Football match foul automatic detection system based on deep learning

Abstract

The invention relates to the field of deep learning, in particular to a football match foul automatic detection system based on deep learning, the football match foul automatic detection system is provided with a plurality of equidistantly arranged judgment devices, and for a single judgment device, the single judgment device comprises a locking shooting module used for shooting match condition images when participants and a target football appear at the same time, a screening processing module, and a judgment module used for judging whether the participants and the target football appear at the same time; the action learning module is used for generating corresponding match condition information, the action prediction module is used for selecting a plurality of learning features, preprocessing the match condition information and generating an action judgment result by utilizing an action learning model, the decision module is used for deciding the action judgment result, and when the action judgment result is collision, the decision module judges that participants foul, and the action prediction module is used for judging that the participants foul. And when the action judgment result is no collision, the judgment module judges that the participants do not foul, so that the foul behavior is automatically detected and judged, a more objective and accurate penalty basis is provided for the football match, the interference of human factors is reduced, and the fairness and ornamental value of the match are improved.

Description

Technical Field

[0001] The present invention relates to the field of deep learning, and in particular to an automatic foul detection system for football matches based on deep learning. Background Art

[0002] The pace of football matches is fast and complex. Referees may make misjudgments or miss calls due to limited viewing angles or momentary misjudgment during high-intensity matches. In recent years, the application of deep learning technology in the field of sports has gradually increased, especially in the analysis and judgment of football matches. The development of these technologies not only provides a more objective basis for the judgment of football matches, but also lays the foundation for the intelligent development of the sports field. The automatic foul detection system based on deep learning can analyze the game scene in real time and accurately identify foul behaviors, thereby providing referees with a more objective basis for judgment, reducing human errors, and ensuring the fairness of the game.

[0003] China Patent Authorization Announcement No.: CN117095472B discloses an AI-based swimming foul action judgment method and system, including the following steps: using visual sensors, depth sensors, and inertial measurement units to collect multimodal information of swimmers including posture data, motion trajectory, and physiological data as multi-source raw data. In this invention, by integrating visual, depth, and inertial sensor data, the swimmer's posture and action information is fully captured, complex swimming foul recognition is supported, the OpenPose three-dimensional posture estimation algorithm is used to accurately analyze postures, improve the accuracy of foul judgment, deep learning and time series analysis are used to analyze instant actions and spatiotemporal trajectories, provide rich contextual information, self-supervised and reinforced learning real-time optimization models, continuously improve judgment capabilities, and provide foul prompts and adjustment suggestions based on real-time data stream processing technology to improve swimming skills.

[0004] China Patent Authorization Announcement No.: CN118155292B discloses an AI-based dribbling foul action determination method and system, including the following steps: Based on the game scene data, the random forest algorithm is used to classify the player position, movement speed and game stage data, analyze the game intensity and the key interaction between the players, build an analysis model of the environment and action, and generate an environment and context analysis record. In this invention, the random forest algorithm is used to achieve an in-depth analysis of the player position, movement speed and game stage in the game scene, and an effective comprehensive analysis model of the environment and action is constructed, thereby enhancing the understanding of the game intensity and the interaction between the players. The spatiotemporal graph convolutional network model is used to convert the motion sequence into a graph structure, making the extraction of motion features more refined and accurate. The application of the dynamic time bending algorithm enhances the recognition ability of action patterns in differentiated time series.

[0005] However, the above method has the following problem: it is impossible to judge the real-time actions of the contestants by analyzing the real-time game images. Summary of the invention

[0006] To this end, the present invention provides an automatic foul detection system for football matches based on deep learning, so as to overcome the problem in the prior art that it is impossible to judge the real-time actions of the contestants by analyzing real-time game images.

[0007] To achieve the above object, the present invention provides a football match foul automatic detection system based on deep learning, wherein the system is provided with a plurality of equally spaced decision devices, and a single decision device comprises:

[0008] A locking shooting module is used to lock the contestants and the target football. When the contestants and the target football appear at the same time, the locking shooting module shoots the game image;

[0009] A screening processing module, connected to the locking and shooting module, for screening the game situation images and generating corresponding game situation information;

[0010] An action prediction module, which is connected to the screening and processing module, is used to select a number of learning features according to the game information, pre-process the game information, generate corresponding game prediction data, use the action learning model to learn the game prediction data, and generate corresponding action judgment results;

[0011] A decision module, which is connected to the action prediction module and is used to make a decision on the action judgment result. When the action judgment result is a collision, the decision module determines that the contestant has committed a foul. When the action judgment result is a non-collision, the decision module determines that the contestant has not committed a foul.

[0012] Wherein, the action learning model is generated through training based on the game situation prediction data;

[0013] The action judgment result is the collision condition between the contestant's arm and the target football.

[0014] Furthermore, the locking and shooting module includes:

[0015] A target monitoring device, which is used to monitor and track the position and movement trajectory of the contestants and the target football in real time;

[0016] A camera connected to the target monitoring device, used to capture the image of the game when the contestants and the target football are monitored simultaneously;

[0017] A transmission device is connected to the camera and is used to store the game image and transmit the game image to the screening and processing module.

[0018] Furthermore, the screening processing module includes:

[0019] A filtering device, which is used to filter the game image using a pixel threshold to form a corresponding filtered image;

[0020] A cutting device, connected to the filtering device, for cutting the filtered image to a standard size;

[0021] An extraction device, connected to the cutting device, for extracting the action features of the contestants and the position information of the target football, and generating corresponding game information;

[0022] The pixel threshold is the minimum standard of the pixels of the game image, and is related to the straight-line distance between the camera and the defense area of ​​the game field;

[0023] The standard size is a size that is compatible with the action learning model;

[0024] The action features include the arm position information of the contestant, the relative position of the contestant's arm and the target football, and the extension direction and angle of the arm.

[0025] Furthermore, the action prediction module comprises:

[0026] A selection device is used to select corresponding learning features according to the game information, wherein:

[0027] The learning features include the relative position and relative distance between the arm of the contestant and the target football;

[0028] A preprocessing device, connected to the selection device, for preprocessing the game information and generating corresponding game pre-data;

[0029] A learning device is connected to the preprocessing device and is used to learn the pre-match data, generate corresponding action judgment results, and mark the action judgment results according to the learning characteristics.

[0030] Furthermore, the locking shooting module is provided with a single camera, and when the target monitoring device simultaneously monitors the contestant and the target football, a shooting instruction is issued to the camera;

[0031] Wherein, the camera faces the defense area.

[0032] Furthermore, the filtering device obtains image pixels of the game image, wherein:

[0033] When the image pixel is greater than the pixel threshold, the corresponding game image is retained, and when the image pixel is less than the pixel threshold, the filtering device filters the corresponding game image.

[0034] Furthermore, the cropping device identifies the contestant and the target football and performs optimized protection, crops the filtered image to the standard size, and the extraction device extracts the standard coordinates of the contestant's arm and the target football based on the action characteristics and the position information, and integrates the standard coordinates into corresponding game information.

[0035] Furthermore, the preprocessing device cuts the game information at a standard sampling rate to generate a number of game pre-data with a sampling rate of the standard sampling rate, and the action learning model learns the game pre-data and generates a corresponding action judgment result;

[0036] The standard sampling rate is a sampling rate that can be recognized by the action learning model, and for a single sampling, the corresponding standard sampling rate is a single sampling rate.

[0037] Further, when the action judgment result is no collision, the adjudication module determines that the contestant has not violated the rules, and keeps the warning light green in response to the action judgment result.

[0038] Furthermore, when the action judgment result is a collision, the adjudication module determines that the contestant has committed a foul, the warning light is adjusted to red in response to the action judgment result, and the corresponding game image is uploaded to the visual device.

[0039] Compared with the prior art, the present invention is provided with a plurality of equally spaced decision devices. For a single decision device, the present invention comprises: a locking shooting module, for shooting a game image when a contestant and a target football appear at the same time; a screening processing module, for generating corresponding game information; an action prediction module, for selecting a plurality of learning features, pre-processing the game information, and generating an action judgment result by using an action learning model; a decision module, for making a decision on the action judgment result; when the action judgment result is a collision, the decision module determines that the contestant has committed a foul; when the action judgment result is no collision, the decision module determines that the contestant has not committed a foul, thereby realizing automatic detection and decision of foul behaviors, providing a more objective and accurate basis for penalty judgment for football matches, reducing interference from human factors, and improving the fairness and viewing quality of the match.

[0040] Furthermore, by setting up a target monitoring device to monitor and track the positions and movement trajectories of the contestants and the target football in real time, and taking pictures of the game images when the contestants and the target football are monitored at the same time, the locked shooting module provides a solid foundation for the automatic foul detection system in football matches based on deep learning, which can efficiently capture and transmit game images, provide high-quality data support for subsequent foul detection, and ensure the efficient operation and accurate judgment of the entire system.

[0041] Furthermore, the game image is filtered using a pixel threshold through a filtering device, and the filtered image is cropped to a standard size, the action features of the contestants and the position information of the target football are extracted, the corresponding game information is generated, the effective information in the game image is screened, and irrelevant content is removed to ensure that the cropped image can be correctly processed by the model, thereby realizing the analysis of the contestants' actions and the football's position.

[0042] Furthermore, by selecting features related to football interaction, preprocessing game information, and using a learning device to generate action judgment results, the complete process of feature selection, data preprocessing, and model learning lays the foundation for subsequent learning and judgment, provides reliable data input for the learning device, improves the quality of data and the learning efficiency of the model, and helps to improve the accuracy of action judgment.

[0043] Furthermore, the shooting is triggered by the target monitoring device to focus on the key game situations in the defensive area, and the filtering device filters the image through the pixel threshold to ensure that the image quality meets the requirements of subsequent processing. It not only improves the targeted image acquisition, but also reduces invalid data through the screening mechanism, providing a high-quality data foundation for subsequent image processing and analysis.

[0044] Furthermore, by learning the preliminary data of the game situation and setting warning lights to display the action judgment results, the action judgment results are fed back in real time, and the color changes are used to intuitively show whether the contestants have committed a foul. At the same time, the visual device is used to store and display key game images. Computer vision and machine learning technology are used to realize real-time monitoring and judgment of the actions of the contestants in football matches. It can efficiently handle complex game scenes and provide auxiliary decision-making support for referees. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a structural schematic diagram of a single adjudication device in the automatic foul detection system for football matches based on deep learning of the present invention;

[0046] Figure 2 This is a structural diagram of the locking camera module of the present invention;

[0047] Figure 3 It is a structural schematic diagram of the screening processing module of the present invention;

[0048] Figure 4 It is a structural schematic diagram of the action prediction module of the present invention. DETAILED DESCRIPTION

[0049] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0050] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0051] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0052] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0053] See also Figure 1 As shown, it is a structural schematic diagram of a single adjudication device in the automatic foul detection system for football matches based on deep learning of the present invention, including:

[0054] A locking shooting module is used to lock the contestants and the target football. When the contestants and the target football appear at the same time, the locking shooting module shoots the game image;

[0055] A screening and processing module, which is connected to the locking and shooting module, is used to screen the game images and generate corresponding game information;

[0056] The action prediction module is connected to the screening and processing module, and is used to select a number of learning features according to the game information, pre-process the game information, generate corresponding game prediction data, use the action learning model to learn the game prediction data, and generate corresponding action judgment results;

[0057] A decision module, which is connected to the action prediction module and is used to make a decision on the action judgment result. When the action judgment result is a collision, the decision module determines that the contestant has committed a foul. When the action judgment result is no collision, the decision module determines that the contestant has not committed a foul.

[0058] Among them, the action learning model is trained and generated based on the preliminary data of the game situation;

[0059] The result of the action judgment is the collision between the contestant's arm and the target football.

[0060] By setting a number of equally spaced decision devices, a single decision device includes: a locking shooting module, which is used to shoot a game image when a contestant and a target football appear at the same time; a screening processing module, which is used to generate corresponding game information; an action prediction module, which is used to select a number of learning features, pre-process the game information, and generate an action judgment result using an action learning model; a decision module, which is used to make a decision on the action judgment result; when the action judgment result is a collision, the decision module determines that the contestant has committed a foul; when the action judgment result is no collision, the decision module determines that the contestant has not committed a foul, thereby realizing automatic detection and judgment of foul behaviors, providing a more objective and accurate basis for judgment and penalty for football matches, reducing the interference of human factors, and improving the fairness and viewing of the game.

[0061] See also Figure 2 As shown, it is a structural schematic diagram of the locking shooting module of the present invention, including:

[0062] A target monitoring device, which is used to monitor and track the position and movement trajectory of the players and the target football in real time;

[0063] A camera connected to the target monitoring device is used to capture images of the game while monitoring the players and the target football at the same time;

[0064] The transmission device is connected to the camera and is used to store the game situation images and transmit the game situation images to the screening and processing module.

[0065] The target monitoring device is the core part of the lock-on shooting module, which is used to monitor and track the position and movement trajectory of the contestants and the target football in real time. It needs to have high precision and high real-time performance to ensure that the key moments in the game can be accurately captured. In specific implementation, the target monitoring device uses multiple image sensors to monitor the position of the contestants and the target football in real time through computer vision technology or target detection and tracking algorithms.

[0066] Object detection and tracking algorithms:

[0067] Use deep learning algorithms (such as YOLOv8) and adopt improved network structures (such as CSP structure and SPPF module) to improve the performance of the algorithm.

[0068] The camera is connected to the target monitoring device, and when it monitors the contestants and the target football at the same time, it captures the game image. In the specific implementation, the camera needs to have high resolution and high frame rate to capture clear images. The camera can flexibly capture the game scene by combining fixed and mobile perspectives. When the target monitoring device detects that the contestants and the football appear in the field of view at the same time, the camera is triggered to shoot. The camera can automatically adjust the shooting angle and focal length according to the position information provided by the target monitoring device to ensure that the best perspective is captured.

[0069] The transmission device is used to store the game images and transmit the image data to the screening and processing module. It needs to have the ability of high-speed transmission and large-capacity storage to ensure the real-time and integrity of the data.

[0070] In the specific implementation, a local server or cloud storage system is used to store the captured game images, support fast data reading and backup, and ensure the reliability of the system. High-speed transmission technology (such as 5G or UWB) is used to transmit image data to the screening and processing module in real time. The transmission device needs to have the characteristics of low latency and high bandwidth to meet the requirements of real-time processing.

[0071] By setting up a target monitoring device to monitor and track the position and movement trajectory of the participants and the target football in real time, and taking pictures of the game images when the participants and the target football are monitored at the same time, the locked shooting module provides a solid foundation for the automatic foul detection system in football matches based on deep learning. It can efficiently capture and transmit game images, provide high-quality data support for subsequent foul detection, and ensure the efficient operation and accurate judgment of the entire system.

[0072] See also Figure 3 As shown, it is a schematic diagram of the structure of the screening processing module of the present invention, including:

[0073] A filtering device, which is used to filter the game image using a pixel threshold to form a corresponding filtered image;

[0074] A cropping device, connected to the filtering device, for cropping the filtered image to a standard size;

[0075] An extraction device, connected to the cutting device, is used to extract the action features of the contestants and the position information of the target football, and generate corresponding game information;

[0076] The pixel threshold is the minimum standard for the pixels of the game image, and is related to the straight-line distance between the camera and the defense area of ​​the game field;

[0077] The standard size is the size that is suitable for the action learning model;

[0078] The motion features include the arm position information of the contestant, the relative position of the contestant's arm and the target football, and the extension direction and angle of the arm.

[0079] In the specific implementation, the pixel threshold is used to filter the game image to generate a filtered image. The pixel threshold is a key parameter and is the minimum standard for the pixels of the game image. It is related to the straight-line distance between the camera and the defensive area of ​​the playing field. The longer the distance, the higher the pixel threshold is required to ensure image quality. The pixel threshold is dynamically adjusted according to the actual distance between the camera and the field to adapt to different shooting conditions.

[0080] The cropping device is connected to the filtering device and is used to crop the filtered image to a standard size. The purpose of cropping is to adjust the image to a size suitable for subsequent processing and ensure that the cropped image can be correctly processed by the model.

[0081] The game image is filtered by the filtering device using pixel threshold, and the filtered image is cropped to a standard size, the action features of the contestants and the position information of the target football are extracted, the corresponding game information is generated, the effective information in the game image is screened, and the irrelevant content is removed to ensure that the cropped image can be correctly processed by the model, thereby realizing the analysis of the contestants' actions and the football's position.

[0082] See also Figure 4 As shown, it is a schematic diagram of the structure of the action prediction module of the present invention, including:

[0083] A selection device is used to select corresponding learning features according to the game information, wherein:

[0084] The learned features include the relative position and distance between the participant’s arm and the target football;

[0085] A preprocessing device, connected to the selection device, for preprocessing the game information and generating corresponding game pre-data;

[0086] The learning device is connected to the preprocessing device and is used to learn the pre-match data, generate corresponding action judgment results, and mark the action judgment results according to the learning characteristics.

[0087] The learning device generates prediction results of the contestants' actions by learning the pre-game data, and marks the action judgment results according to the learning characteristics for subsequent analysis and verification. In the specific implementation, the learning device analyzes the pre-game data based on machine learning algorithms (such as decision trees, support vector machines) or deep learning algorithms (such as convolutional neural networks, recurrent neural networks) and generates prediction results.

[0088] The relative position and relative distance between the contestant's arm and the target football are the key to motion prediction. The relative position can describe the geometric relationship between the contestant's arm and the target football, while the relative distance provides quantitative information.

[0089] By selecting features related to football interaction, preprocessing game information, and using a learning device to generate action judgment results, the complete process of feature selection, data preprocessing, and model learning lays the foundation for subsequent learning and judgment, provides reliable data input for the learning device, improves data quality and model learning efficiency, and helps to improve the accuracy of action judgment.

[0090] Specifically, the locking shooting module is provided with a single camera, and when the target monitoring device simultaneously monitors the contestants and the target football, a shooting instruction is issued to the camera;

[0091] Among them, the camera is facing the defense area.

[0092] Specifically, the filtering device obtains image pixels of the game image, wherein:

[0093] When the image pixel is greater than the pixel threshold, the corresponding game image is retained, and when the image pixel is less than the pixel threshold, the filtering device filters the corresponding game image.

[0094] In the specific implementation, the locked shooting module is equipped with a camera that faces the defensive area, ensuring that the camera can focus on the game situation in the defensive area, such as the players' defensive movements and interactions with the football.

[0095] The filtering device obtains the image pixels of the game image, that is, the resolution or pixel density of the image.

[0096] The role of pixel threshold:

[0097] Quality Control: Pixel threshold is used to ensure that the image quality meets the requirements of subsequent processing. Low-pixel images may contain too much noise or insufficient details, affecting subsequent feature extraction and analysis.

[0098] Relevance screening: Through pixel threshold screening, background images or blurred images that are not related to the game can be removed to improve the validity of the data.

[0099] The shooting is triggered by the target monitoring device, focusing on the key game situations in the defensive area. The filtering device filters the image through the pixel threshold to ensure that the image quality meets the requirements of subsequent processing. It not only improves the targeted image acquisition, but also reduces invalid data through the screening mechanism, providing a high-quality data foundation for subsequent image processing and analysis.

[0100] Specifically, the cropping device identifies the contestants and the target football and performs optimized protection, crops the filtered image to a standard size, and the extraction device extracts the standard coordinates of the contestants' arms and the target football based on the motion characteristics and position information, and integrates the standard coordinates into corresponding game information.

[0101] In a specific implementation, the function of the cropping device is to identify the contestants and the target football, and to crop the filtered image to a standard size. During this process, the cropping device will optimize the protection of key targets (contestants and target football) to ensure that the cropped image can meet the needs of subsequent processing. For example, using an object detection algorithm (such as YOLO) can achieve accurate identification of contestants and footballs, and determine the cropping area through a bounding box.

[0102] The extraction device extracts the standard coordinates of the players' arms and the target football based on the motion features and position information, and integrates these coordinates into the game information. This process can be achieved through deep learning models (such as YOLOv5, YOLOv8), which can identify key points in the image and extract relevant position information. For example, the YOLO model can detect the bounding boxes of players and footballs and further extract their coordinate information.

[0103] Specifically, the preprocessing device cuts the game information according to the standard sampling rate to generate a number of game pre-data with the standard sampling rate, and the action learning model learns the game pre-data and generates corresponding action judgment results;

[0104] The standard sampling rate is a sampling rate that can be recognized by the action learning model, and for a single sampling, the corresponding standard sampling rate is a single sampling rate.

[0105] In specific implementation, when the standard sampling rate is 1700 per second, the system has the best analysis effect on the game situation data.

[0106] The action learning model learns the pre-processed data of the competition and generates the corresponding action judgment results. This process can be achieved through machine learning or deep learning algorithms, such as support vector machines (SVM) or convolutional neural networks (CNN). By learning the pre-processed data, the model can determine whether the actions of the contestants meet specific rules.

[0107] Specifically, when the action judgment result is no collision, the adjudication module determines that the contestant has not violated the rules and keeps the warning light green in response to the action judgment result.

[0108] Specifically, when the action judgment result is a collision, the adjudication module determines that the contestant has committed a foul, the warning light is adjusted to red in response to the action judgment result, and the corresponding game image is uploaded to the visual device.

[0109] By learning the predicted match data and setting warning lights to display the action judgment results, the action judgment results are fed back in real time, and the color changes are used to intuitively show whether the participants have committed a foul. At the same time, the visual device is used to store and display key match images. It uses computer vision and machine learning technology to achieve real-time monitoring and judgment of the actions of participants in football matches. It can efficiently handle complex game scenes and provide auxiliary decision-making support for referees.

[0110] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0111] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An automatic foul detection system for football matches based on deep learning, characterized in that: The system is provided with a plurality of equally spaced decision devices, and a single decision device includes: A locking shooting module is used to lock the contestants and the target football. When the contestants and the target football appear at the same time, the locking shooting module shoots the game image; A screening processing module, connected to the locking and shooting module, for screening the game situation images and generating corresponding game situation information; An action prediction module, which is connected to the screening and processing module, is used to select a number of learning features according to the game information, pre-process the game information, generate corresponding game prediction data, use the action learning model to learn the game prediction data, and generate corresponding action judgment results; A decision module, which is connected to the action prediction module and is used to make a decision on the action judgment result. When the action judgment result is a collision, the decision module determines that the contestant has committed a foul. When the action judgment result is a non-collision, the decision module determines that the contestant has not committed a foul. Wherein, the action learning model is generated through training based on the game situation prediction data; The action judgment result is the collision condition between the contestant's arm and the target football.

2. The deep learning-based automatic football match foul detection system according to claim 1 is characterized in that: The locking and shooting module comprises: A target monitoring device, which is used to monitor and track the position and movement trajectory of the contestants and the target football in real time; A camera connected to the target monitoring device, for capturing the game image when the contestants and the target football are monitored simultaneously; A transmission device is connected to the camera and is used to store the game image and transmit the game image to the screening and processing module.

3. The automatic football match foul detection system based on deep learning according to claim 2 is characterized in that: The screening processing module comprises: A filtering device, which is used to filter the game image using a pixel threshold to form a corresponding filtered image; A cutting device, connected to the filtering device, for cutting the filtered image to a standard size; An extraction device, connected to the cutting device, for extracting the action features of the contestants and the position information of the target football, and generating corresponding game information; The pixel threshold is the minimum standard of the pixels of the game image, and is related to the straight-line distance between the camera and the defense area of ​​the game field; The standard size is a size that is compatible with the action learning model; The action features include the arm position information of the contestant, the relative position of the contestant's arm and the target football, and the extension direction and angle of the arm.

4. The deep learning-based automatic football match foul detection system according to claim 3 is characterized in that: The action prediction module comprises: A selection device is used to select corresponding learning features according to the game information, wherein: The learning features include the relative position and relative distance between the arm of the contestant and the target football; A preprocessing device, connected to the selection device, for preprocessing the game information and generating corresponding game pre-data; A learning device is connected to the preprocessing device and is used to learn the pre-match data, generate corresponding action judgment results, and mark the action judgment results according to the learning characteristics.

5. The deep learning-based automatic football match foul detection system according to claim 4 is characterized in that: The locking shooting module is provided with a single camera, and when the target monitoring device simultaneously monitors the contestant and the target football, a shooting instruction is issued to the camera; Wherein, the camera faces the defense area.

6. The deep learning-based automatic football match foul detection system according to claim 5 is characterized in that: The filtering device obtains image pixels of the game image, wherein: When the image pixel is greater than the pixel threshold, the corresponding game image is retained, and when the image pixel is less than the pixel threshold, the filtering device filters the corresponding game image.

7. The deep learning-based automatic football match foul detection system according to claim 6 is characterized in that: The cropping device identifies the contestant and the target football and performs optimized protection, crops the filtered image to the standard size, and the extraction device extracts the standard coordinates of the contestant's arm and the target football according to the action characteristics and the position information, and integrates the standard coordinates into corresponding game information.

8. The deep learning-based automatic foul detection system for football matches according to claim 7 is characterized in that: The preprocessing device cuts the game information at a standard sampling rate to generate a number of game pre-data with a sampling rate of the standard sampling rate, and the action learning model learns the game pre-data and generates a corresponding action judgment result; The standard sampling rate is a sampling rate that can be recognized by the action learning model, and for a single sampling, the corresponding standard sampling rate is a single sampling rate.

9. The deep learning-based automatic football match foul detection system according to claim 8 is characterized in that: When the action judgment result is no collision, the adjudication module determines that the contestant has not violated the rules, and keeps the warning light green in response to the action judgment result.

10. The deep learning-based automatic foul detection system for football matches according to claim 9 is characterized in that: When the action judgment result is a collision, the adjudication module determines that the contestant has committed a foul, the warning light is adjusted to red in response to the action judgment result, and the corresponding game image is uploaded to the visual device.

Images