Virtual sports event dynamic content configuration and real-time synchronization method and system

Abstract

The invention discloses a virtual sports event dynamic content configuration and real-time synchronization method and system, and relates to the field of dynamic content configuration and real-time synchronization. Acquiring a virtual match video; and obtaining a match target position and a posture feature vector based on the virtual match video. And constructing a Bayesian network. Constructing a time detection network according to the Bayesian network; detecting a match state based on the match target position and the posture feature vector through a time detection network to obtain a key time point; contents such as Banner, news, advertisements and the like are configured at a management end through images, texts, videos and external links, and effective states are set according to key time points; the technical effect that content configuration such as advertisement putting can be more accurately and synchronously carried out at the time point when the key event occurs is achieved.

Description

Technical Field

[0001] This invention relates to the field of dynamic content configuration and real-time synchronization, and more specifically, to a method and system for dynamic content configuration and real-time synchronization of virtual sports events. Background Technology

[0002] Currently, virtual sports events (such as virtual football and virtual horse racing) are essentially processes and results generated by computer programs. Virtual sports events are new forms of exercise that reconstruct traditional sports using digital technology at their core, through virtual reality (VR), augmented reality (AR), artificial intelligence, and other technologies. While humans still need to set initial strategies or intervene in key decisions, algorithms train AI players to perform the actions.

[0003] For long-duration virtual sports events, real-time event streaming allows for real-time prediction. However, accurately predicting the timing of key events and outputting advertisements, news, and banners at those times requires a more accurate network model. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for configuring and synchronizing dynamic content of virtual sports events in real time, so as to solve the above-mentioned problems existing in the prior art.

[0005] In a first aspect, embodiments of the present invention provide a method for configuring and synchronizing dynamic content of virtual sports events in real time, including: Obtain virtual event videos; the virtual event videos are videos of ongoing virtual events; Based on the virtual event video, the target location and posture feature vectors are obtained; the posture feature vectors represent the characteristics of the posture of the personnel at the target location. Construct a Bayesian network; the Bayesian network represents a network that associates location and pose with the probability of key events; Construct a time detection network based on the Bayesian network; By using a time detection network, the status of the competition is detected based on the position and posture feature vectors of the competition target, and key time points are obtained. Based on key time points, generate event display information.

[0006] Optionally, constructing the time detection network based on the Bayesian network includes: The time detection network includes four temporal convolutional networks, one fully connected neural network, a first position probability structure, a second position probability structure, a first pose probability structure, and a second pose probability structure. The first position node, second position node, first pose node and second pose node in the Bayesian network are replaced by a temporal convolutional network, and the key event node in the Bayesian network is replaced by a fully connected neural network. Insert the first position probability structure, the second position probability structure, the first attitude probability structure, and the second attitude probability structure between the first position node, the second position node, the first attitude node, the second attitude node, and the key event node, respectively. The outputs of the first position probability structure, the second position probability structure, the first pose probability structure, and the second pose probability structure are the inputs of the fully connected neural network.

[0007] Optionally, the first position probability structure stores the event probability corresponding to the first position node; the second position probability structure stores the event probability corresponding to the second position node; the first attitude probability structure stores the event probability corresponding to the first attitude node; and the second attitude probability structure stores the event probability corresponding to the second attitude node.

[0008] Optional methods for constructing Bayesian networks include: Obtain a first position node, a second position node, a first attitude node, a second attitude node, and a key event node; the first position node and the first attitude node correspond to the position and attitude changes of one target in the competition; the first position node and the first attitude node correspond to the position and attitude changes of another target in the competition. The first position node, the second position node, the first attitude node, and the second attitude node are the parent nodes of the key event node; Based on the first position node, the second position node, the first attitude node, the second attitude node, and the key event node, determine the probability that a node affects a key event, and obtain the event probability.

[0009] Optionally, the step of detecting the event status and obtaining key time points through a time detection network based on the position and posture feature vectors of the competition target includes: A target set is obtained based on the position and posture feature vectors of multiple competition targets; each target set includes a position sequence and a posture feature sequence corresponding to two targets; the position sequence represents the position of the competition target at multiple time points; the posture feature sequence represents the posture feature vectors at multiple time points. The position sequences and pose feature sequences corresponding to the two targets are respectively input into the four temporal convolutional networks in the temporal detection network to obtain the discriminant value; the discriminant value indicates whether the key event is completed at a future time point. If the discriminant value is greater than the discriminant threshold, the corresponding time point is recorded as the key time point. If the discrimination value is less than the discrimination threshold, the target position and posture feature vector corresponding to the new time point are obtained for discrimination; the new time point refers to the time point after a competition status detection.

[0010] Optionally, obtaining the target position and posture feature vector based on the virtual event video includes: The virtual event video is divided into frames to extract images, resulting in virtual event images; multiple virtual event images are obtained corresponding to multiple time points; The virtual event images are subjected to target detection to obtain the location and category of the competition targets. In the virtual event image, the image corresponding to the target location is segmented to obtain the target location image; Based on the target location image and target category, a key point detection network is used to detect key points of the pose to obtain a pose skeleton image; the pose skeleton image represents the pose or shape of the target at the competition target location. An input pose extraction network is constructed based on the pose skeleton image to extract features and obtain the pose feature vector.

[0011] Optionally, the time interval between two adjacent virtual event images is the same and longer than the time interval of the time detection network.

[0012] Optionally, the training method for the time detection network includes: The system obtains the historical match target location, historical posture feature vector, and corresponding annotation discriminant value; the historical match target location refers to the match target location obtained from the virtual match video at a historical time point; the historical posture feature vector refers to the posture feature vector obtained from the virtual match video at a historical time point; and the annotation discriminant value indicates whether a key event has been completed. The historical target position and historical posture feature vector are input into the time detection network for detection to obtain training discriminant values. The loss is calculated from the training discriminant value and the labeled discriminant value, and the time detection network is trained.

[0013] Optionally, generating event display information based on key time points includes: Generate key event statements and key event images; Display key event statements and key event images at key time points.

[0014] Secondly, embodiments of the present invention provide a system for configuring and synchronizing dynamic content of virtual sports events in real time, including: The acquisition module is used to acquire virtual event videos; the virtual event videos are videos of ongoing virtual events. An extraction module is used to obtain the target location and posture feature vectors based on the virtual event video; the posture feature vectors represent the characteristics of the posture of the personnel at the target location. The Bayesian module is used to construct Bayesian networks; the Bayesian network represents a network that associates location and pose with the probability of key events. The network module is used to construct a time detection network based on the Bayesian network; The detection module is used to detect the state of the competition and obtain key time points by using a time detection network based on the position and posture feature vectors of the competition target. The generation module is used to generate event display information based on key time points.

[0015] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects: This invention also provides a method and system for dynamic content configuration and real-time synchronization of virtual sports events. Since key events affecting virtual sports events (such as a shot in football or overtaking an opponent in a sprint) are generally predictable based on position and posture, position and posture information is extracted from images. Furthermore, while position and posture influence key events, the impact is unknown; therefore, this invention designs a Bayesian network for detection. By designing a time detection network using a Bayesian network, not only can the influence of position and posture be obtained, but also a first position probability structure, a second position probability structure, a first posture probability structure, and a second posture probability structure are designed to obtain event probabilities in the Bayesian network. By jointly determining whether a key event will occur using posture and position, and by reducing the influence of redundant features through probability, the trajectory of position and posture changes of two targets is detected. The future time point where the key event occurs is predicted as the key time point.

[0016] The management interface allows for the configuration of banners, news articles, and advertisements via images, videos, and external links, with activation status set according to key time points. Banners, news articles, and advertisements can also be synchronized to the mini-program in real time via API, supporting homepage pop-up ads and personalized displays. Simultaneously, caching mechanisms and incremental updates are employed to reduce server load and improve content update efficiency. This achieves the technical effect of more accurately synchronizing and configuring advertising and other content delivery at the precise time points of key events. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for configuring and synchronizing dynamic content of virtual sports events in real time, provided by an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the time detection network structure in the flowchart of a method for configuring and synchronizing dynamic content of virtual sports events in real time, provided by an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram illustrating the time relationship of multiple detections using a time detection network in a flowchart of a method for configuring and synchronizing dynamic content of virtual sports events according to an embodiment of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings.

[0021] Example 1 like Figure 1 As shown in the figure, this embodiment of the invention provides a method for configuring and synchronizing dynamic content of virtual sports events in real time. The method includes: S101: Obtain virtual event video; the virtual event video is a video of an ongoing virtual event.

[0022] The virtual event videos are data obtained in real time from the virtual event engine through an interface (API).

[0023] In this embodiment, the virtual match video is a video of a football match.

[0024] S102: Based on the virtual event video, obtain the target location and posture feature vector; the posture feature vector represents the characteristics of the posture of the person at the target location.

[0025] The target location of the competition refers to the location of the participants in the competition.

[0026] S103: Construct a Bayesian network; the Bayesian network represents a network that associates location and pose with the probability of key events.

[0027] Here, the key event refers to an event that affects the match. In this embodiment, the key event refers to whether a goal is scored when a shot is taken. Therefore, the obtained Bayesian network represents the probability of whether changes in the goalkeeper's position and posture, and the position and shape of the shot ball, affect whether a goal is scored.

[0028] S104: Construct a time detection network based on the Bayesian network; S105: Using a time detection network, the event status is detected based on the position and posture feature vectors of the competition target to obtain key time points; S106: Generate event display information based on key state probabilities and key time points.

[0029] Optionally, constructing the time detection network based on the Bayesian network includes: The time detection network includes four temporal convolutional networks (TCNs), one fully connected neural network (FCN), a first position probability structure, a second position probability structure, a first pose probability structure, and a second pose probability structure.

[0030] In this embodiment, a temporal convolutional network is configured to receive feature vectors of the target's position or pose at five time points. Furthermore, because the temporal convolutional network corresponds to a causal convolutional network structure, it can output features representing changes in position or pose, and these features can then be used to predict whether key events will occur at future time points based on the trajectory of these changes.

[0031] The Temporal Convolutional Network (TCN) is mentioned above.

[0032] The structure of the time detection network is as follows: Figure 2 As shown.

[0033] The first position node, second position node, first pose node and second pose node in the Bayesian network are replaced by a Temporal Convolutional Network (TCN), and the key event node in the Bayesian network is replaced by a fully connected neural network.

[0034] The above method preserves the structure of each node in the Bayesian network, and a new network is constructed based on this structure using temporal convolutional networks and fully connected neural networks (FCN).

[0035] Insert the first position probability structure, the second position probability structure, the first attitude probability structure, and the second attitude probability structure between the first position node, the second position node, the first attitude node, the second attitude node, and the key event node, respectively.

[0036] The outputs of the first position probability structure, the second position probability structure, the first pose probability structure, and the second pose probability structure are the inputs of the fully connected neural network.

[0037] In this embodiment, a fully connected neural network detects changes in the position and posture of two corresponding targets. The goalkeeper's position, the soccer ball's position, the goalkeeper's posture, and the changes in the soccer ball's posture are used to jointly determine whether a shot should be taken at a predicted future time.

[0038] Using the above method, the Temporal Convolutional Network (TCN) is used to find position and pose changes, and the corresponding fully connected neural network is used to extract features of position and pose changes.

[0039] Optionally, the first position probability structure stores the event probability corresponding to the first position node; the second position probability structure stores the event probability corresponding to the second position node; the first attitude probability structure stores the event probability corresponding to the first attitude node; and the second attitude probability structure stores the event probability corresponding to the second attitude node.

[0040] Optional methods for constructing Bayesian networks include: Obtain a first position node, a second position node, a first attitude node, a second attitude node, and a key event node; the first position node and the first attitude node correspond to the position and attitude changes of one target in the competition; the first position node and the first attitude node correspond to the position and attitude changes of another target in the competition.

[0041] In this embodiment, the key event node corresponds to a key event, and the key event is whether the shot results in a goal.

[0042] In this embodiment, since the key event is shooting, the first position node represents the goalkeeper's position, the first posture node represents the goalkeeper's posture, the second position node represents the position of the shooting ball, and the second posture node represents the shape of the ball.

[0043] If the event is a running race, then the first position node and the first posture node correspond to one person participating in the race; the first position node and the first posture node correspond to another person participating in the race.

[0044] The first position node, the second position node, the first attitude node, and the second attitude node are the parent nodes of the key event nodes.

[0045] Among them, the first position node, the second position node, the first attitude node, and the second attitude node point to the key event node.

[0046] Based on the first position node, the second position node, the first attitude node, the second attitude node, and the key event node, determine the probability that a node affects a key event, and obtain the event probability.

[0047] The probability of an event can be obtained through expert scoring or virtual simulation. In the virtual simulation, images of successful shots in historical matches are acquired, and the changes in the goalkeeper's position and posture, as well as the changes in the position and posture of the ball, are recorded. Three of the four variables are kept constant, and one is randomly changed. The probability of a successful shot after the change is detected, and the corresponding event probability is obtained.

[0048] In this embodiment, the probability of the goalkeeper's position influencing the critical event is 40%, the probability of the goalkeeper's posture influencing the critical event is 18%, the probability of the football's position influencing the critical event is 40%, and the probability of the goalkeeper's position influencing the critical event is 2%.

[0049] Optionally, the step of detecting the event status and obtaining key time points through a time detection network based on the position and posture feature vectors of the competition target includes: A target set is obtained based on the position and posture feature vectors of multiple competition targets; each target set includes a position sequence and a posture feature sequence corresponding to two targets; the position sequence represents the position of the competition target at multiple time points; the posture feature sequence represents the posture feature vectors at multiple time points. The position sequences and pose feature sequences corresponding to the two targets are respectively input into the four temporal convolutional networks in the temporal detection network to obtain the discriminant value; the discriminant value indicates whether the key event is completed at a future time point.

[0050] In this embodiment, for example, the position sequence and posture feature sequence corresponding to the goalkeeper, and the position sequence and posture feature sequence corresponding to the football are obtained. Figure 2 As shown, the position sequence includes five target positions: the first target position, the second target position, ..., the fifth target position. The posture feature sequence includes five posture feature vectors: the first posture feature vector, the second posture feature vector, ..., the fifth posture feature vector. The temporal convolutional network input to the goalkeeper corresponds to the position sequence of the goalkeeper, the temporal convolutional network input to the goalkeeper corresponds to the first posture probability structure, the temporal convolutional network input to the football corresponds to the second position probability structure, and the temporal convolutional network input to the football corresponds to the second posture probability structure.

[0051] If the discriminant value is greater than the discriminant threshold, the corresponding time point is recorded as the critical time point.

[0052] In this embodiment, the discrimination threshold is 0.9.

[0053] If the discrimination value is less than the discrimination threshold, the target position and posture feature vector corresponding to the new time point are obtained for discrimination; the new time point refers to the time point after a competition status detection.

[0054] Because time is passing during network detection, a new virtual event video will be acquired after the detection is completed or the event video is uploaded. This new virtual event video can be combined with the previously acquired virtual event video to re-detect the discriminant value for future time points. The time relationship is as follows: Figure 3 As shown.

[0055] Optionally, obtaining the target position and posture feature vector based on the virtual event video includes: The virtual event video is divided into frames to extract images, resulting in virtual event images; multiple virtual event images are obtained corresponding to multiple time points.

[0056] The virtual event images are clear images.

[0057] The virtual event image is subjected to target detection to obtain the location and category of the target.

[0058] In this embodiment, the YOLOv5 model is used for target detection. Since the key event is shooting, the target categories include the football and the goalkeeper, and the game target position represents the position of the football and the goalkeeper.

[0059] In the virtual event image, the image corresponding to the target location is segmented to obtain the target location image.

[0060] In this embodiment, the target location image only contains images of the football or goalkeeper at the target location in the game.

[0061] Based on the target location image and target category, a key point detection network is used to detect key points of the pose to obtain a pose skeleton image; the pose skeleton image represents the pose or shape of the target at the competition target location.

[0062] In this embodiment, if the target category is a goalkeeper, the keypoint detection network is the OpenPose model, which is used to detect keypoints on the human body and connect the keypoints on the human limbs to form a pose skeleton image. If the target category is a soccer ball, the keypoint detection network is an edge detection network using the Canndy operator, and the edge shape of the soccer ball is used as the pose skeleton image.

[0063] An input pose extraction network is constructed based on the pose skeleton image to extract features and obtain the pose feature vector.

[0064] In this embodiment, the pose extraction network is a convolutional neural network (CNN).

[0065] The posture feature vector represents the extracted posture information.

[0066] Optionally, the time length between two adjacent virtual event images is the same and greater than the time length during which the time detection network runs.

[0067] Optionally, the training method for the time detection network includes: The system obtains the historical match target location, historical posture feature vector, and corresponding annotation discriminant value; the historical match target location refers to the match target location obtained from the virtual match video at a historical time point; the historical posture feature vector refers to the posture feature vector obtained from the virtual match video at a historical time point; and the annotation discriminant value indicates whether a key event has been completed.

[0068] In this embodiment, if the key event (scoring a goal) is completed, the labeling value is set to 1. If the key event is not completed (no goal is scored), the labeling value is set to 0.

[0069] The historical target position and historical posture feature vector are input into the time detection network for detection to obtain training discriminant values. The loss is calculated from the training discriminant value and the labeled discriminant value, and the time detection network is trained.

[0070] In this embodiment, the cross-entropy loss function is used to calculate the loss.

[0071] Optionally, generating event display information based on key time points includes: Generate key event statements and key event images.

[0072] In this embodiment, templates are used to generate real-time comments and news as key event statements. The images from advertisements are used as key event images.

[0073] Display key event statements and key event images at key time points.

[0074] Among them, key event statements and key event images will pop up at key time points, that is, at the time of a successful shot.

[0075] Using the methods described above, banners, news articles, and advertisements can be configured on the management side via images, videos, and external links, with their activation status set according to key time points. Banners, news articles, and advertisements can also be synchronized to the mini-program in real time via API, supporting homepage pop-up ads and personalized displays. Simultaneously, caching mechanisms and incremental updates are employed to reduce server load and improve content update efficiency.

[0076] Example 2 Based on the above-described method for configuring and synchronizing dynamic content of virtual sports events in real time, this invention also provides a system for configuring and synchronizing dynamic content of virtual sports events in real time, the system comprising: The acquisition module is used to acquire virtual event videos; the virtual event videos are videos of ongoing virtual events. An extraction module is used to obtain the target location and posture feature vectors based on the virtual event video; the posture feature vectors represent the characteristics of the posture of the personnel at the target location. The Bayesian module is used to construct Bayesian networks; the Bayesian network represents a network that associates location and pose with the probability of key events. The network module is used to construct a time detection network based on the Bayesian network; The detection module is used to detect the state of the competition and obtain key time points by using a time detection network based on the position and posture feature vectors of the competition target. The generation module is used to generate event display information based on key time points.

[0077] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0078] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0079] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

Claims

1. A method for configuring and synchronizing dynamic content in virtual sports events in real time, characterized in that, include: Get virtual event videos; The virtual match videos are videos of ongoing virtual matches; Based on the virtual event video, the target location and posture feature vectors are obtained; the posture feature vectors represent the characteristics of the posture of the personnel at the target location. Construct a Bayesian network; the Bayesian network represents a network that associates location and pose with the probability of key events; Construct a time detection network based on the Bayesian network; By using a time detection network, the status of the competition is detected based on the position and posture feature vectors of the competition target, and key time points are obtained. Based on key time points, generate event display information.

2. The method for configuring and synchronizing dynamic content of virtual sports events in real time according to claim 1, characterized in that, The step of constructing a time detection network based on the Bayesian network includes: The time detection network includes four temporal convolutional networks, one fully connected neural network, a first position probability structure, a second position probability structure, a first pose probability structure, and a second pose probability structure. The first position node, second position node, first pose node and second pose node in the Bayesian network are replaced by a temporal convolutional network, and the key event node in the Bayesian network is replaced by a fully connected neural network. Insert the first position probability structure, the second position probability structure, the first attitude probability structure, and the second attitude probability structure between the first position node, the second position node, the first attitude node, the second attitude node, and the key event node, respectively. The outputs of the first position probability structure, the second position probability structure, the first pose probability structure, and the second pose probability structure are the inputs of the fully connected neural network.

3. The method for configuring and synchronizing dynamic content of virtual sports events in real time according to claim 2, characterized in that, The first position probability structure stores the event probability corresponding to the first position node; the second position probability structure stores the event probability corresponding to the second position node; the first attitude probability structure stores the event probability corresponding to the first attitude node; and the second attitude probability structure stores the event probability corresponding to the second attitude node.

4. The method for configuring and synchronizing dynamic content of virtual sports events in real time according to claim 1, characterized in that, Methods for constructing Bayesian networks include: Obtain a first position node, a second position node, a first attitude node, a second attitude node, and a key event node; the first position node and the first attitude node correspond to the position and attitude changes of one target in the competition; the first position node and the first attitude node correspond to the position and attitude changes of another target in the competition. The first position node, the second position node, the first attitude node, and the second attitude node are the parent nodes of the key event node; Based on the first position node, the second position node, the first attitude node, the second attitude node, and the key event node, determine the probability that a node affects a key event, and obtain the event probability.

5. The method for configuring and synchronizing dynamic content of virtual sports events in real time according to claim 1, characterized in that, The process involves using a time-based detection network to detect the event status based on the target's position and posture feature vectors, thereby obtaining key time points, including: A target set is obtained based on the position and posture feature vectors of multiple competition targets; each target set includes a position sequence and a posture feature sequence corresponding to two targets; the position sequence represents the position of the competition target at multiple time points; the posture feature sequence represents the posture feature vectors at multiple time points. The position sequences and pose feature sequences corresponding to the two targets are respectively input into the four temporal convolutional networks in the temporal detection network to obtain the discriminant value; the discriminant value indicates whether the key event is completed at a future time point. If the discriminant value is greater than the discriminant threshold, the corresponding time point is recorded as the key time point. If the discrimination value is less than the discrimination threshold, the target position and posture feature vector corresponding to the new time point are obtained for discrimination; the new time point refers to the time point after a competition status detection.

6. The method for dynamic content configuration and real-time synchronization of virtual sports events according to claim 1, characterized in that, The process of obtaining the target position and posture feature vector based on the virtual event video includes: The virtual event video is divided into frames to extract images, resulting in virtual event images; multiple virtual event images are obtained corresponding to multiple time points; The virtual event images are subjected to target detection to obtain the location and category of the competition targets. In the virtual event image, the image corresponding to the target location is segmented to obtain the target location image; Based on the target location image and target category, a key point detection network is used to detect key points of the pose to obtain a pose skeleton image; the pose skeleton image represents the pose or shape of the target at the competition target location. An input pose extraction network is constructed based on the pose skeleton image to extract features and obtain the pose feature vector.

7. The method for configuring and synchronizing dynamic content of virtual sports events in real time according to claim 6, characterized in that, The time interval between two adjacent virtual event images is the same and greater than the time interval of the time detection network.

8. The method for dynamic content configuration and real-time synchronization of virtual sports events according to claim 1, characterized in that, The training method for the time detection network includes: The system obtains the historical match target location, historical posture feature vector, and corresponding annotation discriminant value; the historical match target location refers to the match target location obtained from the virtual match video at a historical time point; the historical posture feature vector refers to the posture feature vector obtained from the virtual match video at a historical time point; and the annotation discriminant value indicates whether a key event has been completed. The historical target position and historical posture feature vector are input into the time detection network for detection to obtain training discriminant values. The loss is calculated from the training discriminant value and the labeled discriminant value, and the detection network is trained.

9. The method for dynamic content configuration and real-time synchronization of virtual sports events according to claim 1, characterized in that, The generation of event display information based on key time points includes: Generate key event statements and key event images; Display key event statements and key event images at key time points.

10. A system for dynamic content configuration and real-time synchronization of virtual sports events, characterized in that, include: The acquisition module is used to acquire virtual event videos; The virtual match videos are videos of ongoing virtual matches; An extraction module is used to obtain the position and posture feature vectors of the competition target based on the virtual competition video; the posture feature vectors represent the characteristics of the posture of the personnel at the competition target position; The Bayesian module is used to construct Bayesian networks; the Bayesian network represents a network that associates location and pose with the probability of key events. The network module is used to construct a time detection network based on the Bayesian network; The detection module is used to detect the state of the competition and obtain key time points by using a time detection network based on the position and posture feature vectors of the competition target. The generation module is used to generate event display information based on key time points.

Images