Technical and tactical ability monitoring and analyzing system and method for separation net ball games
Through multiple time synchronization cameras and neural network technology, three-dimensional position monitoring of balls and players in tennis-type sports is realized, solving the problems of strong subjectivity and low accuracy in the existing technology, and improving analysis efficiency and accuracy.
Patent Information
- Application Number
- CN202411119469.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-08-14
AI Technical Summary
The existing tennis-type sports monitoring and analysis technologies are highly subjective and inefficient, making it difficult to achieve large-scale and in-depth data analysis, especially when dealing with occlusion situations and obtaining player spatial information.
Multiple time synchronization cameras are used to collect video frames, and camera parameters are calculated through neural network models and deep neural networks to determine the three-dimensional position of the ball and the player. Combined with the three-dimensional spatial geometric relationship constraints, accurate monitoring and analysis of player postures is achieved.
It improves the accuracy and efficiency of monitoring and analysis of tennis-type sports, alleviates the occlusion problem, provides the three-dimensional position of the ball and the three-dimensional posture of the player, and provides a solid foundation for deep technical and tactical analysis.
Smart Images

Figure CN119090972A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to monitoring and analyzing sports technical and tactical capabilities. Background Art
[0002] Tennis sports (e.g., table tennis, badminton, tennis, volleyball, etc.) are popular sports. The high speed of the ball in these sports places high demands on the players' technical and tactical abilities and their speed of reaction to the game situation. In high-level competitions, the outcome of a game often depends on subtle technical and tactical differences. Therefore, accurate monitoring and analysis of the players' technical and tactical abilities during training and competition is crucial to improving training efficiency and competition performance.
[0003] For tennis and other sports, traditional training and game analysis methods rely on manual observation by coaches or analysts. This method is highly subjective, inefficient, and difficult to achieve large-scale and in-depth data analysis.
[0004] In computer vision technology, computers can accurately identify and track objects by analyzing images and videos. As computer vision technology develops, it is also widely used in sports analysis to provide an efficient and objective means of monitoring and analyzing athlete performance.
[0005] Chinese patent application publication CN 109977752 A discloses a method for analyzing badminton player skills and tactics based on sequence pattern mining, which requires manual video viewing and action annotation, and is difficult to automatically process massive data.
[0006] Chinese patent application publication CN 110433471 A discloses a badminton motion trajectory monitoring and analysis system and method, which uses a single high-speed depth camera solution, requires special equipment to execute, and has a relatively strict setup. In addition, the method has limited processing capabilities for situations where there is occlusion, has low accuracy in obtaining badminton's spatial information, and does not estimate the player's spatial information.
[0007] Chinese patent application publication CN 116958872 A discloses an intelligent auxiliary training method and system for badminton, which uses a binocular camera, has limited ability to process occlusion and calculate depth information, and has low accuracy in obtaining spatial information of badminton and players.
[0008] There is a need in the art for more accurate and efficient monitoring and analysis of tennis sports. Summary of the invention
[0009] The present invention is provided in order to provide a more accurate and efficient monitoring and analysis technology for tennis sports.
[0010] According to one aspect of the present invention, there is provided a system for monitoring technical and tactical capabilities of tennis sports, comprising: a camera parameter estimation module, for: receiving a plurality of video frames of motion captured by a plurality of time-synchronized cameras at a plurality of viewing angles; calculating camera parameters based on position information of elements associated with a court in the plurality of video frames using a neural network model; a ball positioning module, for: determining a two-dimensional position of a ball from the plurality of video frames using a deep neural network model; determining a three-dimensional position of the ball based on the determined two-dimensional position of the ball and the camera parameters; and a player posture detection module, for: determining the two-dimensional posture of all detectable human targets from the plurality of video frames; determining the three-dimensional posture of the human target based on the determined two-dimensional posture of the human target and the camera parameters; and determining the three-dimensional posture of the player through the three-dimensional posture of the human target based on three-dimensional spatial geometric relationship constraints associated with the court.
[0011] According to another aspect of the present invention, there is provided a technical and tactical capability analysis system for tennis sports, comprising: the technical and tactical capability monitoring system as described above; and a technical and tactical analysis module for determining technical and tactical data based on the determined three-dimensional position of the ball and / or the three-dimensional posture of the player.
[0012] According to another aspect of the present invention, there is provided a system for presenting a three-dimensional animation of tennis sports, comprising: the technical and tactical ability monitoring system as described above; and a three-dimensional rendering module for: receiving data indicating the three-dimensional position of a ball and / or the three-dimensional posture of a player determined within a time period; and rendering the received data into a three-dimensional animation.
[0013] According to another aspect of the present invention, a television broadcasting system for tennis sports is provided, which includes: a technical and tactical ability monitoring system as described above; a technical and tactical analysis module for determining technical and tactical data based on the determined three-dimensional position of the ball and / or the three-dimensional posture of the player; a three-dimensional rendering module for: receiving data indicating the three-dimensional position of the ball and / or the three-dimensional posture of the player determined within a time period; and rendering the received data into a three-dimensional animation and / or visual rendering of the determined technical and tactical data to obtain a rendered video; an online packaging module for beautifying and packaging the technical and tactical data to obtain a key signal; a super slow motion system for recording and editing the rendered video to generate a super slow motion picture; a switcher for integrating and processing the key signal and the super slow motion picture to generate a broadcast picture; and a television signal broadcasting system for broadcasting television signals to television viewers based on the broadcast picture.
[0014] According to another aspect of the present invention, a method for monitoring technical and tactical capabilities of tennis sports is provided, comprising: receiving multiple video frames of motion captured by multiple time-synchronized cameras at multiple perspectives; using a neural network model to calculate camera parameters based on position information of elements associated with a court in the multiple video frames; using a deep neural network model to determine the two-dimensional position of a ball from the multiple video frames; determining the three-dimensional position of the ball based on the determined two-dimensional position of the ball and the camera parameters; determining the two-dimensional posture of all detectable human targets from the multiple video frames; determining the three-dimensional posture of the human target based on the determined two-dimensional posture of the human target and the camera parameters; and determining the three-dimensional posture of the player through the three-dimensional posture of the human target based on the three-dimensional spatial geometric relationship constraints associated with the court.
[0015] According to another aspect of the present invention, a computer-readable storage medium is provided, on which computer instructions are stored. When the computer instructions are executed by a processor, the steps of the method described above are implemented.
[0016] According to another aspect of the present invention, a computer program product is provided, comprising computer instructions, which implement the steps of the method described above when executed by a processor.
[0017] The system and method according to the embodiments of the present invention make the monitoring and analysis of tennis sports more accurate and efficient. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Embodiments of the present invention are described with reference to the accompanying drawings.
[0019] Figure 1 A block diagram of a system for monitoring technical and tactical abilities of tennis sports according to some embodiments of the present invention is shown.
[0020] Figure 2 A block diagram of a technical and tactical ability analysis system for tennis according to some embodiments of the present invention is shown.
[0021] Figure 3 A block diagram is shown of a system for presenting three-dimensional animation of tennis motions according to some embodiments of the present invention.
[0022] Figure 4 An example animation screen presented by a system for presenting a three-dimensional animation of a tennis game according to some embodiments of the present invention is shown.
[0023] Figure 5 An overall flow chart of a system for presenting three-dimensional animation of tennis motions according to some embodiments of the present invention is shown.
[0024] Figure 6A block diagram of a television broadcast system for tennis sports according to some embodiments of the present invention is shown.
[0025] Figure 7 A schematic diagram showing the arrangement of camera positions of video acquisition cameras used in a television broadcasting system for tennis sports according to some embodiments of the present invention.
[0026] Figure 8 A flow chart showing a method for monitoring technical and tactical abilities in tennis sports according to some embodiments of the present invention.
[0027] Fig. 9 A flow chart illustrating a process of calculating camera parameters according to some embodiments of the present invention.
[0028] Fig.10 A flowchart illustrating a process of training a neural network model for calculating camera parameters according to some embodiments of the present invention.
[0029] Fig.11 A flow chart illustrating a process of determining the two-dimensional position of a ball according to some embodiments of the present invention.
[0030] Fig.12 A flow chart illustrating a process of determining a three-dimensional pose of a player according to some embodiments of the present invention.
[0031] Fig.13 A block diagram of a machine-readable storage medium according to some embodiments of the invention is shown.
[0032] Fig.14 A block diagram illustrating a computer program product according to some embodiments of the invention. DETAILED DESCRIPTION
[0033] In the present application, the term "netball sports" refers to a type of sports that generally includes two teams competing against each other, the playing field is divided into two parts by a net, each team competes with a ball in its respective part, and the players of each team do not have direct physical contact and confrontation with the opposing players. As examples, netball sports may include table tennis, badminton, tennis, volleyball, etc.
[0034] According to one aspect of the present invention, a technical and tactical ability monitoring system for tennis sports is provided.
[0035] Figure 1 A block diagram of a system 100 for monitoring technical and tactical abilities in tennis according to some embodiments of the present invention is shown.
[0036] The technical and tactical capability monitoring system 100 may include a camera parameter estimation module 110, a ball positioning module 120, and a player posture detection module 130. In some embodiments, the technical and tactical capability monitoring system 100 may include a computing device (e.g., a server), which may include various computing resources (e.g., a processor, etc.). Each of the camera parameter estimation module 110, the ball positioning module 120, and the player posture detection module 130 may include software components and / or hardware components. For example, each of the camera parameter estimation module 110, the ball positioning module 120, and the player posture detection module 130 may be programmed to call a corresponding computing resource to perform a specified operation.
[0037] The camera parameter estimation module 110 may be configured to receive a plurality of video frames 140. The plurality of video frames 140 may be a plurality of video frames of motion captured by a plurality of time-synchronized cameras at a plurality of viewing angles. Figure 7 The scene arrangement for collecting a plurality of video frames 140 is described. The camera parameter estimation module 110 may also be used to calculate camera parameters based on position information of elements associated with the court in the plurality of video frames 140 using a neural network. As an example, the elements associated with the court may include various lines of the court field, such as boundary lines, lines defining relevant areas, etc. The camera parameters may include external parameters of the camera.
[0038] The ball positioning module 120 may be used to determine the two-dimensional position of the ball from a plurality of video frames 140 using a deep neural network model. The deep neural network model is trained to accurately capture the two-dimensional information coordinates of the ball from a series of consecutive video frames, laying a solid foundation for subsequent three-dimensional positioning, motion evaluation and analysis, etc. The ball positioning module 120 may also be used to determine the three-dimensional position 150 of the ball based on the determined two-dimensional position of the ball and camera parameters.
[0039] The player pose detection module 130 may be used to determine the two-dimensional poses of all detectable human targets from a plurality of video frames. As an example, this may be achieved by executing a target detection algorithm. The player pose detection module 130 may also be used to determine the three-dimensional pose of the human target based on the determined two-dimensional pose of the human target and camera parameters. The player pose detection module 130 may also be used to determine the three-dimensional pose 160 of the player from the three-dimensional pose of the human target based on the three-dimensional spatial geometric relationship constraints associated with the court. As an example, the three-dimensional spatial geometric relationship constraints may include a three-dimensional size constraint of the court.
[0040] The three-dimensional position 150 of the ball determined by the ball positioning module 120 and the three-dimensional posture 160 of the player determined by the player posture detection module 130 can be used as the output of the technical and tactical ability monitoring system 100, as monitoring indicators of the player's tactics and technical performance during tennis sports, which can be used as a reference for coaches, analysts, and players, and as a basis for further adjusting training and competition plans to improve training and competition levels.
[0041] The technical and tactical ability monitoring system 100 according to some embodiments of the present invention can effectively alleviate the occlusion problem by performing visual processing on video frames collected from multiple perspectives during tennis sports, and the results are highly accurate; in addition, the technical and tactical ability monitoring system 100 can also provide the three-dimensional position of the ball and the three-dimensional posture of the player, thereby providing a solid foundation for further in-depth and extensive technical and tactical analysis.
[0042] In some embodiments, in order to calculate the camera parameters, the camera estimation module 110 may be configured to detect a plurality of key points of the court in the plurality of video frames 140. As an example, the key points may include points on lines on the court, intersections of lines, etc. As an example, 100 key points may be detected. Alternatively, more or fewer key points may be detected according to actual sports scenes and accuracy requirements.
[0043] The camera estimation module 110 may also be used to match a plurality of key points with a predefined standard court template. As an example, the plurality of key points may be densely matched with corresponding points on the standard court template. The camera estimation module 110 may also be used to estimate camera parameters using an algorithm based on the matching result. As an example, the algorithm may include a PnP algorithm, which is a method for solving a correspondence between three-dimensional and two-dimensional points. As an example, the camera parameters may include external parameters of the camera.
[0044] In some embodiments, the camera estimation module 110 may be used to estimate camera parameters based on the homography matching error between the homography transformation of multiple key points and the homography transformation of the standard three-dimensional court template. In some embodiments, the multiple key points detected may cover the entire court. Since the homography matching error of each pixel of the court in each video frame increases as the distance between the pixel and the key point of the generated homography transformation increases, therefore, making the detected key points cover the entire court helps to reduce the spacing between points, so that the homography matching error caused by the occluded key points is limited to the local area, and gradually eliminated as the pixel approaches other key points, thereby reducing the error caused by occlusion. Alternatively or additionally, the multiple key points detected may be evenly distributed on the court. By setting evenly distributed key points, the homography matching error can be further evenly limited to each local area, thereby improving the overall accuracy.
[0045] In order to train the neural network model to perform the above-mentioned process of calculating camera parameters, it is necessary to provide true values during the training process, which involves manually annotating a corresponding number (e.g., 100) of key points for the scene in each video frame. In order to save a lot of manual annotation work, in some embodiments, a template-aware annotation tool can be used. With this annotation tool, only a threshold number of key points need to be manually annotated. As an example, the threshold number can be less than (e.g., much less than) the number of key points detected. As an example, the threshold number can be 4-6, which is much less than the number of key points detected by the neural network model during the reasoning process (e.g., 100).
[0046] Accordingly, the neural network model can be trained through the following steps: arbitrarily manually annotating a threshold number of key points in each video frame, where the threshold number is less than the number of key points detected; calculating the homography transformation between the court in the video frame and the standard three-dimensional court template for the threshold number of key points; based on the result of the homography transformation, mapping the intersection points of the court lines associated with the court from the standard three-dimensional court template to the video frame; comparing the mapped intersection points with the corresponding intersection points in the standard three-dimensional court template; manually fine-tuning the result of the comparison to indicate the position of one or more intersection points with the maximum error in the video frame; and adjusting the homography transformation based on the position of the fine-tuned one or more intersection points by minimizing the remapping error method.
[0047] As an example, for the above process, any four visible key points can be manually calibrated in the court of the video frame, and the homography transformation between the court in the video frame and the standard three-dimensional template court can be calculated based on the four key points. Then, the intersection points (e.g., all intersection points) of the court lines can be mapped from the standard three-dimensional court template to the video frame according to the result of the homography transformation, and the mapped intersection points can be compared with the corresponding intersection points of the court in the video frame. Then, the position of one or more intersection points with the largest error with the actual court intersection points in the video frame can be manually fine-tuned. The fine-tuned intersection points generate new corresponding point relationships. Since there is usually no homography transformation that can exactly meet the matching relationship requirements of the threshold number (e.g., 4) of key points, the method of minimizing the remapping error is used to calculate the required homography transformation. The neural network model trained in this way can extract accurate two-dimensional-three-dimensional point pairs from the accurate homography transformation, so that the algorithm (e.g., PnP algorithm) is further used to calculate the camera parameters (e.g., the camera's external parameters).
[0048] In some embodiments, considering the diversity of video sources in actual application scenarios, the ball positioning module 120 may perform preprocessing. For example, the ball positioning module 120 may adjust the resolution of all video frames to a fixed predetermined resolution. As an example, the fixed predetermined resolution may be 288x512 pixels. The two-dimensional position of the subsequent ball is determined based on the video frame with the adjusted resolution. In this way, the general applicability of the deep neural network to video frames of different resolutions is guaranteed, and the consumption of computing resources is effectively controlled, thereby ensuring the efficient execution of the algorithm.
[0049] In some embodiments, in order to deepen the concept of time series analysis, the ball positioning module 120 can receive multiple consecutive video frames (e.g., 8 consecutive frames) as input of the deep neural network. This not only enhances the tracking coherence and stability of the deep neural network model in the time dimension, but also enables the deep neural network model to more delicately break down the continuity and change trend of the movement.
[0050] In some embodiments, in order to combat complex and changeable background interference to improve the accuracy of ball recognition, background frame correction technology can also be introduced. The ball positioning module 120 can calculate the background in the video frame under the same viewing angle separately, and superimpose the calculated results into the video frame. As an example, the background of the video frame can be obtained by collecting when there are no players on the court, or by collecting multiple frames of the court with players and removing the players therein through an algorithm. The above-mentioned background frame correction technology can significantly enhance the contrast between the ball and the background, so that clear positioning of the ball can be achieved even in a complex game environment.
[0051] In some embodiments, the ball positioning module 120 can be used to generate a heat map that indicates the area where the ball may appear and the probability density of appearing in the area. The generated heat map can be matched to the size of the input video frame. In order to enable the ball positioning module 120 to generate a heat map, in the training phase, the real position information of the marked ball can be used to generate an accurate heat map. After the training is completed, in order to determine the predicted two-dimensional coordinates of the ball from the heat map, post-processing techniques can be implemented in the inference phase. The ball positioning module 120 can be used to determine the position of the ball from the heat map using a contour extraction method and / or an optimal maximum contour method, thereby achieving extremely high positioning accuracy.
[0052] In some embodiments, in order to further improve the reasoning speed of the deep learning model in practical applications, the original deep learning model can be converted and optimized. For example, the original PyTorch model can be converted and optimized using NVIDIA Tensor RT technology, and the model parameters can be quantized to half precision (lower precision than the original precision). This not only verifies the slight attenuation of model performance in the test set, but also improves the reasoning speed of the model without sacrificing too much precision, making the entire ball positioning process smoother and more immediate.
[0053] In some embodiments, the player posture detection module 130 may identify all detectable human targets through a target detection algorithm. The player posture detection module 130 may generate a bounding box sequence for each human target using a two-dimensional tracking algorithm, crop and enlarge the image within the bounding box, and use a two-dimensional posture estimation algorithm to obtain the two-dimensional posture of the human target. The player posture detection module 130 may perform triangulation based on the two-dimensional posture information of the human target in the multi-view video frame coordinate system and the camera parameters to determine the three-dimensional posture coordinates of the human target.
[0054] In some embodiments, the player posture detection module 130 may screen a predetermined number of players' three-dimensional postures from all detectable human targets' three-dimensional postures based on the three-dimensional spatial geometric relationship constraints associated with the court, wherein the predetermined number is determined at least in part based on the number of players specified in the sports event. As an example, the three-dimensional spatial geometric relationship constraints associated with the court may include the three-dimensional size constraints of the court. Taking badminton as an example, the player posture detection module 130 may screen the three-dimensional posture information of the two (for singles) or four (for doubles) players with the highest three-dimensional position matching degree of the badminton court from the three-dimensional postures of all detected human targets. By combining the three-dimensional spatial geometric relationship constraints associated with the court, players can be identified more efficiently and accurately from all detected human targets.
[0055] In some embodiments, the player posture detection module 130 may determine the three-dimensional posture of the player using a process different from the traditional top-down "target detection-target tracking-gesture recognition" process.
[0056] Chinese patent application publication CN 116958872 A uses the traditional top-down "target detection-target tracking-posture recognition" process. For the two-dimensional posture estimation task of the player, the invention adopts the "top-down" target detection-target tracking-posture recognition process. Specifically, the target detection algorithm is first used to obtain the bounding boxes of all identifiable human targets, and the court sidelines are used for filtering to retain the corresponding bounding boxes of the players. Then the two-dimensional tracking algorithm is used to obtain the bounding box sequence corresponding to each player. The image in the bounding box is cropped and enlarged, and the two-dimensional posture estimation algorithm is used to obtain the two-dimensional posture of the player. Thereafter, triangulation is performed based on the two-dimensional information of the badminton and the player in the monocular or binocular image coordinate system and the camera parameters to obtain the three-dimensional position of the badminton and the player, and finally the continuous image or video is used to form the motion trajectory of the badminton and the player.
[0057] In contrast, in some embodiments, an efficient three-dimensional tracking detection mode may be adopted. The player posture detection module 130 may estimate the position range of the player in multiple video frames at the next moment through the back projection technology based on the player's three-dimensional posture and camera parameters and the continuity of the three-dimensional posture change at adjacent times according to each tracked player. The player posture detection module 130 may skip target detection within the position range and directly estimate the two-dimensional posture of the player at the next moment. Thus, by combining the characteristics of the detected players continuously moving within the local image range at adjacent times and the three-dimensional spatial geometric relationship constraints associated with the court (which constrains the number of players in the specified area of the sport, possible positions, etc.), for the detected players, there is no need to perform target detection at the next moment, but based on the posture of the player at the current moment, the single-person two-dimensional posture estimation and correction of the player is performed at the next moment, thereby improving the efficiency of generating the two-dimensional posture of the player in a continuous time series.
[0058] In some embodiments, after generating a continuous time series of two-dimensional postures of players according to the above process, the player posture detection module 130 can also skip the matching task and directly perform single-person triangulation on the player for the estimated two-dimensional posture of the player at the next moment. Therefore, in the triangulation link, the single-person triangulation of the player can be directly performed on the result of the newly generated two-dimensional posture estimation of the single person according to the consistency of the projection area of the same player at different viewing angles, avoiding the matching between multiple detection results under different viewing angles, thereby further improving the matching accuracy and significantly shortening the matching time.
[0059] According to one aspect of the present invention, a technical and tactical ability analysis system for tennis sports is provided.
[0060] Figure 2A block diagram of a technical and tactical ability analysis system 200 for tennis sports according to some embodiments of the present invention is shown.
[0061] The technical and tactical capability sharing system 200 may include a camera parameter estimation module 210, a ball positioning module 220, a player posture detection module 230, and a technical and tactical analysis module 240. In some embodiments, the technical and tactical capability monitoring system 200 may include a computing device (e.g., a server), which may include various computing resources (e.g., a processor, etc.). Each of the camera parameter estimation module 210, the ball positioning module 220, the player posture detection module 230, and the technical and tactical analysis module 240 may include software components and / or hardware components. For example, each of the camera parameter estimation module 210, the ball positioning module 220, the player posture detection module 230, and the technical and tactical analysis module 240 may be programmed to call corresponding computing resources to perform specified operations.
[0062] The camera parameter estimation module 210 can be combined with the above Figure 1 The camera parameter estimation module 110 described above is similar. The ball positioning module 220 can be combined with the above Figure 1 The ball positioning module 110 described above is similar. The player posture detection module 230 can be combined with the above Figure 1 The technical and tactical ability analysis module 200 may receive a plurality of video frames 250. The plurality of video frames 250 may be combined with the above Figure 1 The ball positioning module 220 can determine the three-dimensional position 260 of the ball, which can be combined with the above Figure 1 The three-dimensional position 150 of the ball is similar to that described above. The player posture detection module 230 can determine the three-dimensional posture 270 of the player, which can be combined with the above Figure 1 The three-dimensional pose 160 of the player is described similarly.
[0063] The technical and tactical analysis module 240 may receive the three-dimensional position 260 of the ball and / or the three-dimensional posture 270 of the player to determine technical and tactical data 280 .
[0064] The technical and tactical data that can be determined by the technical and tactical analysis module 240 may include net crossing speed, net crossing height, net crossing angle, number of shots and point statistics, player movement distance, and player take-off events.
[0065] As an example, the technical and tactical analysis module 240 can calculate the three-dimensional speed and direction of the ball at any time through the first-order derivative of the distance with respect to time according to the three-dimensional position information 260 of the ball. Thus, the height of the ball from the ground, the speed, and the projection of the angle between the ball's direction of movement and the net plane at the moment closest to the net plane on the ground can be calculated, thereby calculating the speed, height, and angle of the ball passing the net.
[0066] As an example, the technical and tactical analysis module 240 can determine the occurrence, time and location of a hitting event by judging whether the angle between the ball speed direction and the normal direction of the net changes, thereby providing statistical analysis data on the number and location of hitting.
[0067] As an example, the technical and tactical analysis module 240 can further accumulate the movement distance of the center of gravity of the player within a specified time according to the three-dimensional posture of the player to provide a statistical analysis function of the player's movement distance. For example, by detecting whether the height of the lowest three-dimensional key point (e.g., ankle or toe) of the player from the ground exceeds a preset threshold height (e.g., 25 cm), the occurrence of a take-off event can be determined, and the maximum height from the ground within the time range of the take-off event can be calculated, thereby calculating the take-off height.
[0068] As an example, the technical and tactical data 280 that the technical and tactical analysis module 240 can determine and output may include: player's moving distance (meters), maximum smash speed over the net (kilometers / hour), maximum net crossing angle in front of the net (degrees), player's number of jump attacks (times), player's maximum jump height (centimeters), etc.
[0069] According to some embodiments of the present invention, the technical and tactical ability analysis system 200 can determine technical and tactical data 280 based on the determined three-dimensional position 260 of the ball and / or the three-dimensional posture 270 of the player, thereby providing an accurate reference for improving the player's training and competition performance.
[0070] According to one aspect of the present invention, a system for presenting three-dimensional animation of tennis sports is provided.
[0071] Figure 3 A block diagram is shown of a system 300 for presenting three-dimensional animation of tennis motions according to some embodiments of the present invention.
[0072] The system 300 for presenting three-dimensional animation may include a camera parameter estimation module 310, a ball positioning module 320, a player posture detection module 330, and a three-dimensional rendering module 240. In some embodiments, the technical and tactical capability monitoring system 300 may include a computing device (e.g., a server), which may include various computing resources (e.g., a processor, etc.). Each of the camera parameter estimation module 310, the ball positioning module 320, the player posture detection module 330, and the three-dimensional rendering module 340 may include software components and / or hardware components. For example, each of the camera parameter estimation module 310, the ball positioning module 320, the player posture detection module 330, and the three-dimensional rendering module 340 may be programmed to call the corresponding computing resources to perform a specified operation.
[0073] The camera parameter estimation module 310 can be combined with the above Figure 1 The camera parameter estimation module 110 described above is similar. The ball positioning module 320 can be combined with the above Figure 1 The ball positioning module 120 described above is similar. The player posture detection module 330 can be combined with the above Figure 1 The system 300 for presenting a three-dimensional animation may receive a plurality of video frames 350. The plurality of video frames 350 may be combined with the above Figure 1 The ball positioning module 320 can determine the three-dimensional position 360 of the ball, which can be combined with the above Figure 1 The three-dimensional position 150 of the ball described above is similar. The player posture detection module 330 can determine the three-dimensional posture 370 of the player, which can be combined with the above Figure 1 The three-dimensional pose 160 of the player is described similarly.
[0074] In some embodiments, the 3D rendering module 340 may be used for animation rendering of real-time games.
[0075] The three-dimensional rendering module 340 may be used to receive data indicating a three-dimensional position of a ball 360 determined over a period of time. In some embodiments, the three-dimensional rendering module 340 may be used to receive data indicating a three-dimensional pose 370 of a player determined over a period of time. In some embodiments, the three-dimensional rendering module 340 may be used to receive data indicating a three-dimensional position of a ball 360 and a three-dimensional pose 370 of a player determined over a period of time.
[0076] The 3D rendering module 340 can also be used to render the received data into a 3D animation 380. Depending on the data received by the 3D rendering module 340, the 3D rendering module 340 can render the received data into a 3D animation of the ball over a period of time, a 3D animation of the players, or a 3D animation reflecting the relationship between people and the ball.
[0077] In some embodiments, the system 300 for presenting three-dimensional animation may further include a technical and tactical analysis module 390. The technical and tactical analysis module 390 may be combined with the above Figure 2 The tactical analysis module 390 may receive the three-dimensional position 360 of the ball determined by the ball positioning module 320 and / or the three-dimensional posture 370 of the player determined by the player posture detection module 330 to determine tactical data 395 .
[0078] In some embodiments, the three-dimensional rendering module 340 may be used to visually render the determined technical and tactical data 395 .
[0079] As an example, in order to implement the three-dimensional rendering module 340, the EEVEE rendering engine of Blender 4.1 can be used, and the generation and interaction of the animation can be controlled by Python programming. Specifically, in the data preprocessing and scene setting part, a Python script can be written to parse the sequence data of the input three-dimensional position 360 of the ball and / or the three-dimensional posture 370 of the player, and convert these data into a format available to Blender. At the same time, Blender can be used to create a realistic court scene, set accurate court models, player models and ball models to ensure that the produced animation conforms to the actual game situation. In the animation generation and special effects production part, a Python script can be written to control the motion trajectory of the player and ball models according to the input sequence data, and collision detection and interaction logic can be added to simulate the real game situation. In some embodiments, the three-dimensional rendering module 340 can also design ball trajectory special effects and / or batting special effects. Blender's tracking and noise texture are used to simulate trajectory dragging, and the particle system is used to enhance the visual effect of the ball's flight. UV mapping can be used to expand the transparency gradient mapping based on the Y axis on a three-dimensional cone, so that the batting is presented as a diffuse ripple halo effect.
[0080] According to some embodiments of the present invention, the system 300 for presenting three-dimensional animation further provides users with realistic and highly interactive three-dimensional animation based on the ability to monitor and analyze technical and tactical skills, thereby better helping players and coaches analyze and improve technical and tactical skills.
[0081] Figure 4 An example animation screen presented by a system for presenting a three-dimensional animation of a tennis game according to some embodiments of the present invention is shown.
[0082] Animation screen 410 presents real-time data analysis of a badminton match, and the presented technical and tactical data include the current number of shots (4 shots), the speed of the last ball across the net of both players (50 km / h, 78 km / h), and the movement trajectories of both players.
[0083] Animation screen 450 presents a player's jumping data analysis, and the presented technical and tactical data include the player's jump times (27 times), average jump height (0.07 meters), and the location of each jump point.
[0084] Figure 5 An overall flow chart of a system for presenting three-dimensional animation of tennis motions according to some embodiments of the present invention is shown.
[0085] After the process starts, the video capture module 510 first captures motion videos from multiple perspectives in a time-synchronous manner to generate multiple single-perspective videos.
[0086] Each generated single-view video can be processed by a camera parameter estimation module, a ball positioning module, and a player posture detection module, respectively. Specifically, the camera parameter estimation module determines camera parameters based on each single-view video, the ball positioning module determines a two-dimensional trajectory of the ball based on each single-view video, and the player posture detection module determines a two-dimensional posture of a human target based on each single-view video.
[0087] Subsequently, the camera parameters generated for each single-view video, the 2D trajectory of the ball, and the 2D pose of the human target may be integrated to determine a 3D motion result 520. The 3D motion result may include the 3D trajectory of the ball and the 3D pose of the player.
[0088] The 3D motion result 520 can then be processed and analyzed by the technical and tactical analysis module 530 to obtain technical and tactical data. The 3D motion result 520 can also be rendered by the 3D rendering module 540 to output a rendered video.
[0089] According to another aspect of the present invention, a television broadcasting system for tennis sports is provided.
[0090] Figure 6 A block diagram of a television broadcast system 600 for tennis sports is shown in accordance with some embodiments of the present invention.
[0091] The television broadcasting system 600 may include a server 610 , a data exchange system 620 , a frame synchronization card 630 , an online packaging system 640 , a super slow motion system 650 , a switcher 660 , and a television signal broadcasting system 670 .
[0092] The server 610 may include the system 300 for presenting three-dimensional animation described above in conjunction with 3 (and including the technical and tactical analysis module 390), which may provide the determined technical and tactical data (e.g., the technical and tactical analysis module 390 described above in conjunction with Figure 3 The technical and tactical data 395 described above) and the generated rendered video (for example, the above combination Figure 3 3D animation described in 380).
[0093] The technical and tactical data may be provided to the online packaging system 640 through the data exchange system. The online packaging system 640 may beautify and package the received technical and tactical data to obtain a key signal.
[0094] The rendered video is transmitted to the super slow motion system 650 through the frame synchronization card 630. The super slow motion system 650 performs recording and editing on the received rendered video to generate a super slow motion picture.
[0095] The key signal generated by the online packaging system 640 and the super slow motion picture generated by the super slow motion system 650 are provided to the switcher 660. The switcher integrates the key signal and the super slow motion picture to generate a broadcast picture.
[0096] The broadcast images are transmitted to the television signal broadcast system 670. The television signal broadcast system 670 broadcasts television signals to television viewers based on the broadcast images.
[0097] The television broadcasting system 600 according to some embodiments of the present invention is capable of broadcasting the determined technical and tactical data and rendered video of the tennis sport to television viewers.
[0098] In some embodiments, television broadcast system 600 may include multiple time-synchronized cameras (not shown) for capturing multiple video frames of a motion.
[0099] Figure 7 A schematic diagram showing a camera arrangement 700 for video acquisition cameras used in a television broadcast system for tennis sports according to some embodiments of the present invention is shown.
[0100] Figure 7 The camera arrangement 700 of the video acquisition cameras for the badminton court 710 is shown. The video acquisition cameras 721, 722, 723, 724, 725, 726 can be arranged around the badminton court 710 to capture the motion data in the badminton court 710. The video acquisition cameras 721, 722, 723, 724, 725, 726 can be set as time-synchronized cameras, and their shooting frame rates are variable, and the default frame rate can be 50 frames per second. The video acquisition cameras 721, 722, 723, 724, 725, 726 can be coupled to the server 730 via optical fibers, and the collected signals are transmitted to the server 730 via optical fibers. The server 730 can be a combination of the above Figure 6 The server 610 described above. The server 730 can run relevant software modules to save, process and calculate and analyze multiple channels of video, so as to obtain the three-dimensional position of the ball and the three-dimensional posture of the players. The server 730 can further provide the three-dimensional position of the ball and the three-dimensional posture of the players for further technical and tactical analysis, or perform three-dimensional visualization output through three-dimensional rendering, or broadcast to television viewers through a television broadcast system.
[0101] Figure 7 The number and position of the video acquisition cameras shown in are exemplary. In some embodiments, more or fewer video acquisition cameras can be used. In some embodiments, the position of each video acquisition camera can be adjusted.
[0102] According to another aspect of the present invention, a method for monitoring technical and tactical capabilities of tennis sports is provided.
[0103] Figure 8A flow chart of a method 800 for monitoring technical and tactical capabilities of tennis sports according to some embodiments of the present invention is shown. The method 800 may be combined with the above Figure 1 The technical and tactical capability monitoring system 100 is described as performing, but the scope of the present invention is not limited in this regard.
[0104] The method 800 may include, at block 810 , receiving a plurality of video frames of motion captured by a plurality of time-synchronized cameras at a plurality of viewing angles.
[0105] The method 800 may include, at block 820 , calculating camera parameters based on position information of elements associated with the pitch in a plurality of video frames using a neural network model.
[0106] The method 800 may include, at block 830 , determining a two-dimensional position of a ball from a plurality of video frames using a deep neural network model.
[0107] The method 800 may include, at block 840 , determining a three-dimensional position of the ball based on the determined two-dimensional position of the ball and the camera parameters.
[0108] Method 800 may include, at block 850 , determining two-dimensional poses of all detectable human targets from a plurality of video frames.
[0109] Method 800 may include, at block 860 , determining a three-dimensional pose of the human target based on the determined two-dimensional pose of the human target and camera parameters.
[0110] The method 800 may include, at block 870 , determining a three-dimensional pose of the player through a three-dimensional pose of the human target based on three-dimensional spatial geometric relationship constraints associated with the court.
[0111] According to some embodiments of the present invention, the technical and tactical ability monitoring method 800 can effectively alleviate the occlusion problem by performing visual processing on video frames collected from multiple perspectives during tennis sports, and the results are highly accurate; in addition, the method 800 can also provide the three-dimensional position of the ball and the three-dimensional posture of the player, thereby providing a solid foundation for further in-depth and extensive technical and tactical analysis.
[0112] Fig. 9 A flowchart of a process 900 for calculating camera parameters according to some embodiments of the present invention is shown. The process 900 may be a specific implementation of the steps in block 820 described above in conjunction with the method 800, but the scope of the present invention is not limited thereto.
[0113] Process 900 may include, at block 910 , detecting a plurality of key points of a court in a plurality of video frames.
[0114] Process 900 may include, at block 920 , matching a plurality of key points to a predefined standard three-dimensional court template.
[0115] Process 900 may include, at block 930, estimating camera parameters using an algorithm based on the matching results. In some embodiments, the steps in block 930 may include estimating camera parameters based on homography matching errors between homography transformations of multiple key points and homography transformations of the standard 3D court template.
[0116] Fig.10 A flowchart of a process 1000 for training a neural network model for calculating camera parameters according to some embodiments of the present invention is shown. Process 1000 may be used to train the above-mentioned neural network model. Figure 1 The camera parameter module 110 of the technical and tactical capability monitoring system 100 described herein or in combination with Figure 8 The steps in block 820 of method 800 are described using a neural network, but the scope of the present invention is not limited in this regard.
[0117] Process 1000 may include, at block 1010 , arbitrarily manually annotating a threshold number of keypoints in each video frame, the threshold number being less than the number of the plurality of keypoints detected.
[0118] Process 1000 may include, at block 1020 , computing a homography transform of a court in a video frame to a standard three-dimensional court template for a threshold number of key points.
[0119] Process 1000 may include, at block 1030 , mapping intersection points of court lines associated with the court from a standard three-dimensional court template into a video frame based on a result of the homography transformation.
[0120] Process 1000 may include, at block 1040 , comparing the mapped intersection points to corresponding intersection points in a standard three-dimensional court template.
[0121] Process 1000 may include, at block 1050 , the result of the manual fine-tuning comparison indicates the location of one or more intersection points of maximum error in the video frame.
[0122] Process 1000 may include, at block 1060 , adjusting the homography transformation using a method of minimizing remapping errors based on the fine-tuned positions of the one or more intersection points.
[0123] Fig.11 A flow chart illustrating a process 1100 of determining a two-dimensional position of a ball according to some embodiments of the present invention. Process 1100 may be a specific implementation of the steps in block 830 described above in conjunction with method 800, but the scope of the present invention is not limited in this regard.
[0124] Process 1100 may include, at block 1100 , generating a heat map indicating areas where a ball is likely to appear and the probability density of appearing in the areas.
[0125] Process 1100 may include, at block 1120 , determining the position of the ball from the heat map using a contour extraction method and / or a preferred maximum contour method.
[0126] Optionally, the process 1100 may further include: adjusting the plurality of video frames to a fixed predetermined resolution at block 1130. The two-dimensional position of the ball is determined based on the video frames with the adjusted resolution.
[0127] Optionally, process 1100 may further include: at box 1140, receiving a plurality of consecutive video frames as input of the deep neural network.
[0128] Optionally, the process 1100 may further include: at box 1150, performing a separate calculation on the background in the video frame at the same viewing angle, and superimposing the calculation result on the video frame.
[0129] Fig.12 A flow chart illustrating a process 1200 for determining a three-dimensional pose of a player according to some embodiments of the present invention. Process 1200 may be a specific implementation of the steps in block 870 described above in conjunction with method 800, but the scope of the present invention is not limited in this regard.
[0130] Process 1200 may include: at block 1210 , estimating a position range where the player will appear in a plurality of video frames at a next moment based on the three-dimensional pose of the player and the camera parameters by using a back-projection technique.
[0131] Process 1200 may include, at block 1230 , skipping target detection within the position range and directly estimating a two-dimensional pose of the player at a next moment.
[0132] Process 1200 may include, at block 1250 , skipping the matching task and directly performing single-person triangulation for the player for the estimated next-time two-dimensional pose of the player.
[0133] According to another aspect of the present invention, a computer-readable storage medium is provided.
[0134] Fig.13 A block diagram of a machine-readable storage medium 1300 is shown according to some embodiments of the present invention.
[0135] The computer readable storage medium 1300 may include computer instructions 1350 stored thereon. The computer instructions 1350, when processed, may implement the above-mentioned Figure 8 The method described in 800, combined with Fig. 9 The process described 900, combined Fig.10 Described process 1000, combined Fig.11 The process described in 1100, combined Fig.12 Process 1200 is described.
[0136] According to another aspect of the present invention, a computer program product is provided.
[0137] Fig.14 A block diagram of a computer program product 1400 is shown according to some embodiments of the invention.
[0138] The computer program product 1400 may include computer instructions 1450. The computer instructions 1450, when processed, may implement the above-mentioned Figure 8 The method described in 800, combined with Fig. 9 The process described 900, combined Fig.10 Described process 1000, combined Fig.11 The process described in 1100, combined Fig.12 Process 1200 is described.
[0139] Embodiments of the present invention have been described with reference to the accompanying drawings. The embodiments are illustrative rather than restrictive.
Claims
1. A system for monitoring technical and tactical capabilities of tennis sports, comprising: Camera parameter estimation module for: receiving multiple video frames of motion captured by multiple time-synchronized cameras at multiple viewpoints; calculating camera parameters based on position information of features associated with the court in the plurality of video frames using a neural network model; Ball positioning module for: determining a two-dimensional position of a ball from the plurality of video frames using a deep neural network model; determining a three-dimensional position of the ball based on the determined two-dimensional position of the ball and the camera parameters; as well as Player posture detection module, used to: determining two-dimensional poses of all detectable human objects from the plurality of video frames; Determining a three-dimensional posture of the human target based on the determined two-dimensional posture of the human target and the camera parameters; as well as The three-dimensional posture of the player is determined through the three-dimensional posture of the human target based on the three-dimensional spatial geometric relationship constraints associated with the court.
2. The system of claim 1, wherein: The camera estimation module is used to: detecting a plurality of key points of the court in the plurality of video frames; Matching the plurality of key points with a predefined standard three-dimensional court template; and Based on the matching results, an algorithm is used to estimate the camera parameters.
3. The system of claim 2, wherein: The camera estimation module is used to: The camera parameters are estimated based on a homography matching error between the homography transformation of the plurality of key points and the homography transformation of the standard three-dimensional stadium template.
4. The system of claim 3, wherein: The neural network model is trained through the following steps: Manually annotating a threshold number of key points arbitrarily in each video frame, the threshold number being less than the number of the plurality of key points detected; Calculating a homography transformation between the court in the video frame and the standard three-dimensional court template for the threshold number of key points; Based on the result of the homography transformation, mapping the intersection points of the court lines associated with the court from the standard three-dimensional court template to the video frame; comparing the mapped intersection points with corresponding intersection points in the standard three-dimensional court template; Manually fine-tuning the position of one or more intersection points in the video frame where the comparison result indicates a maximum error; The homography is adjusted based on the fine-tuned positions of the one or more intersection points by minimizing the remapping error.
5. The system of claim 1, wherein: The ball positioning module is also used to perform at least one of the following steps: resizing the plurality of video frames to a fixed predetermined resolution, wherein a two-dimensional position of the ball is determined based on the resized video frames; Receiving a plurality of consecutive video frames as input of the deep neural network; The background in the video frame under the same viewing angle is calculated separately, and the calculation result is superimposed on the video frame.
6. The system of claim 1, wherein: The ball positioning module is used for: generating a heat map indicating areas where the ball is likely to appear and the probability density of appearing in the areas; and The position of the ball is determined from the heat map using a contour extraction method and / or preferably a maximum contour method.
7. The system of claim 1, wherein: The player posture detection module is used to: Based on the three-dimensional spatial geometric relationship constraints associated with the court, a predetermined number of players' three-dimensional postures are screened from all detectable three-dimensional postures of human targets, wherein the predetermined number is determined at least in part based on the number of players specified in the sport.
8. The system of claim 1, wherein: The player posture detection module is used to perform the following steps for each tracked player: Based on the three-dimensional posture of the player and the camera parameters, estimating the position range of the player in the multiple video frames at the next moment by back-projection technology; and The target detection is skipped within the position range and the two-dimensional posture of the player at the next moment is directly estimated.
9. The system of claim 8, wherein: The player posture detection module is also used to perform the following steps for each tracked player: For the estimated two-dimensional pose of the player at the next moment, the matching task is skipped and single-person triangulation of the player is directly performed.
10. A technical and tactical ability analysis system for tennis sports, comprising: A system as claimed in any one of claims 1 to 9; as well as Technical and tactical analysis module for: Technical and tactical data are determined based on the determined three-dimensional position of the ball and / or the three-dimensional posture of the player.
11. A system for presenting a three-dimensional animation of a tennis sport, comprising: A system as claimed in any one of claims 1 to 9; as well as 3D rendering module for: receiving data indicating a three-dimensional position of a ball and / or a three-dimensional pose of a player determined over a period of time; as well as The received data is rendered into a three-dimensional animation.
12. The system of claim 11, further comprising: Technical and tactical analysis module for: determining technical and tactical data based on the determined three-dimensional position of the ball and / or the three-dimensional posture of the player, and The three-dimensional rendering module is also used for: Visual rendering of the determined technical and tactical data.
13. A television broadcasting system for tennis sports, comprising: A system as claimed in any one of claims 1 to 9; A technical and tactical analysis module, for determining technical and tactical data based on the determined three-dimensional position of the ball and / or the three-dimensional posture of the player; 3D rendering module for: receiving data indicating a three-dimensional position of a ball and / or a three-dimensional pose of a player determined over a period of time; as well as Rendering the received data into three-dimensional animation and / or visual rendering of the technical and tactical data determined to obtain a rendered video; An online packaging module is used to beautify and package the technical and tactical data to obtain a key signal; A super slow motion system, used for recording and editing the rendered video to generate a super slow motion picture; A switcher, for integrating and processing the key signal and the super slow motion picture to generate a broadcast picture; as well as A television signal broadcasting system is used to broadcast television signals to television viewers based on the broadcast images.
14. The television broadcasting system according to claim 13, further comprising: Multiple time-synchronized cameras for capturing multiple video frames of motion at multiple viewpoints.
15. A method for monitoring technical and tactical capabilities of tennis sports, comprising: receiving multiple video frames of motion captured by multiple time-synchronized cameras at multiple viewpoints; calculating camera parameters based on position information of features associated with the court in the plurality of video frames using a neural network model; determining a two-dimensional position of a ball from the plurality of video frames using a deep neural network model; determining a three-dimensional position of the ball based on the determined two-dimensional position of the ball and the camera parameters; determining two-dimensional poses of all detectable human objects from the plurality of video frames; Determining a three-dimensional posture of the human target based on the determined two-dimensional posture of the human target and the camera parameters; as well as The three-dimensional posture of the player is determined through the three-dimensional posture of the human target based on the three-dimensional spatial geometric relationship constraints associated with the court.
16. The method of claim 15, wherein: Calculating camera parameters includes: detecting a plurality of key points of the court in the plurality of video frames; Matching the plurality of key points with a predefined standard three-dimensional court template; and Based on the matching results, an algorithm is used to estimate the camera parameters.
17. The method of claim 16, wherein: Estimating the camera parameters includes: estimating the camera parameters based on a homography matching error between the homography transformation of the plurality of key points and the homography transformation of the standard three-dimensional stadium template, and The neural network model is trained through the following steps: Manually annotating a threshold number of key points arbitrarily in each video frame, the threshold number being less than the number of the plurality of key points detected; Calculating a homography transformation between the court in the video frame and the standard three-dimensional court template for the threshold number of key points; Based on the result of the homography transformation, mapping the intersection points of the court lines associated with the court from the standard three-dimensional court template to the video frame; comparing the mapped intersection points with corresponding intersection points in the standard three-dimensional court template; Manually fine-tuning the position of one or more intersection points in the video frame where the comparison result indicates a maximum error; The homography is adjusted based on the fine-tuned positions of the one or more intersection points by minimizing the remapping error.
18. The method of claim 15, wherein: Determining the two-dimensional position of the ball includes at least one of the following steps: resizing the plurality of video frames to a fixed predetermined resolution, wherein a two-dimensional position of the ball is determined based on the resized video frames; Receiving a plurality of consecutive video frames as input of the deep neural network; The background in the video frame under the same viewing angle is calculated separately, and the calculation result is superimposed on the video frame.
19. The method of claim 15, wherein: Determining the 2D position of the ball involves: generating a heat map indicating areas where the ball is likely to appear and the probability density of appearing in the areas; and The position of the ball is determined from the heat map using a contour extraction method and / or preferably a maximum contour method.
20. The method of claim 15, wherein: Determining the three-dimensional posture of the player through the three-dimensional posture of the human target includes: Based on the three-dimensional spatial geometric relationship constraints associated with the court, a predetermined number of players' three-dimensional postures are screened from all detectable three-dimensional postures of human targets, wherein the predetermined number is determined at least in part based on the number of players specified in the sport.
21. The method of claim 15, wherein: Determining the three-dimensional posture of the player through the three-dimensional posture of the human target includes: Based on the three-dimensional posture of the player and the camera parameters, the position range of the player appearing in the multiple video frames at the next moment is estimated by back-projection technology; Skipping target detection within the position range and directly estimating the two-dimensional posture of the player at the next moment; and For the estimated two-dimensional pose of the player at the next moment, the matching task is skipped and single-person triangulation of the player is directly performed.
22. A computer-readable storage medium having computer instructions stored thereon, wherein the computer instructions, when executed by a processor, implement the steps of the method according to any one of claims 15 to 21.
23. A computer program product comprising computer instructions, which, when executed by a processor, implement the steps of the method according to any one of claims 15 to 21.
Citation Information
Patent Citations
Badminton player technical and tactical analysis method based on sequence pattern mining
CN109977752A
Badminton motion track monitoring and analyzing system and method
CN110433471A
Basketball motion analysis method based on 3D attitude estimation
CN111724414A
Intelligent auxiliary training method and system for badminton sports
CN116958872A
Ballgame video analysis device, ballgame video analysis method, and computer program
WO2021192768A1
Cited By
Intelligent inspection and analysis method and system for special equipment based on image acquisition
CN120411891A
Ball game video special effect generation method and device and storage medium
CN121603698A
Methods, equipment, and storage media for generating special effects in ball sports videos
CN121603698B
