compare the cheating probability with a preset cheating threshold, and determine whether cheating occurs.

2. The offline chess and card activity cheating detection method according to claim 1, characterized in that: It is described that: Step 1) is completed by an image acquisition module and a key point detection module. Specifically, the image acquisition module uses a camera to capture the image frame of the card player in real time. For a certain player Pi, the active image frame obtained in real time at a certain moment is Xi, and Xi contains the three targets of the face, hand, and table of the player Pi at the same time. Input Xi into the key point detection module (KPD), and the output detection result includes facial detection key points F, hand detection key points H, and table detection key points T.

3. The offline chess and card activity cheating detection method according to claim 2, characterized in that: It is described that: The specific process of the key point detection module (KPD) is F, H, T = Relu(ConV(Relu(ConV(Relu(ConV(X)))))) 4. The offline chess and card activity cheating detection method according to claim 2, characterized in that: It is described that: Step 2) is completed by a key point attention fusion module. Specifically, the attention fusion module KAT first passes the three key detection points F, H, and T through a convolution module respectively to obtain the corresponding low-dimensional features f, h, and t. Subsequently, input f, h, and t into the attention module to calculate the attention between f and h and the attention between h and t. Then, splice the two calculated attention vectors and obtain the key point mutual trust value A through a convolution module.

5. The offline chess and card activity cheating detection method according to claim 4, characterized in that: It is described that: The convolution module consists of two convolution layers with a convolution kernel of 3 and a stride of 1, as well as a Relu activation function.

6. The offline chess and card activity cheating detection method according to claim 4, characterized in that: It is described that: The attention calculation formula is 7. The offline chess and card activity cheating detection method according to claim 4, characterized in that: It is described that: Step 3) is completed by a cheating probability prediction module. Specifically, input the facial expression detection key points F and the key point mutual trust value A into the cheating probability prediction module (SC). The cheating probability prediction module (SC) first performs convolution processing on the facial detection key points F and the key point mutual trust value A respectively to obtain the convolution results ConV(F) and ConV(A). Then, calculate the attention value A1 = AT(F, ConV(A)) between F and ConV(A), and input the result into the probability harmonic module for probability adjustment of the result. Subsequently, calculate the attention value between the output of the probability harmonic module and the convolution result of F, A2 = AT(P, ConV(F)), and obtain the final cheating probability V = ConV(A2) through convolution of the result, and compare it with the preset cheating threshold M. If V < M, then the player Pi does not cheat when playing cards. If V >= M, then the player Pi is determined to cheat when playing cards.

8. A system for implementing the offline chess and card activity cheating detection method according to claim 1, characterized in that: It includes: An image acquisition module for taking images during activities; A key point detection module for extracting the key point information of the face, hand, and table in the captured image frame; A key point attention fusion module for processing the three key points to obtain the key point mutual trust value; The cheating probability prediction module analyzes the cheating probability based on the nonlinear correlation between the player's facial expression and the mutual trust value of key points.

9. An electronic device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the method according to any one of claims 1 to 7 when executed by the processor.

Images