A return determination training method for a tennis ball serving machine and related device

Abstract

The embodiment of the application discloses a return determination training method of a ball serving machine and a related device, and is used for stabilizing return determination training. The application is applied to a control module of a ball serving machine, the ball serving machine further comprises a long-focus camera and a short-focus camera, and the method comprises the following steps: controlling the long-focus camera to acquire a plurality of long-focus images of a target court in a preset time window, controlling the short-focus camera to acquire a plurality of short-focus images of the target court in the time window, and processing the long-focus images to obtain a target image; determining a training person probability of each candidate person in the target image according to the target image, and determining a target region corresponding to the target court according to the short-focus image; determining a training person from a plurality of candidate persons according to all training person probabilities and the target region; determining a return area time proportion as a ratio of a number of frames in which court coordinates of the training person in the time window are in a preset return area to the time window; and if the return area time proportion is greater than a first preset threshold, it is determined that the training person stably returns.

Description

Technical Field

[0001] This application relates to the field of ball-serving machines, and more particularly to a method and related apparatus for training ball-serving machine return-to-position determination. Background Technology

[0002] In ball sports training (such as tennis, badminton, and table tennis), the trainee needs to move from the center preparation area to the ball's landing area to receive the ball and then return to the center preparation area. The center preparation area is generally called the return-to-position area, and this process is usually called the return-to-position training. With the continuous development of technology, trainees can use ball machines for return-to-position training. Existing ball machines can serve balls according to a fixed rhythm, fixed landing point, and preset frequency.

[0003] However, in the existing solutions, the existing ball-serving machines cannot determine whether the ball has returned to its original position, and even more so, they cannot distinguish between "passing through the return zone" and "stable return". Summary of the Invention

[0004] To address the technical problem of the inability to determine stable return position in the prior art, this application provides a return position determination training method and related device for ball-serving machines, used for stable return position determination training.

[0005] The first aspect of this application provides a training method for determining the return position of a ball-serving machine, applied to the control module of the ball-serving machine. The ball-serving machine also includes a telephoto camera and a short-focus camera. The method includes: Control the telephoto camera to acquire multiple telephoto images of the target stadium within a preset time window, control the short-focus camera to acquire multiple short-focus images of the target stadium within the time window, and process the telephoto images to obtain the target image; The probability of a trainee for each candidate in the target image is determined based on the target image, and the target area corresponding to the target court is determined based on the short-focus image. The trainer is determined from multiple candidates based on the trainer probability and target region of all candidates. The ratio of the number of frames in which the trainee's court coordinates are within the preset return zone to the time window is determined as the return zone time percentage. The trainee's court coordinates are obtained from the target image and the short-focus image. If the proportion of time spent in the return zone is greater than the first preset threshold, then the trainee is determined to have achieved stable return to position.

[0006] Optionally, if the proportion of time spent in the retracement zone is greater than a first preset threshold, the method may further include the following steps before determining that the trainee has achieved stable retracement: The ratio of the vector sum of multiple court coordinates of the trainee within the time window to the time window is determined as the position mean, and the ratio of the sum of the squared Euclidean distances between each court coordinate of the trainee and the position mean to the time window is determined as the position variance; the average of multiple movement speeds of the trainee is determined as the speed mean, where the movement speed is obtained within the time window based on multiple court coordinates corresponding to multiple frames of images of the trainee in the court coordinate system. If the proportion of time spent in the retracement zone is greater than the first preset threshold, then the trainee is determined to have achieved stable retracement, including: If the preset conditions are met, the trainee is determined to have returned to the starting position stably. The preset conditions include at least one of the following: the proportion of time spent in the return zone is greater than the first preset threshold, the position variance is less than the second preset threshold, the average speed is less than the third preset threshold, and the trainee's orientation is aligned with the ball machine. The preset conditions include at least the proportion of time spent in the return zone being greater than the first preset threshold. The trainee's orientation is calculated based on the trainee's key body points.

[0007] Optionally, the target area corresponding to the target court is determined based on the short-focus image, including: Based on the preset key point detection algorithm, key point detection processing is performed on the short focal length image to obtain multiple corresponding short focal length key points. The corresponding first plane homography matrix is ​​determined according to the long focal length intrinsic parameters of the long focal length camera, the short focal length intrinsic parameters of the short focal length camera, the preset short focal length to long focal length rotation matrix, the preset short focal length to long focal length translation matrix, the transpose of the normal vector in the preset court plane formula, and the constants in the court plane formula. For each short focal length key point, the product of the first plane homography matrix and the short focal length coordinates of the short focal length key point is determined as the long focal length coordinates, and the area enclosed by the long focal length coordinates of all short focal length key points is determined as the target area.

[0008] Optionally, the trainer can be determined from multiple candidates based on the trainer probability and target region of all candidates, including: The candidate stadium coordinates in the corresponding stadium coordinate system are calculated based on the candidate image coordinates of multiple candidates in the target image and the second plane homography matrix. The second plane homography matrix is ​​the homography matrix of the plane from the target area to the target stadium. For each candidate, the ratio of the number of frames in which the candidate's court coordinates are within the target area to the time window is determined as the in-court percentage. The corresponding score is obtained by weighting the in-court percentage, the trainee probability, the preset in-court switching coefficient, and three preset weight coefficients. The candidate with the highest score among all candidates is selected as the trainer.

[0009] Optionally, the target image can be obtained by processing the telephoto and short-focus images, including: Calculate the coordinates of candidate images in the telephoto image and determine the region of interest image containing the area where the candidate image coordinates are located; The region of interest image is extracted and processed to obtain the corresponding original RAW image and the red, green and blue three-channel RGB image; Based on the preset feature extraction model, the RAW image is input into the RAW feature extraction module in the feature extraction model to obtain the RAW features output by the RAW feature extraction module, and the RGB image is input into the RGB feature extraction module in the feature extraction model to obtain the RGB features output by the RGB feature extraction module. RAW and RGB features are concatenated and fused to obtain fused features. Based on a preset enhancement algorithm, corresponding enhanced features are generated from the fused features. The enhanced features are then subjected to convolution and pooling to obtain the corresponding Image Signal Processor (ISP) parameters. The ISP parameters include exposure compensation, gamma correction, contrast, saturation, sharpening, noise reduction, contrast limiting adaptive histogram equalization cropping threshold, and image enhancement weights. The region of interest (ROI) image is processed according to the ISP parameters to obtain an enhanced image. Based on a weighted sum algorithm, the corresponding target image is calculated from the enhanced image, image enhancement weights, and ROI image.

[0010] Optionally, if the ball-serving machine also includes a ball-serving module, the method further includes, after confirming the trainee's stable return to position: The target area is divided into a return zone and multiple receiving zones based on the user's input division instructions; Based on the mode selection command input by the user, the mode corresponding to the mode selection command is determined as the target mode, and the target receiving area is determined from multiple receiving areas based on the target mode; Based on the preset relationship between trainee performance and difficulty adjustment parameters, the corresponding target difficulty adjustment parameters are determined according to the trainee's target historical consecutive successful receptions and the average time taken to return to position. Trainee performance includes the historical successful receptions and the average time taken to return to position. The current serving parameters are adjusted according to the target difficulty adjustment parameters to obtain the target serving parameters, which include serving speed, serving frequency, left and right landing point span, front and back landing point span, and spin intensity. The trainee's dominant hand is determined based on the user's input of the dominant hand selection command, real-time racket hand recognition data, or historical hitting data; the same side or opposite side of the dominant hand is determined as the serving side; The serve is executed based on the target receiving area, target serving parameters, and the serving side control module.

[0011] Optionally, if the number of receiving zones is two and the target mode is an alternating training mode, then the corresponding target receiving zone is determined from multiple receiving zones according to the target mode, including: Obtain the most recent receiving area from historical hitting data, and identify the other receiving area among multiple receiving areas besides the most recent receiving area as the target receiving area; or, If the target pattern is random, the corresponding target receiving area is determined from multiple receiving areas based on the target pattern, including: The current target receiving area is any receiving area that has been continuously identified as a historical target receiving area for no more than a preset threshold number of times. or, If the target pattern is a weak-side strengthening pattern, then the corresponding target receiving area is determined from multiple receiving areas based on the target pattern, including: For each receiving area, based on a preset weighted sum algorithm, the corresponding target performance score is calculated according to the time from the return area to the receiving area, the time from the receiving area to the return area, and the shot success rate obtained from historical shot data; Based on the preset relationship between performance score and serve probability, the target serve probability of each receiving area is determined according to the target performance score of all receiving areas, and any receiving area is determined as the target receiving area based on the target serve probability. The performance score and serve probability are negatively correlated.

[0012] Optionally, the trainee probability of each candidate in the target image is determined based on the target image, including: Based on the preset recognition model, the target image is input into the feature extraction module of the recognition model to obtain the multi-scale visual features output by the feature extraction module. The multi-scale visual features are then input into the fusion module of the recognition model for fusion processing to obtain the trainee probability of each candidate output by the fusion module.

[0013] A second aspect of this application provides a repositioning determination training device, comprising: The acquisition unit is used to acquire multiple long-focus images of the target stadium within a preset time window, acquire multiple short-focus images of the target stadium within the time window, and process the long-focus images to obtain the target image. The determination unit is used to determine the trainee probability of each candidate in the target image based on the target image, and to determine the target area corresponding to the target court based on the short-focus image; The unit is also used to determine the trainer from multiple candidates based on the trainer probability and target region of all candidates; The determining unit is also used to determine the ratio of the number of frames in which the trainee's court coordinates are within a preset return zone to the time window as the return zone time proportion. The trainee's court coordinates are obtained based on the target image and the short-focus image. The unit is also used to determine that the trainee has a stable return to position if the proportion of the return time is greater than a first preset threshold.

[0014] A third aspect of this application provides a return-to-position determination training device, comprising: Central processing unit, memory, and input / output interfaces; The memory can be either temporary or permanent storage. The central processing unit is configured to communicate with the memory and execute instructions in the memory to perform the aforementioned method.

[0015] A fourth aspect of this application provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the aforementioned method.

[0016] As can be seen from the above technical solutions, the embodiments of this application have the following advantages: First, a telephoto camera acquires multiple frames of telephoto images of the target stadium within a preset time window. A short-focus camera acquires multiple frames of short-focus images of the target stadium within the same time window. The telephoto images are then processed to obtain the target image. Next, the probability of each candidate trainee in the target image is determined, and the target region corresponding to the target stadium is determined based on the short-focus images. Then, based on the trainee probabilities of all candidates and the target region, the trainee is selected from multiple candidates. The ratio of the number of frames in which the trainee's stadium coordinates fall within a preset return zone to the time window is determined as the return zone time percentage. Finally, if the return zone time percentage is greater than a first preset threshold, the trainee is considered to have stably returned to position. Analyzing multiple real-time images and determining whether the trainee has stably returned to position based on the return zone time percentage allows for differentiation between passing through the return zone and stable return, providing a correct foundation for subsequent analysis.

[0017] Additional aspects and advantages of embodiments of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of embodiments of this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of an embodiment of the repositioning determination training method disclosed in this application; Figure 2This is a schematic diagram of another embodiment of the repositioning determination training method disclosed in this application; Figure 3 This is a schematic diagram of the long and short focal length fusion court boundary identification disclosed in this application; Figure 4 A schematic diagram of a specific embodiment of the position determination training disclosed in this application; Figure 5 This is a schematic diagram of the follow-back logic disclosed in this application; Figure 6 This is a schematic diagram of the receiving and returning process disclosed in this application; Figure 7 This is a schematic diagram of an embodiment of the repositioning determination training device disclosed in this application; Figure 8 This is a schematic diagram of another embodiment of the return-to-position determination training device disclosed in this application.

[0020] The realization of the objectives, functional features and advantages of the embodiments of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] Ball machines have become an indispensable part of ball sports training. Existing ball machines can serve balls according to a fixed rhythm, fixed landing point, and preset frequency, and can perform return-to-position training. However, existing solutions only analyze single-frame images to determine whether the trainee's position is within the return-to-position zone. This means that even when the trainee merely passes through the return-to-position zone, it is still identified as returning to the return-to-position zone, rather than a true stable return, failing to distinguish between "passing through the return-to-position zone" and "stable return." To solve the above technical problems, this application provides a return-to-position training method and related apparatus for ball machines, used for stable return-to-position training. By analyzing multiple real-time images and determining whether the trainee has achieved a stable return based on the proportion of time spent in the return-to-position zone, it can distinguish between passing through the return-to-position zone and a stable return, providing a correct basis for subsequent analysis.

[0022] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the embodiments of this application, and should not be construed as limiting the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the protection scope of the embodiments of this application.

[0023] In the description of the embodiments of this application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "circumferential", "radial", etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the embodiments of this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0024] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "first," "second," "third," "fourth," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. In the description of embodiments of this application, "a plurality of" means two or more, unless otherwise expressly specified.

[0025] In the embodiments of this application, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0026] In the embodiments of this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature being directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0027] The following describes a method for training position determination based on this application. Please refer to [link to relevant documentation]. Figure 1 One embodiment of the return-to-position determination training method of this application is applied to the control module of a ball-serving machine. The ball-serving machine also includes a telephoto camera and a short-focus camera. The method includes: 101. Control the telephoto camera to acquire multiple telephoto images of the target stadium within a preset time window, control the short-focus camera to acquire multiple short-focus images of the target stadium within the time window, and process the telephoto images to obtain the target image. The target court is the court where the ball machine and the trainee are located, while the time window can be set according to actual needs; no specific limit is set here. Specifically, the telephoto image is processed to facilitate subsequent operations.

[0028] 102. Determine the trainee probability of each candidate in the target image based on the target image, and determine the target area corresponding to the target court based on the short-focus image; Here, "candidates" refers to people in and around the target golf course, including trainers, coaches, and spectators. "Target area" corresponds to "target golf course" and indicates the area surrounding the target golf course.

[0029] 103. Based on the trainer probability and target region of all candidates, determine the trainer from multiple candidates; The trainer probability is the probability that the candidate is the trainer. Based on the trainer probability and the target area, the real trainer can be identified.

[0030] 104. The ratio of the number of frames in which the trainee's court coordinates are within the preset return zone within the time window to the time window is determined as the return zone time percentage. The trainee's court coordinates are obtained from the target image and the short-focus image. Simply put, the number of frames that meet the condition (the trainee's court coordinates are within the return zone in that frame) is divided by the time window to obtain the return zone time percentage, which can represent the stability of the trainee's return.

[0031] 105. If the proportion of time in the return zone is greater than the first preset threshold, then the trainee is determined to have stably returned to the position.

[0032] The first preset threshold can be set according to actual needs, generally any value between 0.6 and 0.8, and is not specifically limited here. If the proportion of time in the return zone is greater than the first preset threshold, it indicates that the trainee has stably returned to the position within the time window.

[0033] In this embodiment, a telephoto camera is first controlled to acquire multiple frames of telephoto images of the target court within a preset time window, and a short-focus camera is controlled to acquire multiple frames of short-focus images of the target court within the time window. The telephoto images are then processed to obtain the target image. Next, the trainee probability of each candidate in the target image is determined based on the target image, and the target region corresponding to the target court is determined based on the short-focus images. Then, based on the trainee probabilities of all candidates and the target region, the trainee is determined from multiple candidates. The ratio of the number of frames in which the trainee's court coordinates are within a preset return zone to the time window is determined as the return zone time percentage. Finally, if the return zone time percentage is greater than a first preset threshold, the trainee is determined to have stably returned to position. By analyzing multiple real-time images and determining whether the trainee has stably returned to position based on the return zone time percentage, it is possible to distinguish between passing through the return zone and stably returning to position, providing a correct basis for subsequent analysis.

[0034] Please see Figures 2 to 6 Another embodiment of the return-to-position determination training method of this application is applied to the control module of a ball-serving machine. The ball-serving machine also includes a telephoto camera, a short-focus camera, and a ball-serving module. The method includes: 201. Control the telephoto camera to acquire multiple telephoto images of the target stadium within a preset time window, control the short-focus camera to acquire multiple short-focus images of the target stadium within the time window, and process the telephoto images to obtain the target image. The short-focus camera has a larger field of view, allowing for a more complete observation of the target court area, while the long-focus camera captures details of the candidates and the area. The system identifies key points of the court using the short-focus image and, combined with camera calibration relationships, maps the short-focus court boundaries onto the long-focus image.

[0035] Please see Figure 3 Specifically, the step "processing the telephoto image to obtain the target image" can be implemented by calculating the coordinates of the candidate image in the telephoto image and determining the region of interest image containing the area where the candidate image coordinates are located. The region of interest image is extracted and processed to obtain the corresponding original RAW image and the red, green and blue three-channel RGB image; Based on a pre-defined feature extraction model, the RAW image is input into the RAW feature extraction module of the model to obtain the RAW features output by the RAW feature extraction module. Similarly, the RGB image is input into the RGB feature extraction module of the model to obtain the RGB features output by the RGB feature extraction module. The model can be implemented using neural networks such as CNNs; specific implementation is not limited here.

[0036] RAW and RGB features are concatenated and fused to obtain fused features. Based on a preset enhancement algorithm, corresponding enhanced features are generated from the fused features. Convolution and pooling processing is applied to the enhanced features to obtain corresponding Image Signal Processor (ISP) parameters. ISP parameters include exposure compensation, gamma correction, contrast, saturation, sharpening, noise reduction, contrast-limited adaptive histogram equalization cropping threshold, and image enhancement weights. The region of interest (ROI) image is processed according to the ISP parameters to obtain an enhanced image. Based on a weighted sum algorithm, the corresponding target image is calculated using the enhanced image, image enhancement weights, and ROI image. Specifically, the sum of the product of the enhanced image and the image enhancement weights and the product of the ROI image and the difference between 1 and the image enhancement weights is the target image.

[0037] This step can improve the stability of human detection and tracking under conditions such as backlighting, low contrast, and when the human body and background colors are similar, while avoiding the high computing power and high latency issues caused by deep image enhancement networks directly generating images.

[0038] 202. Determine the training probability of each candidate in the target image based on the target image, and determine the target area corresponding to the target court based on the short-focus image; The trainer probability is the probability that the candidate is the trainer, which can be obtained by fusing lightweight backbone and neck features.

[0039] Specifically, the step "determining the trainee probability of each candidate in the target image based on the target image" involves, in one implementation, inputting the target image into the feature extraction module of the recognition model based on a preset recognition model. This yields multi-scale visual features output by the feature extraction module, which are then input into the fusion module of the recognition model for fusion processing. This results in the trainee probability of each candidate output by the fusion module. Depending on actual needs, the fusion module can output trainee probabilities, or it can output human detection boxes, human keypoints, hand positions, foot positions, and racket positions, etc. To reduce the computational overhead of embedded devices, modified structures of MobileNet, ShuffleNet, CSP-Lite, YOLOv8n, or other lightweight network structures can be used; specific details are not limited here. Furthermore, model training should avoid splitting multiple outputs into multiple serial models; a unified loss formula is required. The specific loss formula is not limited, but the quantities involved can include human detection loss, candidate classification loss, human keypoint loss, etc.

[0040] Specifically, the step "determining the target area corresponding to the target stadium based on the short-focus image" in one embodiment involves performing keypoint detection processing on the short-focus image based on a preset keypoint detection algorithm to obtain multiple corresponding short-focus keypoints. Then, based on the telephoto intrinsic parameters of the telephoto camera and the short-focus intrinsic parameters of the short-focus camera, a preset short-focus to telephoto rotation matrix, a preset short-focus to telephoto translation matrix, the transpose of the normal vector in the preset stadium plane formula, and constants in the stadium plane formula, the corresponding first planar homography matrix is ​​determined. Specifically, in one embodiment, the planar homography matrix is: ; in, This is the first plane homography matrix. For the court surface, This is a rotation matrix from short focal length to long focal length. This is the translation matrix from short focal length to long focal length. For long-range internal reference, It is a short-focus internal reference. It is the transpose of the normal vector. It is a constant.

[0041] For each short-focus keypoint, the product of the first plane homography matrix and the short-focus coordinates of the keypoint is used to determine the long-focus coordinates. The region enclosed by the long-focus coordinates of all short-focus keypoints is then defined as the target region. Specifically, after obtaining the long-focus coordinates, normalization is performed, and the region enclosed by all long-focus coordinates is determined as the target region.

[0042] This allows for the use of a short-focus camera to recover the effective court boundaries from the telephoto view when the field of view of the telephoto camera is narrow or the net obstructs the view, thus improving the stability of personnel area judgment and training area division.

[0043] 203. Based on the trainer probability and target region of all candidates, determine the trainer from multiple candidates; Specifically, in one implementation, the candidate image coordinates of multiple candidates in the target image and the second plane homography matrix are first calculated to obtain the candidate court coordinates in the corresponding court coordinate system. The second plane homography matrix is ​​the homography matrix of the plane from the target area to the target court. For each candidate, the ratio of the number of frames in which the candidate's court coordinates fall within the target area to the time window is determined as the in-court percentage. This in-court percentage is then weighted and calculated based on the trainee probability, a preset in-court / out-of-court switching coefficient, and three preset weighting coefficients to obtain the corresponding score. In one implementation, the formula is: ; ; in, Let i be the score for the i-th candidate. These are the weighting coefficients. For the percentage of the venue, This indicates whether the i-th candidate is within the target area at frame t. A value of 1 indicates the candidate's court coordinates are within the target area, and a value of 0 indicates otherwise. For time window, The probability of training this candidate. This is the switching coefficient between the on-site and off-site areas, used to reduce the weight of temporary visitors.

[0044] Finally, the candidate with the highest score among all candidates is selected as the trainer.

[0045] This step filters out players, coaches, ball retrievers, and passersby from the adjacent court, preventing the system from incorrectly controlling the serve based on the position of non-trained personnel.

[0046] 204. The ratio of the number of frames in which the trainee's court coordinates are within the preset return zone within the time window to the time window is determined as the return zone time percentage. The trainee's court coordinates are obtained from the target image and the short-focus image. The acquisition of the trainee's court coordinates is similar to that of the candidate court coordinates, and will not be elaborated further here. Simply put, the number of frames that meet the condition (the trainee's court coordinates are within the homing zone in that frame) is divided by the time window to obtain the homing zone time percentage, which indicates the stability of the trainee's homing.

[0047] 205. The ratio of the vector sum of multiple court coordinates of the trainee within the time window to the time window is determined as the position mean; the ratio of the sum of the squared Euclidean distances between each court coordinate of the trainee and the position mean to the time window is determined as the position variance; the mean of multiple movement speeds of the trainee is determined as the speed mean. The movement speed is obtained within a time window based on multiple court coordinates corresponding to multiple frames of images of the trainee in the court coordinate system, and can be calculated using the court coordinate sequence.

[0048] 206. If the preset conditions are met, the trainee is confirmed to have returned to a stable position; The preset conditions include at least one of the following: the proportion of time spent in the return zone is greater than the first preset threshold, the position variance is less than the second preset threshold, the average speed is less than the third preset threshold, and the trainee's orientation is aligned with the ball-launching machine. The preset conditions include at least the proportion of time spent in the return zone being greater than the first preset threshold, and the trainee's orientation is calculated based on the trainee's key body points.

[0049] Please see Figure 4 In one implementation, assuming the preset conditions include a return zone time ratio greater than a first preset threshold, a position variance less than a second preset threshold, and a mean velocity less than a third preset threshold, then the corresponding formula is: ; ; ; ; in, This represents the percentage of time spent in the return zone. The time window is typically 0.05 seconds to 2 seconds. This indicates whether the trainee is in the repositioning zone at frame t; a value of 1 indicates they are, and a value of 0 indicates they are. The mean of the positions, For the coordinates of the stadium, For location variance, The average speed Let be the velocity of the t-th frame.

[0050] 207. Divide the target area into a return zone and multiple receiving zones according to the user's input division instructions; The receiving area can be left, center, or right, or it can be divided into the baseline area, service line area, net area, etc. It can also be the forehand area, backhand area, overhead area, etc. It can be set according to actual needs, and there are no specific restrictions here.

[0051] 208. Determine the target receiving area, target serving parameters, and serving side, and control the serving module to serve based on the target receiving area, target serving parameters, and serving side.

[0052] Specifically, based on the user's input of a mode selection command, the mode corresponding to the command is determined as the target mode, and the target receiving area is determined from multiple receiving areas based on the target mode. There are at least three modes, which are illustrated below.

[0053] Specifically, based on the preset relationship between trainee performance and difficulty adjustment parameters, the corresponding target difficulty adjustment parameters are determined according to the trainee's historical consecutive successful receptions and the average time taken to return to position. Trainee performance includes the historical successful receptions and the average time taken to return to position. The current serving parameters are then adjusted according to the target difficulty adjustment parameters to obtain the target serving parameters, which include serving speed, serving frequency, left and right landing point span, front and back landing point span, and spin intensity. This enables adaptive difficulty training.

[0054] Specifically, the trainee's dominant hand is determined based on the user's input of the dominant hand selection command, real-time racket hand recognition data, or historical shot data. The same side or opposite side of the dominant hand is then designated as the serving side. Specifically, if the user inputs a forehand command, the same side of the dominant hand is designated as the serving side; if the user inputs a backhand command, the opposite side of the dominant hand is designated as the serving side. The horizontal coordinate can be used to determine whether the ball is on the dominant hand's side. Alternatively, a pre-defined mapping relationship between the dominant hand and the ball's landing point relative to left and right sides and forehand or backhand can be established; the specifics are not limited here. This avoids the problem of traditional ball machines being unsuitable for left-handed users.

[0055] There are at least three modes, and the specific ones are not limited here. The following are examples of three implementation methods.

[0056] In the first implementation, if there are two receiving areas and the target mode is an alternating training mode, then the step "determine the corresponding target receiving area from multiple receiving areas according to the target mode" specifically involves obtaining the most recent receiving area from historical shot data and determining the other receiving area among the multiple receiving areas besides the most recent receiving area as the target receiving area. For example, if the receiving areas are only A and B, and the target receiving area was A last time, then it will be B this time.

[0057] In the second implementation, if the target mode is a random mode, the step "determine the corresponding target receiving area from multiple receiving areas according to the target mode" specifically involves determining any receiving area among the multiple receiving areas whose number of consecutive determinations as a historical target receiving area does not exceed a preset threshold as the current target receiving area. The preset threshold is generally 2 to 4. For example, assuming a preset threshold of 3, if the ball has been randomly assigned to area A for 3 consecutive times, the next ball will definitely be assigned to area B. If the ball has been randomly assigned to area A for 2 consecutive times, the next ball will be randomly assigned to either area A or area B. Please refer to [link to relevant documentation]. Figure 5 and Figure 6 The ball machine serves the ball to area A. The trainee moves, returns the ball, and then returns to their original position. The ball machine determines whether the trainee has returned to their original position stably before deciding whether to serve from area A or area B.

[0058] In the third implementation, if the target mode is the weak-side enhancement mode, the step "determine the corresponding target receiving area from multiple receiving areas according to the target mode" specifically involves first calculating the corresponding target performance score for each receiving area based on a preset weighted sum algorithm, according to the time from the return area to the receiving area, the time from the receiving area to the return area, and the hitting success rate obtained from historical hitting data. Based on the preset relationship between performance score and serve probability, the target serve probability of each receiving area is determined according to the target performance score of all receiving areas, and any receiving area is determined as the target receiving area based on the target serve probability. The performance score and serve probability are negatively correlated.

[0059] It is understood that this embodiment is not limited to telephoto or short-focus cameras, but can also be a monocular wide-angle camera, a monocular telephoto camera, a binocular camera, a depth camera, a human eye camera, etc. The training position is not limited to the foot points of the human body frame, but can also be the ankle point, hip point, the center point of the human body, or a wearable tag, etc. The serving strategy can also include personalized strategies based on historical data, fatigue-based strategies, reinforcement learning, etc. Each step can be implemented by traditional algorithms, deep learning algorithms, rule-based state machines, statistical models, or a combination of multiple methods. For example, the court boundary can be implemented by traditional line detection, deep learning segmentation, or keypoint detection; target tracking can be implemented by Kalman Filter, SORT, ByteTrack, DeepSORT, or optical flow; the return state can be implemented by rule-based thresholding, HMM, RNN, Transformer, or a temporal classifier; and policy control can be implemented by rules, probabilistic models, reinforcement learning, or expert systems. Furthermore, it can be combined with gesture interaction control strategies. Moreover, local enhancement and lightweight multi-task models improve the stability of human body recognition and trajectory tracking under complex lighting conditions.

[0060] In this embodiment, the ball is served only after a stable return to position is determined, ensuring it matches the trainee's hitting rhythm. Multiple serving strategies are available for training to meet the trainee's needs. Furthermore, it can identify forehand and backhand for adaptation, adaptively adjusting difficulty and strengthening the weak side. Of course, this must be done after identifying the primary trainee to ensure accuracy.

[0061] The above describes a method for training a return-to-position determination according to an embodiment of this application. The following describes a device for training a return-to-position determination according to an embodiment of this application. Please refer to... Figure 7 One embodiment of a repositioning determination training device in this application includes: The acquisition unit 701 is used to acquire multiple long-focus images of the target stadium within a preset time window, acquire multiple short-focus images of the target stadium within the time window, and process the long-focus images to obtain the target image. The determining unit 702 is used to determine the trainee probability of each candidate in the target image based on the target image, and to determine the target area corresponding to the target court based on the short-focus image; The determination unit 702 is also used to determine the trainer from multiple candidates based on the trainer probability of all candidates and the target region; The determining unit 702 is also used to determine the ratio of the number of frames in which the trainee's court coordinates are within a preset return zone to the time window as the return zone time ratio, and the trainee's court coordinates are obtained based on the target image and the short-focus image. The determination unit 702 is also used to determine that the trainee has stably returned to position if the proportion of the return time is greater than the first preset threshold.

[0062] In this embodiment, the acquisition unit 701 first controls a telephoto camera to acquire multiple frames of telephoto images of the target stadium within a preset time window, and controls a short-focus camera to acquire multiple frames of short-focus images of the target stadium within the time window. The telephoto and short-focus images are then processed to obtain the target image. Next, the determination unit 702 determines the trainee probability of each candidate in the target image based on the target image, and determines the target region corresponding to the target stadium based on the short-focus images. Then, based on the trainee probabilities of all candidates and the target region, the trainee is determined from multiple candidates. The ratio of the number of frames in which the trainee's stadium coordinates are within a preset return zone to the time window is determined as the return zone time percentage. Finally, if the return zone time percentage is greater than a first preset threshold, the trainee is determined to have stably returned to position. By analyzing multiple real-time images and determining whether the trainee has stably returned to position based on the return zone time percentage, it is possible to distinguish between passing through the return zone and stably returning to position, providing a correct basis for subsequent analysis.

[0063] Based on the above-described embodiment of the return-to-position determination training device, another embodiment of the return-to-position determination training device of this application further includes: Furthermore, the device also includes a control unit for: The ratio of the vector sum of multiple court coordinates of the trainee within the time window to the time window is determined as the position mean, and the ratio of the sum of the squared Euclidean distances between each court coordinate of the trainee and the position mean to the time window is determined as the position variance; the average of multiple movement speeds of the trainee is determined as the speed mean, where the movement speed is obtained within the time window based on multiple court coordinates corresponding to multiple frames of images of the trainee in the court coordinate system. Determine the unit, specifically for: If the preset conditions are met, the trainee is determined to have returned to the starting position stably. The preset conditions include at least one of the following: the proportion of time spent in the return zone is greater than the first preset threshold, the position variance is less than the second preset threshold, the average speed is less than the third preset threshold, and the trainee's orientation is aligned with the ball machine. The preset conditions include at least the proportion of time spent in the return zone being greater than the first preset threshold. The trainee's orientation is calculated based on the trainee's key body points.

[0064] Furthermore, the specific use of the unit is as follows: Based on the preset key point detection algorithm, key point detection processing is performed on the short focal length image to obtain multiple corresponding short focal length key points. The corresponding first plane homography matrix is ​​determined according to the long focal length intrinsic parameters of the long focal length camera, the short focal length intrinsic parameters of the short focal length camera, the preset short focal length to long focal length rotation matrix, the preset short focal length to long focal length translation matrix, the transpose of the normal vector in the preset court plane formula, and the constants in the court plane formula. For each short focal length key point, the product of the first plane homography matrix and the short focal length coordinates of the short focal length key point is determined as the long focal length coordinates, and the area enclosed by the long focal length coordinates of all short focal length key points is determined as the target area.

[0065] Furthermore, the specific use of the unit is as follows: The candidate field coordinates in the corresponding field coordinate system are calculated based on the candidate image coordinates of multiple candidates in the target image and the second plane homography matrix. The second plane homography matrix is ​​the homography matrix of the plane from the target area to the target field. For each candidate, the ratio of the number of frames in which the candidate's court coordinates are within the target area to the time window is determined as the in-court percentage. The corresponding score is obtained by weighting the in-court percentage, the trainee probability, the preset in-court switching coefficient, and three preset weight coefficients. The candidate with the highest score among all candidates is selected as the trainer.

[0066] Furthermore, the acquisition unit is specifically used for: Calculate the coordinates of the candidate images in the telephoto image, and determine the region of interest image containing the area where the candidate image coordinates are located; The region of interest image is extracted and processed to obtain the corresponding original RAW image and the red, green and blue three-channel RGB image; Based on the preset feature extraction model, the RAW image is input into the RAW feature extraction module in the feature extraction model to obtain the RAW features output by the RAW feature extraction module, and the RGB image is input into the RGB feature extraction module in the feature extraction model to obtain the RGB features output by the RGB feature extraction module. RAW and RGB features are concatenated and fused to obtain fused features. Based on a preset enhancement algorithm, corresponding enhanced features are generated from the fused features. The enhanced features are then subjected to convolution and pooling to obtain the corresponding Image Signal Processor (ISP) parameters. The ISP parameters include exposure compensation, gamma correction, contrast, saturation, sharpening, noise reduction, contrast limiting adaptive histogram equalization cropping threshold, and image enhancement weights. The region of interest (ROI) image is processed according to the ISP parameters to obtain an enhanced image. Based on a weighted sum algorithm, the corresponding target image is calculated from the enhanced image, image enhancement weights, and ROI image.

[0067] Furthermore, the control unit is also used for: The target area is divided into a return zone and multiple receiving zones based on the user's input division instructions; Based on the mode selection command input by the user, the mode corresponding to the mode selection command is determined as the target mode, and the target receiving area is determined from multiple receiving areas based on the target mode; Based on the preset relationship between trainee performance and difficulty adjustment parameters, the corresponding target difficulty adjustment parameters are determined according to the trainee's target historical consecutive successful receptions and the average time taken to return to position. Trainee performance includes the historical successful receptions and the average time taken to return to position. The current serving parameters are adjusted according to the target difficulty adjustment parameters to obtain the target serving parameters, which include serving speed, serving frequency, left and right landing point span, front and back landing point span, and spin intensity. The trainee's dominant hand is determined based on the user's input of the dominant hand selection command, real-time racket hand recognition data, or historical hitting data; the same side or opposite side of the dominant hand is determined as the serving side; The serve is executed based on the target receiving area, target serving parameters, and the serving side control module.

[0068] Furthermore, if the number of receiving zones is two, and the target mode is an alternating training mode, then the specific use of the unit is determined as follows: Obtain the most recent receiving area from historical hitting data, and identify the other receiving area among multiple receiving areas besides the most recent receiving area as the target receiving area; or, If the target pattern is a random pattern, the determining unit is specifically used for: The current target receiving area is any receiving area that has been continuously identified as a historical target receiving area for no more than a preset threshold number of times. or, If the target mode is a weak-side reinforcement mode, determine the specific use of the unit: For each receiving area, based on a preset weighted sum algorithm, the corresponding target performance score is calculated according to the time from the return area to the receiving area, the time from the receiving area to the return area, and the shot success rate obtained from historical shot data; Based on the preset relationship between performance score and serve probability, the target serve probability of each receiving area is determined according to the target performance score of all receiving areas, and any receiving area is determined as the target receiving area based on the target serve probability. The performance score and serve probability are negatively correlated.

[0069] Furthermore, the specific use of the unit is as follows: Based on the preset recognition model, the target image is input into the feature extraction module of the recognition model to obtain the multi-scale visual features output by the feature extraction module. The multi-scale visual features are then input into the fusion module of the recognition model for fusion processing to obtain the trainee probability of each candidate output by the fusion module.

[0070] The functions and processes performed by each unit in the return-to-position determination training device in this embodiment are the same as those described above. Figures 1 to 2 The functions and processes performed by the mid-position determination training device are similar, and will not be described in detail here.

[0071] Figure 8 This is a schematic diagram of the structure of a return-to-position determination training device provided in an embodiment of this application. The return-to-position determination training device 800 may include one or more central processing units (CPUs) 801 and a memory 805, in which one or more application programs or data are stored.

[0072] The memory 805 can be volatile or persistent storage. The program stored in the memory 805 can include one or more modules, each module including a series of instruction operations on the return-to-position determination training device 800. Furthermore, the central processing unit 801 can be configured to communicate with the memory 805 and execute the series of instruction operations in the memory 805 on the return-to-position determination training device 800.

[0073] The return-to-position determination training device 800 may also include one or more power supplies 802, one or more wired or wireless network interfaces 803, one or more input / output interfaces 804, and / or one or more operating systems, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0074] The central processing unit 801 can perform the aforementioned... Figures 1 to 6 The specific operations performed by the return-to-position determination training device in the illustrated embodiment will not be described in detail here.

[0075] This application also provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the methods described in the foregoing embodiments.

[0076] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0077] It should be noted that although the steps in the flowcharts of the various embodiments are drawn sequentially according to the arrows, unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the various embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps.

[0078] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0079] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0080] The above are merely preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structural transformations made using the description and drawings of the present application under the inventive concept of the present application, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present application.

Claims

1. A training method for determining the return position of a ball-serving machine, characterized in that, The control module applied to the ball-serving machine, which also includes a telephoto camera and a short-focus camera, includes the following methods: The telephoto camera is controlled to acquire multiple frames of telephoto images of the target stadium within a preset time window, and the short-focus camera is controlled to acquire multiple frames of short-focus images of the target stadium within the time window. The telephoto images are then processed to obtain the target image. The probability of a trainee for each candidate in the target image is determined based on the target image, and the target area corresponding to the target court is determined based on the short-focus image. A trainer is determined from the plurality of candidates based on the trainer probabilities of all the candidates and the target region; The ratio of the number of frames in which the trainee's court coordinates are within a preset return zone within the time window to the time window itself is determined as the return zone time percentage. The trainee's court coordinates are obtained based on the target image and the short-focus image. If the proportion of the return-to-position time is greater than the first preset threshold, then the trainee is determined to have stably returned to position.

2. The method for determining return position according to claim 1, characterized in that, Before determining that the trainee has stably returned to their original position if the proportion of the return time is greater than a first preset threshold, the method further includes: The ratio of the vector sum of multiple court coordinates of the trainee within the time window to the time window is determined as the position mean, and the ratio of the sum of the squared Euclidean distances of each court coordinate of the trainee and the position mean to the time window is determined as the position variance; the average of multiple movement speeds of the trainee is determined as the speed mean, wherein the movement speed is obtained within the time window based on multiple court coordinates corresponding to multiple frames of images of the trainee in the court coordinate system; The step of determining that the trainee has stably returned to position if the proportion of the return-to-position time is greater than a first preset threshold includes: If the preset conditions are met, the trainee is determined to have returned to a stable position. The preset conditions include at least one of the following: the return zone time percentage is greater than a first preset threshold, the position variance is less than a second preset threshold, the average speed is less than a third preset threshold, and the trainee's orientation is aligned with the ball-serving machine. The preset conditions include at least the return zone time percentage being greater than the first preset threshold, and the trainee's orientation is calculated based on the trainee's key body points.

3. The method for determining return position according to claim 1, characterized in that, Determining the target area corresponding to the target court based on the short-focus image includes: Based on a preset key point detection algorithm, key point detection processing is performed on the short-focus image to obtain multiple corresponding short-focus key points. The corresponding first plane homography matrix is ​​determined according to the long-focus intrinsic parameters of the long-focus camera, the short-focus intrinsic parameters of the short-focus camera, the preset short-focus to long-focus rotation matrix, the preset short-focus to long-focus translation matrix, the transpose of the normal vector in the preset court plane formula, and the constants in the court plane formula. For each of the short focal length key points, the product of the first planar homography matrix and the short focal length coordinates of the short focal length key point is determined as the long focal length coordinates, and the area enclosed by the long focal length coordinates of all the short focal length key points is determined as the target area.

4. The method for determining return position according to claim 3, characterized in that, The step of determining the trainer from multiple candidates based on the trainer probabilities of all candidates and the target region includes: The candidate stadium coordinates in the corresponding stadium coordinate system are calculated based on the candidate image coordinates of multiple candidates in the target image and the second plane homography matrix. The second plane homography matrix is ​​the homography matrix of the plane from the target area to the target stadium. For each candidate, the ratio of the number of frames in which the candidate's court coordinates are within the target area to the time window is determined as the in-court percentage. The corresponding score is obtained by weighting the in-court percentage, the trainee probability, the preset in-court switching coefficient, and three preset weight coefficients. The candidate corresponding to the highest score among all the candidates is identified as the trainer.

5. The method for determining return position according to claim 1, characterized in that, The process of processing the telephoto image to obtain the target image includes: Calculate the coordinates of the candidate image in the telephoto image, and determine the region of interest image containing the area where the candidate image coordinates are located; The region of interest image is extracted and processed to obtain the corresponding original RAW image and the red, green and blue three-channel RGB image; Based on a preset feature extraction model, the RAW image is input into the RAW feature extraction module of the feature extraction model to obtain the RAW features output by the RAW feature extraction module, and the RGB image is input into the RGB feature extraction module of the feature extraction model to obtain the RGB features output by the RGB feature extraction module. The RAW and RGB features are concatenated and fused to obtain fused features. Based on a preset enhancement algorithm, corresponding enhanced features are generated according to the fused features. The enhanced features are then subjected to convolution and pooling to obtain corresponding Image Signal Processor (ISP) parameters. The ISP parameters include exposure compensation, gamma correction, contrast, saturation, sharpening, noise reduction, contrast-limited adaptive histogram equalization cropping threshold, and image enhancement weights. The region of interest (ROI) image is processed according to the ISP parameters to obtain an enhanced image. Based on a weighted sum algorithm, the corresponding target image is calculated according to the enhanced image, the image enhancement weights, and the ROI image.

6. The method for determining return position according to claim 1, characterized in that, If the ball-serving machine further includes a ball-serving module, then after determining that the trainee has stably returned to their position, the method further includes: The target area is divided into a return zone and multiple receiving zones according to the division instructions input by the user; Based on the mode selection command input by the user, the mode corresponding to the mode selection command is determined as the target mode, and the corresponding target receiving area is determined from the plurality of receiving areas based on the target mode; Based on the preset relationship between trainee performance and difficulty adjustment parameters, the corresponding target difficulty adjustment parameters are determined according to the trainee's target historical consecutive successful reception count and the average time taken to return to position. The trainee performance includes the historical successful reception count and the average time taken to return to position. The current serving parameters are adjusted according to the target difficulty adjustment parameters to obtain the target serving parameters, which include serving speed, serving frequency, left and right landing point span, front and back landing point span, and spin intensity. The trainee's dominant hand is determined based on the user's input of the dominant hand selection command, real-time racket hand recognition data, or historical hitting data; the same side or opposite side of the dominant hand is determined as the serving side; The serving module is controlled to serve based on the target receiving area, the target serving parameters, and the serving side.

7. The method for determining return position according to claim 6, characterized in that, If the number of receiving areas is two, and the target mode is an alternating training mode, then the corresponding target receiving area is determined from the plurality of receiving areas according to the target mode, including: Obtain the most recent receiving area from historical ball-hitting data, and determine the other receiving area among the multiple receiving areas besides the most recent receiving area as the target receiving area; or, If the target pattern is a random pattern, then the corresponding target receiving area is determined from the plurality of receiving areas according to the target pattern, including: The target receiving area is determined as the current receiving area if the number of times it has been continuously identified as a historical target receiving area has not exceeded a preset threshold. or, If the target pattern is a weak-side strengthening pattern, then the corresponding target receiving area is determined from the plurality of receiving areas according to the target pattern, including: For each receiving area, based on a preset weighted sum algorithm, the corresponding target performance score is calculated according to the time from the return area to the receiving area, the time from the receiving area to the return area, and the shot success rate obtained from historical shot data. Based on the preset relationship between performance score and serve probability, the target serve probability of each receiving area is determined according to the target performance score of all receiving areas, and any receiving area is determined as the target receiving area based on the target serve probability, wherein the performance score is negatively correlated with the serve probability.

8. The method for determining return position according to claim 1, characterized in that, The step of determining the trainee probability of each candidate in the target image based on the target image includes: Based on a preset recognition model, the target image is input into the feature extraction module of the recognition model to obtain multi-scale visual features output by the feature extraction module. The multi-scale visual features are then input into the fusion module of the recognition model for fusion processing to obtain the trainee probability of each candidate output by the fusion module.

9. A return-to-position determination training device, characterized in that, include: The acquisition unit is used to acquire multiple long-focus images of the target stadium within a preset time window, acquire multiple short-focus images of the target stadium within the time window, and process the long-focus images to obtain the target image. The determining unit is configured to determine the trainee probability of each candidate in the target image based on the target image, and to determine the target area corresponding to the target court based on the short-focus image; The determining unit is further configured to determine a trainer from among the multiple candidates based on the trainer probabilities of all the candidates and the target region; The determining unit is further configured to determine the ratio of the number of frames in which the trainee's court coordinates are within a preset return zone within the time window to the time window as the return zone time ratio, wherein the trainee's court coordinates are obtained based on the target image and the short-focus image; The determining unit is further configured to determine that the trainee has stably returned to position if the proportion of the return time is greater than a first preset threshold.

10. A computer-readable storage medium, characterized in that, Includes instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 8.

Images