A badminton serve control method and system based on artificial intelligence

Abstract

The present invention relates to the technical field of computer vision data processing, and specifically discloses a badminton serve control method and system based on artificial intelligence. The method and system realize the analysis of the hitting of the players in the standard badminton match video, and form a standard serve strategy set and a standard receiving feature set; the automatic serve machine uses the standard serve strategy set to realize the serve operation, and at the same time, collects the real-time receiving video of the trainee, obtains the real-time receiving feature set of the trainee and the overall receiving feature set of the trainee; uses the real-time receiving feature set of the trainee to determine the serving strategy of the next ball and execute it; uses the standard receiving feature set and the overall receiving feature set of the trainee to determine the serving strategy of the next stage and execute it. The present invention can simulate the serving strategy of real athletes, and adjust the serving strategy of the next ball and the next stage according to the receiving reaction of the trainee, so as to meet the personalized needs of different trainees.

Description

Technical Field

[0001] The invention belongs to the technical field of computer vision data processing, and in particular relates to a badminton serve control method and system based on artificial intelligence. Background Art

[0002] Badminton is a common ball sport. When training badminton, it mainly relies on manual sparring. With the development of science and technology, there is a way to use automatic serving machines for serving training. The automatic serving machine is a complex device including a mechanical unit and an electronic unit, in which the electronic unit processes sensor signals, calculates various parameters when serving, and converts the parameters into control signals and outputs them to the mechanical unit to realize automatic serving operation. However, traditional automatic serving machines usually serve according to pre-set mechanical parameters, and the serving mode is fixed and single, which is not conducive to personalized training of athletes.

[0003] With the continuous development of artificial intelligence and big data technology, data processing technology based on computer vision is also constantly improving, and people have put forward higher requirements for the intelligence of badminton automatic serving machines. How to improve the electronic unit of the badminton automatic serving machine so that it can simulate the serving method of real athletes and adjust the serving strategy according to the current status and needs of the training personnel, so that the badminton automatic serving machine can be more intelligent, is the technical problem to be solved by the present invention. Therefore, the present invention aims to provide a badminton serving control method and system based on artificial intelligence. Summary of the invention

[0004] The purpose of the present invention is to overcome the above-mentioned problems existing in the prior art, and provide a badminton serve control method and system based on artificial intelligence, which realizes the analysis of the players' shots in the standard badminton match video, and forms a standard serve strategy set and a standard receiving feature set; the automatic serve machine uses the standard serve strategy set to realize the serve operation, and at the same time, collects the real-time receiving video of the trainee, obtains the real-time receiving feature set of the trainee and the overall receiving feature set of the trainee; uses the real-time receiving feature set of the trainee to determine the serving strategy of the next ball and execute it; uses the standard receiving feature set and the overall receiving feature set of the trainee to determine the serving strategy of the next stage and execute it. The present invention can simulate the serving strategy of real athletes, and adjust the serving strategy of the next ball and the next stage according to the receiving reaction of the trainee, so as to meet the personalized needs of different trainees.

[0005] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0006] A badminton serve control method based on artificial intelligence comprises the following steps:

[0007] A standard badminton match video set is collected, wherein the standard badminton match video set includes n standard badminton match videos of different difficulty levels, wherein the i-th standard badminton match video includes m players from both sides. i The standard badminton match video is used to analyze the sports postures, body movement information and racket movement information of both players, and form n standard serving strategy sets and standard receiving feature sets of different difficulty levels; the standard serving strategy set includes the number of serves m i , serving interval time, serving position, serving height, serving angle and serving strength; the standard ball receiving feature set includes posture action category feature vector, key point movement speed feature vector and racket movement feature vector; when the training task starts, the automatic serving machine realizes serving control according to the standard serving strategy set under a certain difficulty level, and the trainee performs receiving training as the receiving party; the real-time receiving video of the trainee is collected and analyzed to obtain the real-time receiving feature set of the trainee and the overall receiving feature set of the trainee; the real-time receiving feature set of the trainee corresponds to the real-time receiving reaction of the trainee in each serving strategy; the overall receiving feature set of the trainee corresponds to the overall receiving reaction of the trainee in this round of serving strategy; the serving strategy of the next ball is determined and executed by using the real-time receiving feature set of the trainee; the serving strategy of the next stage is determined and executed by using the standard receiving feature set and the overall receiving feature set of the trainee.

[0008] Furthermore, a set of n standard serving strategies of different difficulty levels is formed, specifically including:

[0009] The two athletes in the i-th standard badminton match video are marked as athlete A and athlete B respectively; athlete A in the standard badminton match video is identified and tracked, and the location m i The m corresponding to player A in the round i The batting image frames IA1~IA mi , respectively, with each hitting image frame IA mi Centered on the battle video, extract IA mi The back k i image frame, and the current hitting image frame IA mi Together, the total number is k i +1: the key frame image set IA_i of the player A’s i-th shot;

[0010] Calculate m i The batting image frames IA1~IA mi The timestamp difference between the adjacent frames before and after is obtained i -1 timestamp difference, used as the serve interval in the standard serve strategy set;

[0011] Construct a spatial data model corresponding to a badminton court, and establish a coordinate mapping model between a pixel coordinate system and a court plane coordinate system;

[0012] For m i The batting image frames IA1~IA mi Performing half-court court area identification and segmentation on each of the images to obtain a half-court court area image including the sneakers and badminton of athlete A; performing contour detection on the sneakers and badminton of athlete A in the half-court court area image to obtain a sneaker area and a badminton area, respectively calculating two minimum circumscribed rectangles corresponding to the sneaker area and the badminton area, respectively obtaining the sneaker centroid pixel coordinates and the badminton centroid pixel coordinates, obtaining the badminton height pixel coordinates through the relative positions between the sneaker centroid pixel coordinates and the badminton centroid pixel coordinates, obtaining the sneaker ground coordinates and the badminton height coordinates by using the coordinate mapping model, and using the sneaker ground coordinates and the badminton height coordinates as the serving position and serving height in the standard serving strategy set, respectively;

[0013] Badminton recognition and tracking are performed on each image in the key frame image set IA_i to obtain a badminton motion key point trajectory sequence ((x1, y1, t1), (x2, y2, t2), (x3, y3, t3), ..., (x j ,y j ,t j )), where (x j ,y j ,t j ) represents the abscissa, ordinate and timestamp corresponding to the j-th key point; using the badminton motion key point trajectory sequence to calculate the flight angle and initial speed of the badminton, and using the flight angle as the serving angle in the standard serving strategy set; using the flight angle and initial speed to obtain the hitting force of the badminton, and using the hitting force as the serving force in the standard serving strategy set;

[0014] According to the above method, the n standard badminton match videos of different difficulty levels are processed respectively to obtain n standard serving strategy sets of different difficulty levels.

[0015] Furthermore, n standard ball catching feature sets of different difficulty levels are formed, specifically including:

[0016] The two athletes in the i-th standard badminton match video are marked as athlete A and athlete B respectively; athlete B in the standard badminton match video is identified and tracked, and the location m i The m corresponding to player B in the round i The batting image frames IB1~IB mi , respectively, with each hitting image frame IBmi As the center, extract IB in the battle video mi The k before and after i image frames, and the current hitting image frame IB mi Together the total number is 2k i +1: the key frame image set IB_i of the i-th shot of player B;

[0017] Respectively identify and segment the athlete B for each image in the key frame image set IB_i to obtain a human body region image of the athlete B, perform key point detection on the human body region image to obtain a key point spatial coordinate set corresponding to a total of 14 key points including the nose, neck, left and right shoulders, left and right elbows, left and right wrists, left and right hips, left and right knees, and left and right ankles;

[0018] Using the key point spatial coordinate set to obtain a joint angle feature set and a joint distance feature set, splicing the joint angle feature set and the joint distance feature set to obtain a posture action feature vector, inputting the posture action feature vector into a pre-trained posture action recognition model for recognition, and obtaining a corresponding posture action category, wherein the posture action category includes jumping, sideways, and arm raising; combining the posture action category corresponding to each image in the key frame image set IB_i in timestamp order to obtain a posture action category feature vector in a standard ball catching feature set;

[0019] Using the key point spatial coordinate set of each adjacent image in the key frame image set IB_i, a moving speed sequence of each key point in the x and y directions is obtained as a key point moving speed feature vector in the standard ball receiving feature set;

[0020] Respectively perform racket recognition and segmentation on each image in the key frame image set IB_i to obtain a racket area image, process the racket area image, and calculate the minimum circumscribed rectangle area value and the minimum circumscribed rectangle centroid coordinate value corresponding to the racket area;

[0021] For m i The batting image frames IB1~IB mi For each image in the image, the racket and badminton body are identified, the coordinate value of the center of mass of the minimum circumscribed rectangle of the racket area and the coordinate value of the badminton collision point are calculated, and the relative coordinate value of the collision point is obtained;

[0022] The area value, the center of mass coordinate value and the relative coordinate value of the ball-collision point corresponding to each image in the key frame image set IB_i are combined in the order of timestamps to obtain a racket motion feature vector in a standard ball-catching feature set;

[0023] According to the above method, the n standard badminton game videos of different difficulty levels are processed respectively to obtain n standard ball receiving feature sets of different difficulty levels.

[0024] Furthermore, the real-time ball receiving feature set of the training personnel and the overall ball receiving feature set of the training personnel specifically include:

[0025] The trainer's real-time ball catching feature set includes the trainer's ball catching results and the trainer's posture and action category feature vectors corresponding to each ball; the trainer's posture and action category feature vectors are obtained by processing and analyzing the trainer's real-time ball catching video, and the specific calculation method is the same as the calculation method of the athlete's posture and action category feature vectors in the standard ball catching feature set; the trainer's overall ball catching feature set is obtained by processing and analyzing the trainer's real-time ball catching video, and the specific calculation method is the same as the calculation method of the standard ball catching feature set.

[0026] Furthermore, the real-time receiving feature set of the training personnel is used to determine the serving strategy for the next ball and execute it, specifically including: if the receiving result in the real-time receiving feature set is a failure, and the similarity between the training personnel's posture action category feature vector and the corresponding player's posture action category feature vector in the standard receiving feature set is less than a preset threshold T p , the serving strategy for the next ball repeats the current serving strategy, otherwise, the serving strategy for the next ball corresponding to the predetermined standard serving strategy is used; the use of the standard receiving feature set and the overall receiving feature set of the training personnel to determine the serving strategy for the next stage and execute it specifically includes: if the similarity between the overall receiving feature set of the training personnel and the standard receiving feature set is less than a preset threshold value T w , and the overall ball receiving error rate is greater than the preset threshold T n , then the serving strategy of the next stage repeats the serving strategy of this stage. Otherwise, it is considered that the receiving training corresponding to the serving strategy of this stage is qualified, and a more difficult training can be carried out in the next stage.

[0027] The present invention also provides the following technical solutions:

[0028] A badminton serve control system based on artificial intelligence, including the following functional modules:

[0029] Standard battle video processing and analysis module: collects a standard badminton battle video set, wherein the standard badminton battle video set includes n standard badminton battle videos of different difficulty levels, wherein the i-th standard badminton battle video includes m players from both sides. iThe standard badminton match video is used to analyze the sports postures, body movement information and racket movement information of both players, and form n standard serving strategy sets and standard receiving feature sets of different difficulty levels; the standard serving strategy set includes the number of serves m i , serving interval time, serving position, serving height, serving angle and serving strength; the standard receiving feature set includes posture action category feature vector, key point moving speed feature vector and racket movement feature vector;

[0030] Serving control analysis module: when the training task starts, the automatic serving machine implements serving control according to a standard serving strategy set under a certain difficulty level, and the trainee performs receiving training as the receiving party; the real-time receiving video of the trainee is collected and analyzed to obtain the real-time receiving feature set of the trainee and the overall receiving feature set of the trainee; the real-time receiving feature set of the trainee corresponds to the real-time receiving reaction of the trainee in each serving strategy; the overall receiving feature set of the trainee corresponds to the overall receiving reaction of the trainee in this round of serving strategy;

[0031] Serving strategy adjustment module: Use the real-time receiving feature set of the trainer to determine and execute the serving strategy for the next ball; use the standard receiving feature set and the overall receiving feature set of the trainer to determine and execute the serving strategy for the next stage.

[0032] Furthermore, the standard battle video processing and analysis module specifically includes:

[0033] Standard serve strategy set calculation unit: mark the two players in the i-th standard badminton match video as player A and player B respectively; identify and track player A in the standard badminton match video, and locate m i The m corresponding to player A in the round i The batting image frames IA1~IA mi , respectively, with each hitting image frame IA mi Centered on the battle video, extract IA mi The back k i image frame, and the current hitting image frame IA mi Together, the total number is k i +1: the key frame image set IA_i of the player A’s i-th shot;

[0034] Calculate m i The batting image frames IA1~IA mi The timestamp difference between the adjacent frames before and after is obtained i -1 timestamp difference, used as the serve interval in the standard serve strategy set;

[0035] Construct a spatial data model corresponding to a badminton court, and establish a coordinate mapping model between a pixel coordinate system and a court plane coordinate system;

[0036] For m i The batting image frames IA1~IA mi Performing half-court court area identification and segmentation on each of the images to obtain a half-court court area image including the sneakers and badminton of athlete A; performing contour detection on the sneakers and badminton of athlete A in the half-court court area image to obtain a sneaker area and a badminton area, respectively calculating two minimum circumscribed rectangles corresponding to the sneaker area and the badminton area, respectively obtaining the sneaker centroid pixel coordinates and the badminton centroid pixel coordinates, obtaining the badminton height pixel coordinates through the relative positions between the sneaker centroid pixel coordinates and the badminton centroid pixel coordinates, obtaining the sneaker ground coordinates and the badminton height coordinates by using the coordinate mapping model, and using the sneaker ground coordinates and the badminton height coordinates as the serving position and serving height in the standard serving strategy set, respectively;

[0037] Badminton recognition and tracking are performed on each image in the key frame image set IA_i to obtain a badminton motion key point trajectory sequence ((x1, y1, t1), (x2, y2, t2), (x3, y3, t3), ..., (x j ,y j ,t j )), where (x j ,y j ,t j ) represents the abscissa, ordinate and timestamp corresponding to the j-th key point; using the badminton motion key point trajectory sequence to calculate the flight angle and initial speed of the badminton, and using the flight angle as the serving angle in the standard serving strategy set; using the flight angle and initial speed to obtain the hitting force of the badminton, and using the hitting force as the serving force in the standard serving strategy set;

[0038] According to the above method, the n standard badminton match videos of different difficulty levels are processed respectively to obtain n standard serving strategy sets of different difficulty levels.

[0039] Furthermore, the standard battle video processing and analysis module specifically includes:

[0040] Standard catch feature set calculation unit: mark the two players in the i-th standard badminton match video as player A and player B respectively; identify and track player B in the standard badminton match video, and locate m i The m corresponding to player B in the round i The batting image frames IB1~IB mi , respectively, with each hitting image frame IBmi As the center, extract IB in the battle video mi The k before and after i image frames, and the current hitting image frame IB mi Together the total number is 2k i +1: the key frame image set IB_i of the i-th shot of player B;

[0041] Respectively identify and segment the athlete B for each image in the key frame image set IB_i to obtain a human body region image of the athlete B, perform key point detection on the human body region image to obtain a key point spatial coordinate set corresponding to a total of 14 key points including the nose, neck, left and right shoulders, left and right elbows, left and right wrists, left and right hips, left and right knees, and left and right ankles;

[0042] Using the key point spatial coordinate set to obtain a joint angle feature set and a joint distance feature set, splicing the joint angle feature set and the joint distance feature set to obtain a posture action feature vector, inputting the posture action feature vector into a pre-trained posture action recognition model for recognition, and obtaining a corresponding posture action category, wherein the posture action category includes jumping, sideways, and arm raising; combining the posture action category corresponding to each image in the key frame image set IB_i in timestamp order to obtain a posture action category feature vector in a standard ball catching feature set;

[0043] Using the key point spatial coordinate set of each adjacent image in the key frame image set IB_i, a moving speed sequence of each key point in the x and y directions is obtained as a key point moving speed feature vector in the standard ball receiving feature set;

[0044] Respectively perform racket recognition and segmentation on each image in the key frame image set IB_i to obtain a racket area image, process the racket area image, and calculate the minimum circumscribed rectangle area value and the minimum circumscribed rectangle centroid coordinate value corresponding to the racket area;

[0045] For m i The batting image frames IB1~IB mi For each image in the image, the racket and badminton body are identified, the coordinate value of the center of mass of the minimum circumscribed rectangle of the racket area and the coordinate value of the badminton collision point are calculated, and the relative coordinate value of the collision point is obtained;

[0046] The area value, the center of mass coordinate value and the relative coordinate value of the ball-collision point corresponding to each image in the key frame image set IB_i are combined in the order of timestamps to obtain a racket motion feature vector in a standard ball-catching feature set;

[0047] According to the above method, the n standard badminton game videos of different difficulty levels are processed respectively to obtain n standard ball receiving feature sets of different difficulty levels.

[0048] Furthermore, the serve control analysis module specifically includes:

[0049] Real-time ball receiving feature set calculation unit: the real-time ball receiving feature set of the trainer includes the ball receiving result of the trainer and the trainer's posture action category feature vector corresponding to each ball; the trainer's posture action category feature vector is obtained by processing and analyzing the real-time ball receiving video of the trainer, and the specific calculation method is the same as the calculation method of the athlete's posture action category feature vector in the standard ball receiving feature set;

[0050] Overall ball receiving feature set calculation unit: The overall ball receiving feature set of the trainee is obtained by processing and analyzing the real-time ball receiving video of the trainee, and the specific calculation method is the same as the standard ball receiving feature set calculation method.

[0051] Furthermore, the serving strategy adjustment module specifically includes:

[0052] If the result of the catch in the real-time catch feature set is a failure, and the similarity between the training personnel's posture action category feature vector and the corresponding player's posture action category feature vector in the standard catch feature set is less than the preset threshold T p , the serving strategy for the next ball repeats the current serving strategy, otherwise, the serving strategy for the next ball corresponding to the predetermined standard serving strategy is used; the use of the standard receiving feature set and the overall receiving feature set of the training personnel to determine the serving strategy for the next stage and execute it specifically includes: if the similarity between the overall receiving feature set of the training personnel and the standard receiving feature set is less than a preset threshold value T w , and the overall ball receiving error rate is greater than the preset threshold T n , then the serving strategy of the next stage repeats the serving strategy of this stage. Otherwise, it is considered that the receiving training corresponding to the serving strategy of this stage is qualified, and a more difficult training can be carried out in the next stage.

[0053] By adopting the above technical solution, compared with the prior art, the beneficial effects of the present invention are:

[0054] The present invention processes and analyzes a standard badminton match video set to obtain a standard serving strategy set as a preset strategy set for a badminton automatic serving machine. The preset strategy set is derived from a match video of real players, so that the serving method is more consistent with the real-person match method.

[0055] The present invention processes and analyzes a standard badminton match video set to obtain a standard ball receiving feature set, uses the standard ball receiving feature set to compare and analyze the ball receiving performance of trainees, and adjusts the next ball and the next stage serving strategy of a badminton automatic serving machine according to the analysis result.

[0056] The present invention adopts artificial intelligence technology to process and analyze a standard badminton match video set, and obtains the serving position, serving height, serving angle and serving force in a standard serving strategy set based on an efficient image processing method.

[0057] The present invention adopts artificial intelligence technology to process and analyze standard badminton match video sets and real-time ball catching videos of trainees, and obtains feature vectors corresponding to movement posture, body movement information and racket movement information through a variety of feature extraction and characterization methods, thereby improving the comprehensiveness of feature vectors. At the same time, it simplifies the expression methods of standard ball catching feature sets, real-time ball catching feature sets of trainees and overall ball catching feature sets of trainees, thereby improving the calculation efficiency of feature vectors. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0059] Figure 1 A flowchart of a badminton serve control method based on artificial intelligence of the present invention;

[0060] Figure 2 A flow chart of calculating a standard serving strategy set for the present invention;

[0061] Figure 3 A flow chart for calculating a standard catch feature set for the present invention;

[0062] Figure 4 A flow chart of calculating a real-time ball receiving feature set of a training staff and an overall ball receiving feature set of a training staff according to the present invention;

[0063] Figure 5 A flow chart for determining the next ball and the next stage serving strategy for the present invention;

[0064] Figure 6 The present invention is a functional module diagram of a badminton serve control system based on artificial intelligence. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] Example 1

[0067] like Figure 1 As shown, this embodiment provides a badminton serve control method based on artificial intelligence, comprising the following steps:

[0068] A standard badminton match video set is collected, wherein the standard badminton match video set includes n standard badminton match videos of different difficulty levels, wherein the i-th standard badminton match video includes m players from both sides. i The standard badminton match video is used to analyze the sports postures, body movement information and racket movement information of both players, and form n standard serving strategy sets and standard receiving feature sets of different difficulty levels; the standard serving strategy set includes the number of serves m i , serving interval time, serving position, serving height, serving angle and serving strength; the standard ball receiving feature set includes posture action category feature vector, key point movement speed feature vector and racket movement feature vector; when the training task starts, the automatic serving machine realizes serving control according to the standard serving strategy set under a certain difficulty level, and the trainee performs receiving training as the receiving party; the real-time receiving video of the trainee is collected and analyzed to obtain the real-time receiving feature set of the trainee and the overall receiving feature set of the trainee; the real-time receiving feature set of the trainee corresponds to the real-time receiving reaction of the trainee in each serving strategy; the overall receiving feature set of the trainee corresponds to the overall receiving reaction of the trainee in this round of serving strategy; the serving strategy of the next ball is determined and executed by using the real-time receiving feature set of the trainee; the serving strategy of the next stage is determined and executed by using the standard receiving feature set and the overall receiving feature set of the trainee.

[0069] The standard badminton match video collection is a video collection obtained by recording and processing the matches between real athletes. Each standard badminton match video corresponds to the match between athletes of different levels. In order to facilitate the analysis of the video and form a coherent serving strategy, the match video records several rounds in which the athletes of both sides make no mistakes.

[0070] Example 2

[0071] like Figure 2 As shown, this embodiment provides a badminton serve control method based on artificial intelligence, wherein a set of standard serve strategies of n different difficulty levels is formed, specifically including:

[0072] The two athletes in the i-th standard badminton match video are marked as athlete A and athlete B respectively; athlete A in the standard badminton match video is identified and tracked, and the location m i The m corresponding to player A in the round i The batting image frames IA1~IA mi , respectively, with each hitting image frame IA mi Centered on the battle video, extract IA mi The back k i image frame, and the current hitting image frame IA mi Together, the total number is k i +1 is the key frame image set IA_i of the i-th hitting of athlete A.

[0073] Calculate m i The batting image frames IA1~IA mi The timestamp difference between the adjacent frames before and after is obtained i -1 timestamp difference, used as the serve interval in the standard serve strategy set;

[0074] Construct a spatial data model corresponding to a badminton court, and establish a coordinate mapping model between a pixel coordinate system and a court plane coordinate system;

[0075] For m i The batting image frames IA1~IA mi A half-court area is identified and segmented for each of the images to obtain a half-court area image including the sneakers and badminton of athlete A; contour detection is performed on the sneakers and badminton of athlete A in the half-court area image to obtain a sneaker area and a badminton area, two minimum circumscribed rectangles corresponding to the sneaker area and the badminton area are calculated respectively to obtain the sneaker centroid pixel coordinates and the badminton centroid pixel coordinates, the height pixel coordinates of the badminton are obtained through the relative positions between the sneaker centroid pixel coordinates and the badminton centroid pixel coordinates, the sneaker ground coordinates and the badminton height coordinates are obtained by using the coordinate mapping model, and the sneaker ground coordinates and the badminton height coordinates are used as the serving position and serving height in the standard serving strategy set respectively.

[0076] The hitting situation of player A is analyzed to form a serving strategy set. In order to simplify the calculation amount, the hitting image frame of player A is first located, and the hitting position and hitting height of player A can be analyzed through the single frame image at the hitting moment.

[0077] Badminton recognition and tracking are performed on each image in the key frame image set IA_i to obtain a badminton motion key point trajectory sequence ((x1, y1, t1), (x2, y2, t2), (x3, y3, t3), ..., (x j ,y j ,t j )), where (x j ,y j ,t j ) represents the abscissa, ordinate and timestamp corresponding to the j-th key point; the flight angle and initial speed of the badminton are calculated using the badminton motion key point trajectory sequence, and the flight angle is used as the serving angle in the standard serving strategy set; the hitting force of the badminton is obtained using the flight angle and the initial speed, and the hitting force is used as the serving force in the standard serving strategy set.

[0078] Taking the batting image frame as the starting point, extracting a certain number of key frames thereafter, the dynamic characteristics of athlete A's batting can be analyzed, including the flight angle and initial flight speed of the ball. First, the motion equations of the badminton in the x direction and y direction are calculated, respectively. The batting strength of the badminton is closely related to the swing speed, batting conversion rate, and the flight angle and speed of the badminton. When specifically calculating, factors such as the mass and air resistance of the badminton need to be considered. By measuring the flight speed and angle of the badminton, combined with the motion equation, the strength when batting can be estimated. The specific motion equation refers to the physics equation in the prior art, which will not be repeated here.

[0079] According to the above method, the n standard badminton match videos of different difficulty levels are processed respectively to obtain n standard serving strategy sets of different difficulty levels.

[0080] Example 3

[0081] like Figure 3 As shown, this embodiment provides a badminton serve control method based on artificial intelligence, wherein n standard ball receiving feature sets of different difficulty levels are formed, specifically including:

[0082] The two athletes in the i-th standard badminton match video are marked as athlete A and athlete B respectively; athlete B in the standard badminton match video is identified and tracked, and the location m i The m corresponding to player B in the round i The batting image frames IB1~IB mi , respectively, with each hitting image frame IB mi As the center, extract IB in the battle video mi The k before and after i image frames, and the current hitting image frame IBmi Together the total number is 2k i +1 is the key frame image set IB_i of the i-th batting of athlete B.

[0083] The hitting situation of player B is analyzed to form a standard catch feature set. First, the hitting image frame of player B is located. The single frame image at the hitting moment is taken as the center, and a certain number of image frames before and after it are located. The static and dynamic features of player B's hitting can be analyzed through these image frames, thus forming a standard catch feature set.

[0084] Respectively identify and segment the athlete B for each image in the key frame image set IB_i to obtain a human body region image of the athlete B, perform key point detection on the human body region image to obtain a key point spatial coordinate set corresponding to a total of 14 key points including the nose, neck, left and right shoulders, left and right elbows, left and right wrists, left and right hips, left and right knees, and left and right ankles;

[0085] The key point spatial coordinate set is used to obtain a joint angle feature set and a joint distance feature set, the joint angle feature set and the joint distance feature set are spliced ​​to obtain a posture action feature vector, the posture action feature vector is input into a pre-trained posture action recognition model for recognition, and the corresponding posture action category is obtained, and the posture action category includes jumping, sideways and raising arms; the posture action category corresponding to each image in the key frame image set IB_i is combined in timestamp order to obtain the posture action category feature vector in the standard ball catching feature set.

[0086] According to the human skeleton structure and the spatial coordinate set of the key points of the human body, the angle of extension of each joint of the human body and the distance between the joint points can be calculated, which characterizes the spatial geometric characteristics of the human body posture, that is, the static characteristics, and the posture categories of the human body can be well identified by the static characteristics. In badminton, the more critical posture categories when hitting the ball include jumping, sideways and raising arms. The present embodiment pre-trains the posture action recognition model for the above-mentioned category actions, and the posture recognition model is based on artificial convolutional neural networks, such as mature models such as CNN. After identifying the posture category corresponding to each image, it is combined into an action sequence according to the time sequence, and characterized in the form of a posture action category feature vector, which is convenient for subsequent similarity calculation.

[0087] Using the key point spatial coordinate set of each adjacent image in the key frame image set IB_i, a moving speed sequence of each key point in the x and y directions is obtained as a key point moving speed feature vector in the standard ball receiving feature set;

[0088] The moving speed characteristics of the key points of the human body in the video reflect the changes of the human body at continuous moments and are dynamic features. They are calculated by dividing the moving distance of the key points by the moving time. The moving speed corresponding to each key point is summarized and normalized and represented in the form of a feature vector.

[0089] Respectively perform racket recognition and segmentation on each image in the key frame image set IB_i to obtain a racket area image, process the racket area image, and calculate the minimum circumscribed rectangle area value and the minimum circumscribed rectangle centroid coordinate value corresponding to the racket area;

[0090] For m i The batting image frames IB1~IB mi For each image in the image, the racket and badminton body are identified, the coordinate value of the center of mass of the minimum circumscribed rectangle of the racket area and the coordinate value of the badminton collision point are calculated, and the relative coordinate value of the collision point is obtained;

[0091] The area value, the center of mass coordinate value and the relative coordinate value of the ball-collision point corresponding to each image in the key frame image set IB_i are combined in the order of timestamps to obtain the racket motion feature vector in the standard ball-catching feature set.

[0092] The use of the racket by the athlete when hitting the ball can reflect the athlete's hitting skills. The movement of the racket, the point where the racket contacts the ball, the angle of the swing, and other characteristics all reflect the way of hitting the ball to a certain extent. The tracking of the center of mass of the racket can obtain the characteristics of the racket's motion trajectory. The minimum circumscribed rectangle area value corresponding to the racket area can reflect the angle of the racket face. The coordinate value of the center of mass of the minimum circumscribed rectangle of the racket area and the coordinate value of the badminton collision point can reflect which areas of the racket the hitting point is located.

[0093] According to the above method, the n standard badminton game videos of different difficulty levels are processed respectively to obtain n standard ball receiving feature sets of different difficulty levels.

[0094] Example 4

[0095] like Figure 4 As shown, this embodiment provides a badminton serve control method based on artificial intelligence, wherein the real-time ball receiving feature set of the trainer and the overall ball receiving feature set of the trainer specifically include:

[0096] The trainer's real-time ball catching feature set includes the trainer's ball catching results and the trainer's posture and action category feature vectors corresponding to each ball; the trainer's posture and action category feature vectors are obtained by processing and analyzing the trainer's real-time ball catching video, and the specific calculation method is the same as the calculation method of the athlete's posture and action category feature vectors in the standard ball catching feature set; the trainer's overall ball catching feature set is obtained by processing and analyzing the trainer's real-time ball catching video, and the specific calculation method is the same as the calculation method of the standard ball catching feature set.

[0097] Example 5

[0098] like Figure 5 As shown, this embodiment provides a badminton serve control method based on artificial intelligence, wherein the serve strategy for the next ball is determined and executed using the real-time ball receiving feature set of the trainer, specifically comprising: if the receiving result in the real-time ball receiving feature set is a failure, and the similarity between the trainer's posture action category feature vector and the athlete's posture action category feature vector of the corresponding ball in the standard ball receiving feature set is less than a preset threshold value T p , the serving strategy for the next ball repeats the current serving strategy, otherwise, the serving strategy for the next ball corresponding to the predetermined standard serving strategy is used; the use of the standard receiving feature set and the overall receiving feature set of the training personnel to determine the serving strategy for the next stage and execute it specifically includes: if the similarity between the overall receiving feature set of the training personnel and the standard receiving feature set is less than a preset threshold value T w , and the overall ball receiving error rate is greater than the preset threshold T n , then the serving strategy of the next stage repeats the serving strategy of this stage. Otherwise, it is considered that the receiving training corresponding to the serving strategy of this stage is qualified, and a more difficult training can be carried out in the next stage.

[0099] Example 6

[0100] like Figure 6 As shown, this embodiment provides a badminton serve control system based on artificial intelligence, including the following functional modules:

[0101] The standard match video processing and analysis module includes a standard serve strategy set calculation unit and a standard receiving feature set calculation unit: collect a standard badminton match video set, wherein the standard badminton match video set includes n segments of standard badminton match videos of different difficulty levels, wherein the i-th segment of the standard badminton match video includes m players from both sides. i The standard badminton match video is used to analyze the sports postures, body movement information and racket movement information of both players, and form n standard serving strategy sets and standard receiving feature sets of different difficulty levels; the standard serving strategy set includes the number of serves m i, serving interval time, serving position, serving height, serving angle and serving strength; the standard receiving feature set includes posture action category feature vector, key point moving speed feature vector and racket movement feature vector;

[0102] The serving control analysis module includes a real-time receiving feature set calculation unit and an overall receiving feature set calculation unit: when the training task starts, the automatic serving machine realizes serving control according to a standard serving strategy set under a certain difficulty level, and the trainee performs receiving training as the receiving party; the real-time receiving video of the trainee is collected and analyzed to obtain the real-time receiving feature set of the trainee and the overall receiving feature set of the trainee; the real-time receiving feature set of the trainee corresponds to the real-time receiving reaction of the trainee in each serving strategy; the overall receiving feature set of the trainee corresponds to the overall receiving reaction of the trainee in this round of serving strategy;

[0103] Serving strategy adjustment module: Use the real-time receiving feature set of the trainer to determine and execute the serving strategy for the next ball; use the standard receiving feature set and the overall receiving feature set of the trainer to determine and execute the serving strategy for the next stage.

[0104] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, computer devices, or computer-readable storage media. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0105] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0106] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. According to the technical essence of the present invention, within the spirit and principles of the present invention, any simple modification, equivalent replacement and improvement made to the above embodiment still falls within the protection scope of the technical solution of the present invention.

Claims

1. A badminton serve control method based on artificial intelligence, characterized in that: The steps include: A standard badminton match video set is collected, wherein the standard badminton match video set includes n standard badminton match videos of different difficulty levels, wherein the i-th standard badminton match video includes m players from both sides. i The standard badminton match video is used to analyze the sports postures, body movement information and racket movement information of both players, and form n standard serving strategy sets with different difficulty levels and n standard receiving feature sets with different difficulty levels; the standard serving strategy set includes the number of serves m i , serve interval time, serve position, serve height, serve angle and serve strength; the standard ball receiving feature set includes posture action category feature vector, key point movement speed feature vector and racket movement feature vector; the key point movement speed feature vector refers to the movement speed feature vector of a specific body part of the human body; when the training task starts, the automatic serving machine realizes serve control according to the standard serve strategy set under a certain difficulty level, and the trainee performs ball receiving training as the receiver; the real-time ball receiving video of the trainee is collected and analyzed to obtain the real-time ball receiving feature set of the trainee and the overall ball receiving feature set of the trainee; the real-time ball receiving video of the trainee is received and analyzed, and the real-time ball receiving feature set of the trainee and the overall ball receiving feature set of the trainee are obtained; the real-time ball receiving feature set of the trainee is received and analyzed, and the real-time ball receiving feature set of the trainee is obtained; ... The ball feature set corresponds to the trainee's real-time ball receiving reaction in each serving strategy; the trainee's overall ball receiving feature set corresponds to the trainee's overall ball receiving reaction in this round of serving strategy; the trainee's real-time ball receiving feature set includes the trainee's receiving results and the trainee's posture and action category feature vectors corresponding to each ball; the specific calculation method of the trainee's overall ball receiving feature set is the same as the calculation method of the standard ball receiving feature set; the trainee's real-time ball receiving feature set is used to determine and execute the serving strategy for the next ball; the standard ball receiving feature set and the trainee's overall ball receiving feature set are used to determine and execute the serving strategy for the next stage.

2. A badminton serve control method based on artificial intelligence as claimed in claim 1, characterized in that: The formation of n standard serving strategy sets of different difficulty levels specifically includes: The two athletes in the i-th standard badminton match video are marked as athlete A and athlete B respectively; athlete A in the standard badminton match video is identified and tracked, and the location m i The m corresponding to athlete A in the round trip i The batting image frames IA1~IA mi , respectively, with each hitting image frame IA mi Centered on the battle video, extract IA mi The back k i image frame, and the current hitting image frame IA mi Together, the total number is k i +1: the key frame image set IA_i of the player A’s i-th shot; Calculate m i The batting image frames IA1~IA mi The timestamp difference between the adjacent frames before and after is obtained i -1 timestamp difference, used as the serve interval in the standard serve strategy set; Construct a spatial data model corresponding to a badminton court, and establish a coordinate mapping model between a pixel coordinate system and a court plane coordinate system; For m i The batting image frames IA1~IA mi Performing half-court court area identification and segmentation on each of the images to obtain a half-court court area image including the sneakers and badminton of athlete A; performing contour detection on the sneakers and badminton of athlete A in the half-court court area image to obtain a sneaker area and a badminton area, respectively calculating two minimum circumscribed rectangles corresponding to the sneaker area and the badminton area, respectively obtaining the sneaker centroid pixel coordinates and the badminton centroid pixel coordinates, obtaining the badminton height pixel coordinates through the relative positions between the sneaker centroid pixel coordinates and the badminton centroid pixel coordinates, obtaining the sneaker ground coordinates and the badminton height coordinates by using the coordinate mapping model, and using the sneaker ground coordinates and the badminton height coordinates as the serving position and serving height in the standard serving strategy set, respectively; Badminton recognition and tracking are performed on each image in the key frame image set IA_i to obtain a badminton motion key point trajectory sequence ((x1, y1, t1), (x2, y2, t2), (x3, y3, t3), ..., (x j ,y j ,t j )), where (x j ,y j ,t j ) represents the abscissa, ordinate and timestamp corresponding to the j-th key point; using the badminton motion key point trajectory sequence to calculate the flight angle and initial speed of the badminton, and using the flight angle as the serving angle in the standard serving strategy set; using the flight angle and initial speed to obtain the hitting force of the badminton, and using the hitting force as the serving force in the standard serving strategy set; According to the above method, the n standard badminton match videos of different difficulty levels are processed respectively to obtain n standard serving strategy sets of different difficulty levels.

3. A badminton serve control method based on artificial intelligence as claimed in claim 1, characterized in that: The forming of n standard ball catching feature sets of different difficulty levels specifically includes: The two athletes in the i-th standard badminton match video are marked as athlete A and athlete B respectively; the athlete B in the standard badminton match video is identified and tracked, and the location m i The m corresponding to athlete B in the round trip i The batting image frames IB1~IB mi , respectively, with each hitting image frame IB mi As the center, extract IB in the battle video mi The k before and after i image frames, and the current hitting image frame IB mi Together the total number is 2k i +1: the key frame image set IB_i of the i-th shot of player B; Respectively identify and segment the athlete B for each image in the key frame image set IB_i to obtain a human body region image of the athlete B, perform key point detection on the human body region image to obtain a key point spatial coordinate set corresponding to a total of 14 key points including the nose, neck, left and right shoulders, left and right elbows, left and right wrists, left and right hips, left and right knees, and left and right ankles; Using the key point spatial coordinate set to obtain a joint angle feature set and a joint distance feature set, splicing the joint angle feature set and the joint distance feature set to obtain a posture action feature vector, inputting the posture action feature vector into a pre-trained posture action recognition model for recognition, and obtaining a corresponding posture action category, wherein the posture action category includes jumping, sideways, and arm raising; combining the posture action category corresponding to each image in the key frame image set IB_i in timestamp order to obtain a posture action category feature vector in a standard ball catching feature set; Using the key point spatial coordinate set of each adjacent image in the key frame image set IB_i, a moving speed sequence of each key point in the x and y directions is obtained as a key point moving speed feature vector in the standard ball receiving feature set; Respectively perform racket recognition and segmentation on each image in the key frame image set IB_i to obtain a racket area image, process the racket area image, and calculate the minimum circumscribed rectangle area value and the minimum circumscribed rectangle centroid coordinate value corresponding to the racket area; For m i The batting image frames IB1~IB mi For each image in the image, the racket and badminton body are identified, the coordinate value of the center of mass of the minimum circumscribed rectangle of the racket area and the coordinate value of the badminton collision point are calculated, and the relative coordinate value of the collision point is obtained; The area value, the center of mass coordinate value and the relative coordinate value of the ball-collision point corresponding to each image in the key frame image set IB_i are combined in the order of timestamps to obtain a racket motion feature vector in a standard ball-catching feature set; According to the above method, the n standard badminton game videos of different difficulty levels are processed respectively to obtain n standard ball receiving feature sets of different difficulty levels.

4. A badminton serve control method based on artificial intelligence as claimed in claim 2 or 3, characterized in that: The real-time ball receiving feature set of the training personnel and the overall ball receiving feature set of the training personnel specifically include: The trainer's posture and action category feature vector is obtained by processing and analyzing the trainer's real-time ball catching video, and the specific calculation method is the same as the calculation method of the athlete's posture and action category feature vector in the standard ball catching feature set; the trainer's overall ball catching feature set is obtained by processing and analyzing the trainer's real-time ball catching video.

5. The badminton serve control method based on artificial intelligence as claimed in claim 1, characterized in that: The method of using the real-time receiving feature set of the training personnel to determine the serving strategy for the next ball and execute it specifically includes: if the receiving result in the real-time receiving feature set is a failure, and the similarity between the training personnel's posture action category feature vector and the corresponding player's posture action category feature vector in the standard receiving feature set is less than a preset threshold value T p , the serving strategy for the next ball repeats the current serving strategy, otherwise, the serving strategy for the next ball corresponding to the predetermined standard serving strategy is used; the use of the standard receiving feature set and the overall receiving feature set of the training personnel to determine the serving strategy for the next stage and execute it specifically includes: if the similarity between the overall receiving feature set of the training personnel and the standard receiving feature set is less than a preset threshold value T w , and the overall ball receiving error rate is greater than the preset threshold T n , then the serving strategy of the next stage repeats the serving strategy of this stage. Otherwise, it is considered that the receiving training corresponding to the serving strategy of this stage is qualified, and a more difficult training can be carried out in the next stage.

6. A badminton serve control system based on artificial intelligence, characterized in that: Includes the following functional modules: Standard battle video processing and analysis module: collects a standard badminton battle video set, wherein the standard badminton battle video set includes n standard badminton battle videos of different difficulty levels, wherein the i-th standard badminton battle video includes m players from both sides. i The standard badminton match video is used to analyze the sports postures, body movement information and racket movement information of both players to form n standard serving strategy sets and standard receiving feature sets of different difficulty levels; the standard serving strategy set includes the number of serves m i , serving interval time, serving position, serving height, serving angle and serving force; the standard receiving feature set includes posture action category feature vector, key point moving speed feature vector and racket movement feature vector; the key point moving speed feature vector refers to the moving speed feature vector of a specific body part of the human body; Serving control analysis module: when the training task starts, the automatic serving machine implements serving control according to a standard serving strategy set at a certain difficulty level, and the trainee performs receiving training as the receiving party; the real-time receiving video of the trainee is collected and analyzed to obtain the real-time receiving feature set of the trainee and the overall receiving feature set of the trainee; the real-time receiving feature set of the trainee corresponds to the real-time receiving reaction of the trainee in each serving strategy; the overall receiving feature set of the trainee corresponds to the overall receiving reaction of the trainee in this round of serving strategy; the real-time receiving feature set of the trainee includes the receiving result of the trainee corresponding to each ball and the feature vector of the trainee's posture action category; the specific calculation method of the overall receiving feature set of the trainee is the same as the calculation method of the standard receiving feature set; Serving strategy adjustment module: Use the real-time receiving feature set of the trainer to determine and execute the serving strategy for the next ball; use the standard receiving feature set and the overall receiving feature set of the trainer to determine and execute the serving strategy for the next stage.

7. The artificial intelligence-based badminton serve control system according to claim 6, characterized in that: The standard battle video processing and analysis module specifically includes: Standard serve strategy set calculation unit: mark the two players in the i-th standard badminton match video as player A and player B respectively; identify and track player A in the standard badminton match video, and locate m i The m corresponding to athlete A in the round trip i The batting image frames IA1~IA mi , respectively, with each hitting image frame IA mi Centered on the battle video, extract IA mi The back k i image frame, and the current hitting image frame IA mi Together, the total number is k i +1: the key frame image set IA_i of the player A’s i-th shot; Calculate m i The batting image frames IA1~IA mi The timestamp difference between the adjacent frames before and after is obtained i -1 timestamp difference, used as the serve interval in the standard serve strategy set; Construct a spatial data model corresponding to a badminton court, and establish a coordinate mapping model between a pixel coordinate system and a court plane coordinate system; For m i The batting image frames IA1~IA mi Performing half-court court area identification and segmentation on each of the images to obtain a half-court court area image including the sneakers and badminton of athlete A; performing contour detection on the sneakers and badminton of athlete A in the half-court court area image to obtain a sneaker area and a badminton area, respectively calculating two minimum circumscribed rectangles corresponding to the sneaker area and the badminton area, respectively obtaining the sneaker centroid pixel coordinates and the badminton centroid pixel coordinates, obtaining the badminton height pixel coordinates through the relative positions between the sneaker centroid pixel coordinates and the badminton centroid pixel coordinates, obtaining the sneaker ground coordinates and the badminton height coordinates by using the coordinate mapping model, and using the sneaker ground coordinates and the badminton height coordinates as the serving position and serving height in the standard serving strategy set, respectively; Badminton recognition and tracking are performed on each image in the key frame image set IA_i to obtain a badminton motion key point trajectory sequence ((x1, y1, t1), (x2, y2, t2), (x3, y3, t3), ..., (x j ,y j ,t j )), where (x j ,y j ,t j ) represents the abscissa, ordinate and timestamp corresponding to the j-th key point; using the badminton motion key point trajectory sequence to calculate the flight angle and initial speed of the badminton, and using the flight angle as the serving angle in the standard serving strategy set; using the flight angle and initial speed to obtain the hitting force of the badminton, and using the hitting force as the serving force in the standard serving strategy set; The above processing is performed on the n standard badminton match videos of different difficulty levels respectively to obtain n standard serving strategy sets of different difficulty levels.

8. The artificial intelligence-based badminton serve control system according to claim 6, characterized in that: The standard battle video processing and analysis module specifically includes: Standard catch feature set calculation unit: mark the two players in the i-th standard badminton match video as player A and player B respectively; identify and track player B in the standard badminton match video, and locate m i The m corresponding to athlete B in the round trip i The batting image frames IB1~IB mi , respectively, with each hitting image frame IB mi As the center, extract IB in the battle video mi The k before and after i image frames, and the current hitting image frame IB mi Together the total number is 2k i +1: the key frame image set IB_i of the i-th shot of player B; Respectively identify and segment the athlete B for each image in the key frame image set IB_i to obtain a human body region image of the athlete B, perform key point detection on the human body region image to obtain a key point spatial coordinate set corresponding to a total of 14 key points including the nose, neck, left and right shoulders, left and right elbows, left and right wrists, left and right hips, left and right knees, and left and right ankles; Using the key point spatial coordinate set to obtain a joint angle feature set and a joint distance feature set, splicing the joint angle feature set and the joint distance feature set to obtain a posture action feature vector, inputting the posture action feature vector into a pre-trained posture action recognition model for recognition, and obtaining a corresponding posture action category, wherein the posture action category includes jumping, sideways, and arm raising; combining the posture action category corresponding to each image in the key frame image set IB_i in timestamp order to obtain a posture action category feature vector in a standard ball catching feature set; Using the key point spatial coordinate set of each adjacent image in the key frame image set IB_i, a moving speed sequence of each key point in the x and y directions is obtained as a key point moving speed feature vector in the standard ball receiving feature set; Respectively perform racket recognition and segmentation on each image in the key frame image set IB_i to obtain a racket area image, process the racket area image, and calculate the minimum circumscribed rectangle area value and the minimum circumscribed rectangle centroid coordinate value corresponding to the racket area; For m i The batting image frames IB1~IB mi For each image in the image, the racket and badminton body are identified, the coordinate value of the center of mass of the minimum circumscribed rectangle of the racket area and the coordinate value of the badminton collision point are calculated, and the relative coordinate value of the collision point is obtained; The area value, the center of mass coordinate value and the relative coordinate value of the ball-collision point corresponding to each image in the key frame image set IB_i are combined in the order of timestamps to obtain a racket motion feature vector in a standard ball-catching feature set; The above processing is performed on the n standard badminton match videos of different difficulty levels respectively to obtain n standard ball receiving feature sets of different difficulty levels.

9. A badminton serve control system based on artificial intelligence as claimed in claim 7 or 8, characterized in that: The serve control analysis module specifically includes: Real-time ball receiving feature set calculation unit: the trainer posture action category feature vector is obtained by processing and analyzing the real-time ball receiving video of the trainer, and the specific calculation method is the same as the calculation method of the athlete posture action category feature vector in the standard ball receiving feature set; Overall ball catching feature set calculation unit: The overall ball catching feature set of the training personnel is obtained by processing and analyzing the real-time ball catching video of the training personnel.

10. The artificial intelligence-based badminton serve control system according to claim 6, characterized in that: The serving strategy adjustment module specifically includes: If the result of the catch in the real-time catch feature set is a failure, and the similarity between the training personnel's posture action category feature vector and the corresponding player's posture action category feature vector in the standard catch feature set is less than the preset threshold T p , the serving strategy for the next ball repeats the current serving strategy, otherwise, the serving strategy for the next ball corresponding to the predetermined standard serving strategy is used; the use of the standard receiving feature set and the overall receiving feature set of the training personnel to determine the serving strategy for the next stage and execute it specifically includes: if the similarity between the overall receiving feature set of the training personnel and the standard receiving feature set is less than a preset threshold value T w , and the overall ball receiving error rate is greater than the preset threshold T n , then the serving strategy of the next stage repeats the serving strategy of this stage. Otherwise, it is considered that the receiving training corresponding to the serving strategy of this stage is qualified, and a more difficult training can be carried out in the next stage.

Images