Football goalkeeper penalty-kick fighting real-time training method, system and device

Abstract

The invention provides a real-time training method, system and device for penalty kick fighting of a football goalkeeper. The method comprises the following steps: acquiring penalty kick shooting data; processing the penalty shot data based on a preset prediction model to obtain target drop point information corresponding to the penalty shot data; and displaying the target drop point information to a goalkeeper in a visual mode. According to the invention, the body movement of a penalty player, the initial speed of a football, the ball outlet angle and other information are captured in real time through the camera, the flight path and the goal drop point area of the football are rapidly predicted based on the physical movement law and the deep learning algorithm, and the prediction result is displayed in real time through the display. The penalty kick prediction method improves accuracy of penalty kick prediction, can adapt to complex scenes, can optimize model parameters according to kicking characteristics of different players, is suitable for professional team training, and can also be popularized in football schools and amateur training scenes.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a real-time training method, system and device for penalty kick saving for a soccer goalkeeper. Background Art

[0002] Goalkeeper training for penalty kicks has always been a challenging aspect of soccer. Goalkeepers must quickly identify the kicker's movements and determine the ball's trajectory and landing zone in order to save the shot. However, the speed of a penalty kick typically reaches 20 to 35 meters per second, and this, combined with the goalkeeper's reaction time (usually less than 0.5 seconds), results in a low save success rate.

[0003] Existing penalty kick training methods have the following shortcomings: (1) Traditional training methods are limited: Currently, goalkeepers rely heavily on coaching, making it difficult to improve their decision-making abilities through quantitative data. (2) Lack of real-time analysis tools: During training, it is impossible to analyze the penalty kick player's movements, ball speed, and trajectory data in real time. (3) Inadequate adaptation to complex scenarios. (4) Large errors in the prediction of the landing point: Systems that rely solely on player movement or trajectory modeling ignore complex factors such as ball rotation and airflow interference, resulting in insufficient prediction accuracy.

[0004] An existing soccer penalty analysis method uses a server to obtain input from the target attacker's historical penalty record information to determine penalty training. The analysis module identifies the shooting method and actual shooting location from penalty strategy information or historical penalty record information and establishes a correlation between the two. The shooting method includes the target attacker's initial standing position, the trajectory and body movement information leading to the ball's position, and the ball's speed and trajectory changes upon reaching the goal, thereby determining the target attacker's estimated input position. While this solution can predict penalties, it relies on predefined "action key indicators" and "reference indicator groups" and extracts body movement parameters for analysis. This static analysis method is dependent on the quality of historical records and rule design and lacks adaptive capabilities. Summary of the Invention

[0005] The first purpose of the present invention is to provide a real-time penalty kick saving training method for a soccer goalkeeper that can adapt to more complex scenarios and improve the defensive training effect.

[0006] The second object of the present invention is to provide a system for implementing the above-mentioned real-time training method for soccer goalkeeper penalty kick saves.

[0007] The third object of the present invention is to provide a device for implementing the above-mentioned real-time training method for soccer goalkeeper penalty kick saves.

[0008] In order to achieve the above-mentioned first purpose, the present invention provides a real-time training method for penalty kick saves for a football goalkeeper, which includes the following steps: obtaining penalty kick shot data; processing the penalty kick shot data based on a preset prediction model to obtain target landing point information corresponding to the penalty kick shot data; and displaying the target landing point information to the goalkeeper in a visual manner.

[0009] As can be seen from the above scheme, the present invention predicts the landing point of penalty kicks by obtaining real-time shooting data and displays the target landing line information to the goalkeeper in a visual manner, thereby improving the accuracy of penalty kick predictions, being able to adapt to complex scenarios, and optimizing model parameters according to the kicking characteristics of different players.

[0010] A further solution is that the penalty kick data includes real-time shot video data; the penalty kick data is processed based on a preset prediction model to obtain the target landing point information corresponding to the penalty kick data, including: extracting the key body postures of the shooting player from the real-time shot video data, and combining the three-dimensional convolutional neural network to determine the real-time action features and real-time football status of the shooting player; the real-time action features, real-time football status and real-time shooting video data are fused through a cross-modal attention mechanism to obtain fused features; the fused features are processed based on a model combining physical dynamics and deep learning, and the target landing point information is output.

[0011] It can be seen that the prediction accuracy can be further improved by fusing multimodal data for prediction.

[0012] A further solution is that the penalty kick data also includes historical shot data; real-time action features, real-time football status, real-time shot video data, and historical shot data are fused through a cross-modal attention mechanism to obtain fused features.

[0013] This shows that combining historical shot data for prediction further improves prediction accuracy. For the application scenario of soccer shot trajectory prediction, a cross-modal attention mechanism is used to fuse real-time action features, real-time soccer status, real-time shot video data, and historical shot data. Optimizations are made in the fusion strategy and real-time assurance, improving feature expression capabilities and prediction accuracy in complex dynamic environments.

[0014] A further solution is that the target landing point information represents the area with the highest probability among the six goal landing point areas predicted by the Softmax classifier.

[0015] A further solution is to display the target landing point information in a visual manner, including: prompting the area with the highest probability through the display area of ​​the display, and each display area corresponds to a different goal area.

[0016] In order to achieve the above-mentioned second purpose, the present invention provides a real-time penalty kick saving training system for a football goalkeeper, which includes: a trajectory prediction module, a data acquisition module, and an area display module; the data acquisition module is used to obtain penalty kick shooting data, the trajectory prediction module is used to process the penalty kick shooting data through a preset prediction model to obtain target landing point information corresponding to the penalty kick shooting data; the area display module is used to display the target landing point information to the goalkeeper in a visual manner.

[0017] In order to achieve the above-mentioned third purpose, the present invention provides a real-time training device for football goalkeeper penalty kick saves, which includes: a processor, a camera, and a display, and the processor is connected to the camera and the display respectively; the processor is used to execute the above-mentioned real-time training method for football goalkeeper penalty kick saves.

[0018] As can be seen from the above scheme, the present invention uses a high-speed camera to capture the penalty kick player's body movements (limb joints, kicking posture), the ball's initial speed, release angle, and other information in real time. Based on the laws of physical motion and a deep learning algorithm, it rapidly predicts the ball's flight trajectory and goal impact zone, and displays the prediction results in real time on a display. This improves the accuracy of penalty kick predictions, adapts to complex scenarios, and optimizes model parameters based on the kicking characteristics of different players. This method is suitable for professional team training and can also be promoted in football schools and amateur training settings.

[0019] A further solution is a motion capture sensor, which is connected to a processor.

[0020] This shows that the accuracy of capturing the body movements of penalty kick players can be improved.

[0021] A further solution is that the display is arranged facing the goal area, and each display area on the display corresponds to a different goal area.

[0022] This shows that it is convenient for goalkeepers to check in real time and determine the predicted landing point, thereby improving the effectiveness of defensive training.

[0023] A further solution is that the cameras include a shot recording camera, a goalkeeper take-off time camera, and a goalkeeper save camera.

[0024] It can be seen that through the accumulation of training data by the system, the model parameters can be adjusted according to the kicking characteristics of different players to generate personalized training plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a structural framework diagram of the first embodiment of the real-time penalty kick saving training device for a soccer goalkeeper of the present invention.

[0026] Figure 2This is another structural framework diagram of the first embodiment of the soccer goalkeeper penalty save real-time training device of the present invention.

[0027] Figure 3 1 is a schematic diagram of a system framework of an embodiment of a soccer goalkeeper penalty kick saving system according to the present invention.

[0028] Figure 4 The present invention is a flowchart of an embodiment of a real-time training method for soccer goalkeeper penalty kick saves.

[0029] Figure 5 yes Figure 4 Specific flow chart of step S12 in FIG.

[0030] Figure 6 It is a schematic structural framework diagram of the second embodiment of the real-time penalty kick saving training device for a soccer goalkeeper of the present invention.

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments. DETAILED DESCRIPTION

[0032] The present invention captures the body movements of the penalty kick player when shooting and the initial information of the penalty kick, calculates the trajectory and predicts the landing point, and displays the prediction results to the goalkeeper in real time, thereby assisting the football goalkeeper in defensive training.

[0033] First embodiment of a real-time training device for soccer goalkeeper penalty kick saves: See also Figure 1 The soccer goalkeeper penalty kick saving real-time training device of this embodiment includes a data camera 1, a processor 2, and a display 3. The processor 2 is connected to the camera 1 and the display 3 respectively.

[0034] The specific connection mode between the processor 2, the camera 1 and the display 3 can be a wired connection or a wireless connection.

[0035] Camera 1 is used to collect real-time shot video data. Processor 2 is used to obtain penalty kick data and, based on a preset prediction model, process the penalty kick data to predict the target landing point information corresponding to the current shot. The penalty kick data includes real-time shot video data and historical shot data. Display 3, under the control of processor 2, visually displays the target landing point information.

[0036] See also Figure 2 The camera 1 includes several shot recording cameras 11 , and the display module 3 includes a display screen 31 .

[0037] The shot recording cameras 11 are arranged on both sides of the penalty kick location to collect real-time shot video data during the penalty kick process. The real-time shot video data includes the movement of the ball 5 and the penalty kick player 6. The shot recording cameras 11 of this embodiment are preferably high-speed cameras with a frame rate of 1000 FPS.

[0038] The display screen 31 is positioned directly opposite the goal 4, and displays the target landing point information corresponding to the current shot to the goalkeeper in real time. Specifically, the display screen 31 can be provided with six display areas, each corresponding to the six goal areas divided in front of the goal 4. The first display area corresponds to the upper left goal area, the second display area corresponds to the upper middle goal area, the third display area corresponds to the upper right goal area, the fourth display area corresponds to the lower left goal area, the fifth display area corresponds to the lower middle goal area, and the sixth display area corresponds to the lower right goal area. In other embodiments, the six goal areas can also be numbered, and the display screen 31 displays the target landing point information with the specific number.

[0039] When used, the processor 2 analyzes the real-time shooting video data collected by the shooting recording camera 11 in real time, extracts the key body postures of the shooting player from the real-time shooting video data, determines the action characteristics of the shooting player and the initial state of the football in combination with the three-dimensional convolutional neural network, and processes the action characteristics and the initial state of the football based on a model combining physical dynamics and deep learning to output target landing point information. The target landing point information is used to control the display of the corresponding target display area on the display screen 31, so that the goalkeeper can determine which goal area to defend after checking.

[0040] Example of a soccer goalkeeper penalty kick saving system: This embodiment runs on the processor of the above-mentioned soccer goalkeeper penalty kick saving device embodiment, wherein the processor can specifically be a processor in a server, a personal computer or other device.

[0041] See also Figure 3 The soccer goalkeeper penalty kick saving system includes a data acquisition module 10 , a trajectory prediction module 20 and an area display module 30 .

[0042] The data acquisition module 10 is used to acquire historical shot data and real-time shot video data from a shot recording camera to obtain penalty kick data. The trajectory prediction module 20 is used to process penalty kick data using a preset prediction model to predict the ball's trajectory in real time, obtaining target landing point information corresponding to the penalty kick data. The area display module 30 is used to control the display area on the display screen to indicate to the goalkeeper the goal area corresponding to the target landing point information. This indication is provided visually, such as by controlling the corresponding display area to display specific text or a specific color, or by controlling the corresponding display area to highlight.

[0043] Example of a real-time training method for soccer goalkeeper penalty kick saves: This embodiment is based on the above-mentioned embodiment of the soccer goalkeeper penalty save device and is implemented by executing a computer program on a processor. Figure 4 , the processor includes the following execution steps: S11: Obtain penalty kick data.

[0044] Among them, penalty kick data includes historical shot data and real-time shot video data obtained from the camera.

[0045] Historical shot data includes the following player-related data, football movement data, goalkeeper-related data and game scene data.

[0046] Player-related data includes: (1) Historical penalty kick records of the shooter: including data such as the direction, speed, angle, and landing point distribution of the shot, which is used to analyze the player's shooting habits and tendencies. (2) Statistics of the shooter's action characteristics: including historical statistical values ​​of dynamic characteristics such as starting method, stride length, body center of gravity change, and hitting action. (3) Player's psychological state and performance: including records of factors such as psychological pressure in the game, success rate, and key game performance.

[0047] Football sports data includes: (1) Penalty kick trajectory data: including the flight path, arc trajectory, rotation characteristics (such as rotation direction and rotation speed) and the final landing point of the ball. (2) The impact of the environment on the ball trajectory: data on the impact of the external environment (wind speed, humidity, etc.) on the penalty kick trajectory in historical matches.

[0048] Goalkeeper-related data include: (1) Goalkeeper’s historical defensive performance: (2) including save direction, reaction time, movement distance, success rate, etc. (3) Goalkeeper’s preference and weakness analysis: such as whether to save on the left or right side, and the effectiveness of responding to specific shot types (such as curve balls, high-speed direct shots).

[0049] Match scene data includes: (1) Match venue related information: such as the historical records of venue information such as goal position deviation, turf condition, lighting environment, etc. (2) Penalty kick data at key moments of the match: statistics of penalty kick results in specific matches (such as finals, overtime), and analysis of feature changes in key scenes.

[0050] In different embodiments, the historical shot data may vary. Specifically, in different embodiments, the aforementioned player-related data, football motion data, goalie-related data, and match scenario data may be combined according to actual needs. For example, the aforementioned goalie-related data may not be included. More specifically, for a particular type of data, different data types may be selected. For example, the player-related data may not include the aforementioned player's psychological state and performance.

[0051] Historical shot data plays the following important roles in dynamic prediction and decision-making (historical shot data is processed, high-quality, prior data that takes up little space and is stored in the system cache, without affecting system performance and real-time processing capabilities): 1. Provide long-term background information (context) for the model: Historical shooting data contains regular patterns and tendencies of shooting players, goalkeepers, game scenes, etc., which cannot be directly obtained from real-time shooting video data. For example: (1) The probability that a certain shooting player tends to shoot to a certain area (such as tending to shoot to the upper left corner of the goal). (2) The performance of a certain goalkeeper in saving a specific type of shot (such as reacting slowly to a high-speed direct shot). (3) The success rate and psychological performance of the shooting player under specific conditions (such as the final) in historical games.

[0052] 2. Compensating for the shortcomings of real-time data: Although real-time shooting video data is dynamic and accurate, it may have limitations: (1) One-sidedness of real-time shooting video data: Real-time shooting video data of the current scene can only reflect the current state and cannot describe the historical behavior of the shooting player or goalkeeper. (2) Noise and anomalies: Real-time shooting video data may be affected by environmental interference (such as light, angle, sensor error, etc.). Historical shooting data can help the model distinguish abnormal situations by supplementing the global pattern.

[0053] 3. Enhance model robustness (generalization): During the deep learning model's prediction process, historical shot data can help improve the model's generalization capabilities in complex scenarios. When real-time shot video data contains uncertainty (such as low-resolution video or sensor anomalies), historical shot data can serve as a stable reference for the model, reducing misjudgments. The model can combine real-time shot video data with historical shot data to generate more balanced and reasonable prediction results.

[0054] 4. Support for personalized predictions: Each player's shooting habits, footwork, and other characteristics have their own unique patterns. Relying solely on real-time shooting video data may not be able to identify these individual characteristics. However, historical shooting data can help the model generate personalized defensive strategies for specific players. For example, if a player habitually uses a "fake move" to trick the goalkeeper into moving and then shoots in the opposite direction, the model can use historical shooting data to identify this pattern in advance. Even if the "fake move" has not yet been completed in the real-time shooting video data, it can still predict the player's intention in advance.

[0055] 5. Improve the accuracy of the probability distribution of landing points: Historical shot data can be used as prior information for probability distribution to optimize prediction results.

[0056] 6. Dynamic Adaptation to Different Scenarios: Real-time shot video data may vary depending on the environment (such as the playing field, weather conditions, etc.), while historical shot data provides more stability and a global perspective. This is particularly important in the following situations: (1) New scenarios, new players: When real-time shot video data is insufficient to represent the current scenario, historical data can provide additional support. (2) Real-time model updates: Incorporating historical video data allows the model to adapt more quickly to new environments or new opponents.

[0057] S12: Processing the penalty kick data through the prediction model to obtain target landing point information corresponding to the penalty kick data.

[0058] S13: Displaying target landing information to the goalkeeper in a visual manner.

[0059] In the above step S12, the prediction model processes the penalty kick data, see Figure 5 , specifically including: S121: Extracting key body postures of the shooting player from the real-time shooting video data in the penalty kick shooting data.

[0060] Among them, the existing algorithms such as OpenPose can be used to detect the key body postures of the shooting player in real time.

[0061] S122: Determine the real-time action features of the shooting player and the real-time state of the ball based on a three-dimensional convolutional neural network.

[0062] By processing the body key point postures and real-time shooting video data through a trained three-dimensional convolutional neural network, we can obtain the real-time action features of the shooting player and the real-time state of the ball. The real-time action features describe the actions of the shooting player during the shooting process, and the real-time state of the ball describes the movement state of the ball, including speed, position, etc.

[0063] The three-dimensional convolutional neural network of this embodiment is obtained through a model training method. Its input is the camera video stream and the key point postures of the shooting player's limbs. Its output is the real-time action features of the goal player and the real-time status of the football. Training is performed by setting a training set and a test set.

[0064] S123: Real-time action features, real-time football status, real-time shot video data, and historical shot data are fused through a cross-modal attention mechanism to obtain fused features.

[0065] Among them, the real-time action features, real-time football status, real-time shooting video data, and historical shooting data are fused through the cross-modal attention mechanism. The specific implementation steps of obtaining the fused features include feature alignment, attention mechanism, and fused feature generation.

[0066] First, the input real-time action features, real-time football status, real-time shot video data and historical shot data are aligned between modalities, including time synchronization and spatial dimension mapping. During the time synchronization process, the data of each modality is timestamped to ensure that the time starting point of all data is consistent. Interpolation or resampling technology is used to convert data with different time resolutions to a unified time scale; in the spatial dimension mapping process, the spatial representation of each modality is standardized into a unified coordinate or dimension. For example, the spatial position information in the video frame (such as the coordinates of limb key points) is mapped to a unified spatial coordinate system.

[0067] A temporal dynamic attention mechanism is then introduced to calculate attention weights. This adjusts the weights of each modality based on intermodal correlations and real-time changes. Specifically, a multi-head self-attention mechanism is used to capture the interactions between modalities, while an LSTM network dynamically adjusts the weight coefficients of each modality. For example, at the moment of a shot, a higher weight might be assigned to the live shot video data to capture the immediate action details.

[0068] Finally, fusion features are generated. The corresponding feature vectors of each modality after alignment and weight adjustment are input into the fusion module. The fusion module multiplies the feature vectors by their corresponding attention weights to fuse them and generate a joint feature vector representing the spatiotemporal trajectory of the football shot, namely the fusion feature, which provides comprehensive data support for subsequent landing point prediction.

[0069] Among them, the fusion features include the action features of the shooting player and the initial state of the football . Movement characteristics of the shooting player It is used to describe the dynamic information of the player who shoots the ball, including the position and movement trajectory of the player's limb key points, such as the toes, knees, and hips, as well as the kicking action pattern, such as the kicking force, the change of the body's center of gravity, etc., and the historical action characteristics, which represent the player's kicking habits in previous training or games. Used to describe the important physical characteristics of the ball when it is just kicked, including: the initial speed of the ball ( ), used to describe the initial shooting speed of the ball; the shooting angle ( ), used to describe the direction angle of the player kicking the ball; rotation information ( ), which is used to describe the ball’s rotation rate and direction. The initial state of the ball provides the starting condition for the entire trajectory prediction.

[0070] S124: A model based on the combination of physical dynamics and deep learning processes fusion features and outputs target landing point information.

[0071] Among them, the model based on the combination of physical dynamics and deep learning is as follows: ,in, Indicates the prediction result, specifically the time The flight state or position of the penalty kick at that moment (e.g., the ball's spatial coordinates, speed, etc.). In penalty kick landing point prediction, this may be further mapped to a certain goal area. It represents a model based on the combination of physical dynamics and deep learning, which is used to describe the dynamic characteristics of penalty kick trajectories. Represents the parameters of the model, such as the weights and biases of a neural network. They can be optimized using training data.

[0072] The model based on the combination of physical dynamics and deep learning calculates the ball's flight trajectory based on the dynamic equation, integrating airflow resistance and ball rotation: ,in, represents gravity, represents air resistance, represents the lift caused by rotation.

[0073] In the model based on the combination of physical dynamics and deep learning, the deep model is specifically used to process the complex trajectory deviations during football movement.

[0074] Complex trajectory deviation refers to the fact that when predicting the landing point of a penalty kick, the trajectory may show nonlinear and irregular deviations due to the influence of various factors during the penalty kick movement (such as shooting power, shooting angle, ball rotation, wind speed, player psychological state, etc.). This deviation makes the prediction of the landing point of the penalty kick more complicated, especially when it is necessary to judge the landing point in real time and accurately. In penalty kick prediction, complex trajectory deviation may include: (1) nonlinear effects in the ball movement: for example, side spin or arc ball, which causes the trajectory to be curved instead of straight line. (2) external environmental influences: wind speed, humidity, etc. may further change the trajectory of the ball. (3) player factors: the shooting power and habits of different players will also lead to trajectory differences. Deep learning models (such as Transformer or LSTM) can well capture these complex time series features, thereby improving the prediction ability of trajectory deviation.

[0075] This model, based on the integration of physical dynamics and deep learning, utilizes deep learning algorithms to learn and simulate the trajectory changes of objects affected by physical laws. Its training process includes steps such as data collection and preprocessing, model design, training, and validation. By combining physical dynamics equations with deep learning, the model can effectively train and predict when data is scarce or noisy. By incorporating physical laws into model training, the network is trained by minimizing an overall loss function that includes both data error and physical information error terms, ensuring that the model adheres to physical laws while fitting the data. After extensive testing and adjustments, this model, based on the integration of physical dynamics and deep learning, is capable of learning complex patterns such as the ball's rotation pattern and trajectory changes affected by external forces. Parameters can also be optimized based on individual player characteristics (such as power and shooting angles) to enhance personalized prediction accuracy.

[0076] Based on a model that combines physical dynamics with deep learning, a Softmax classifier is used to predict the area where the penalty kick will land: ,in, For the The score of the region.

[0077] For example, the Softmax classifier outputs a probability value for each goal area, reflecting the likelihood that the penalty kick lands in each goal area. For example: (1) the goal area is in the upper left: 40%; (2) the goal area is in the upper middle): 50%; (3) the goal area is in the lower right: 10%.

[0078] Therefore, at the moment of kicking, a model combining physical dynamics and deep learning can output the area with the highest probability of landing and display it on the screen. During this process, TensorRT is used to optimize the model's execution speed, sending the predicted results to the display with a latency of less than 50 milliseconds. Finally, the predicted goal area is displayed on the screen.

[0079] Second embodiment of a real-time training device for soccer goalkeeper penalty kick saves: The difference between this embodiment and the first embodiment of the above-mentioned football goalkeeper penalty save real-time training device is that it also includes a motion capture sensor 7, which is connected to the processor 2 to realize data transmission with the processor 2. Also, the camera also includes a goalkeeper take-off time camera 12 and a goalkeeper save camera 13.

[0080] The motion capture sensor 7 is set at the footsteps position of the penalty kick player, which can specifically be an optical marker point or a sensor for detecting the position of the human body, and is used to extract the body posture and motion parameters of the penalty kick player, such as pace, knee angle, swing amplitude of the hitting foot, etc., to improve the accuracy of the posture of the key points of the shooting player's limbs.

[0081] The goalkeeper take-off timing camera 22 is installed on the side of the goal to capture the time point of the goalkeeper's take-off, the details of the action and the time synchronization with the ball movement. The camera records the goalkeeper's real-time reaction and provides data support for analyzing the goalkeeper's performance.

[0082] Goalkeeper save cameras 13 are placed on both sides of the goal to capture the goalkeeper's saves from different angles, including body posture and movement path. These cameras, combined with deep learning algorithms, can analyze the goalkeeper's technical characteristics and provide improvement suggestions for training.

[0083] In summary, this invention uses a high-speed camera to capture, in real time, information such as the penalty kick player's body movements (limb joints, kicking posture), the ball's initial velocity, and release angle. Based on the laws of physical motion and a deep learning algorithm, it rapidly predicts the ball's flight trajectory and goal impact zone, displaying the prediction results in real time on a display. This improves the accuracy of penalty kick predictions, adapts to complex scenarios, and optimizes model parameters based on the kicking characteristics of individual players. This method is suitable for training professional teams and can also be adopted in football schools and amateur training settings.

[0084] Finally, it should be emphasized that the above 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 changes and modifications. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A real-time penalty kick saving training method for a soccer goalkeeper, characterized in that: The following steps are involved: Get penalty kick data; Processing the penalty kick data based on a preset prediction model to obtain target landing point information corresponding to the penalty kick data; The target landing point information is displayed to the goalkeeper in a visual manner.

2. The soccer goalkeeper penalty kick saving real-time training method according to claim 1, wherein: The penalty kick shooting data includes real-time shooting video data; Processing the penalty kick data based on the preset prediction model to obtain the target landing point information corresponding to the penalty kick data includes: Extracting the key body postures of the shooting player from the real-time shooting video data, and combining it with a three-dimensional convolutional neural network to determine the real-time action features of the shooting player and the real-time state of the ball; The real-time action features, the real-time football state, and the real-time shot video data are fused through a cross-modal attention mechanism to obtain a fused feature; The fusion features are processed based on a model combining physical dynamics and deep learning, and the target landing point information is output.

3. The soccer goalkeeper penalty kick saving real-time training method as claimed in claim 2, wherein: The penalty kick data also includes historical shot data; The real-time action features, the real-time football status, the real-time shot video data, and the historical shot data are fused through a cross-modal attention mechanism to obtain fused features.

4. The soccer goalkeeper penalty kick saving real-time training method according to claim 1, wherein: The target landing point information represents the area with the highest probability among the six goal landing point areas predicted by the Softmax classifier.

5. The soccer goalkeeper penalty kick saving real-time training method as claimed in claim 4, characterized in that: When the target landing point information is displayed in a visual manner, it includes: The area with the highest probability is indicated by a display area of ​​the display, and each display area corresponds to a different goal area.

6. A real-time penalty kick saving training system for a soccer goalkeeper, characterized in that: include: Trajectory prediction module, data acquisition module, and area display module; The data acquisition module is used to obtain penalty kick data, the trajectory prediction module is used to process the penalty kick data through a preset prediction model to obtain target landing point information corresponding to the penalty kick data; the area display module is used to display the target landing point information to the goalkeeper in a visual manner.

7. A real-time penalty kick saving training device for a soccer goalkeeper, characterized in that: include: A processor, a camera, and a display, wherein the processor is connected to the camera and the display respectively; The processor is used to execute the real-time training method for penalty kick saves for a soccer goalkeeper according to any one of claims 1 to 5.

8. The soccer goalkeeper penalty kick saving real-time training system according to claim 7, characterized in that: Also includes: A motion capture sensor is connected to the processor.

9. The soccer goalkeeper penalty kick saving real-time training system according to claim 7, characterized in that: The display is arranged facing the goal area, and each display area on the display corresponds to a different one of the goal areas.

10. The soccer goalkeeper penalty kick saving real-time training system according to claim 7, characterized in that: The cameras include a shot recording camera, a goalkeeper take-off time camera, and a goalkeeper save camera.

Images