Intelligent auxiliary training method, system, equipment and medium for badminton

Abstract

The present invention discloses an intelligent auxiliary training method, system, device and medium for badminton, which belongs to the field of computer vision technology, including: setting multiple cameras, selecting one as a default camera, obtaining a first target image shot by the default camera, and target pixels of a target object; forming a target area based on the target pixels, obtaining a first posture of the target object in the target area, comparing the first posture with a second posture in a first database, and calculating the highest similarity of the first posture; obtaining a second target image, the second target image is the Kth frame image shot by the default camera, judging whether the target object in the second target image is blocked, if so, obtaining a supplementary video image and analyzing it, otherwise obtaining the first posture from the second target image; adding the similarities of all first postures to obtain an average value, and obtaining a motion score of the target object. The present invention improves the accuracy of posture recognition in badminton training.

Description

Technical Field

[0001] The present invention belongs to the field of computer vision technology, and in particular relates to an intelligent auxiliary training method, system, equipment and medium for badminton. Background Art

[0002] In badminton training, the athlete's posture and movement accuracy are crucial to improving sports and auxiliary training. With the rapid development of computer vision, obtaining the athlete's body posture through video analysis and generating auxiliary training data has become a new training method.

[0003] Currently, intelligent assisted training methods for badminton, such as the Chinese patent document with publication number CN116958872A, discloses an intelligent assisted training method for badminton. The position information and timing information of the ball are output from the two-dimensional ball path detection and tracking, and the position information and trajectory are restored and optimized in three-dimensional space. A method for acquiring the athlete's technical movements based on binocular and monocular perspectives is implemented, and additional conditional constraints are introduced from multiple angles such as timing, venue, and competition rules, in an attempt to provide a more accurate estimation result for the three-dimensional human posture of badminton players under a monocular perspective.

[0004] However, the above-mentioned prior art method needs to introduce additional condition constraints from multiple angles such as timing, venue, and competition rules. This may cause the accuracy and effectiveness of the system to be affected under certain special conditions, such as non-standard venues or changes in competition rules. In addition, the limitations of monocular vision, such as perspective occlusion, lighting changes and other factors, may cause the accuracy of human posture estimation to be affected. Therefore, an intelligent assisted training method for badminton is needed to provide higher accuracy in human posture estimation. Summary of the invention

[0005] To solve the above problems, the present invention provides an intelligent auxiliary training method, system, device and medium for badminton to solve the problems in the prior art.

[0006] In order to achieve the above-mentioned object of the invention, the present invention proposes an intelligent auxiliary training method for badminton, comprising:

[0007] S1: multiple cameras are set, and one camera is selected as a default camera. The remaining cameras simultaneously shoot supplementary video images, and a first target image shot by the default camera is obtained. The first target image is a first frame image including a target object, and a target pixel of the target object is obtained based on the first target image.

[0008] S2: forming a target area based on the target pixels, acquiring a first posture of the target object in the target area, establishing a first database, comparing the first posture with a second posture in the first database, calculating a highest similarity of the first posture, and storing the first posture of the first target image and the highest similarity in the first database;

[0009] S3: Acquire a second target image, where the second target image is the Kth frame image taken by the default camera and after the first target image, locate the target area in the second target image based on the target pixel, and determine whether the target object is blocked; if blocked, acquire the supplementary video image, analyze the supplementary video image to acquire the first posture of the target object at this time; if not blocked, acquire the first posture from the second target image, and store the first posture in the first database;

[0010] S4: Adding the highest similarities of all the first postures in the first database, calculating an average value, and using the average value as a motion score corresponding to the target object.

[0011] Further, locating the target area in the second target image comprises the following steps:

[0012] In the first target image, the size of the target area is set based on the target object, the color of the pixel points in the target area of ​​the first target image is obtained and defined as the target pixel points, and a first number of the target pixel points of different colors in the target area is calculated; in the second target image, an alternative position where the target pixel points appear is located, and an alternative area is generated at the alternative position, the size of the alternative area is the same as the size of the target area, and a second number of the pixel points of different colors in each of the alternative areas is calculated, and the difference in the number of the pixel points of different colors is calculated based on the first number and the second number, and the alternative area where the sum of the differences is the smallest is set as the target area.

[0013] Furthermore, acquiring the first posture of the target object in the target area includes the following steps:

[0014] A first pixel is obtained, where the first pixel is any pixel in the target area, and a first difference between the first pixel and each surrounding pixel is calculated. If the first difference is greater than or equal to a first threshold and less than a second threshold, the first pixel is defined as a boundary pixel of the target object, and the boundary pixels are connected in sequence to obtain a boundary contour, and the boundary contour is defined as the first posture of the target object.

[0015] Further, calculating and obtaining the highest similarity of the first posture includes the following steps:

[0016] A plurality of first detection points are set on the boundary area, a plurality of second detection points are set on the boundary area of ​​the second posture in the first database, a reference position is set, the first posture and the second posture are placed based on the reference position, and a similarity α between the first posture and each of the second postures is calculated based on a first formula, wherein the first formula is: Wherein, N is the number of the first detection points, q I is the three-dimensional spatial position coordinate of the first detection point, q J is the three-dimensional spatial position coordinate of the Jth second detection point, Q1 is the set of the first detection points, Q2 is the set of the second detection points, min is the minimum value function for obtaining the second detection point closest to the first detection point, ||q I -q J || 2 is the first detection point q I With the second detection point q J The square of the distance between the two positions is calculated, and the size of each similarity is compared to obtain the highest similarity of the first posture.

[0017] Further, determining whether the target object is blocked includes the following steps:

[0018] Obtain the K-2 and K-1 frames of images taken by the default camera, which are defined as the third target image and the fourth target image respectively, identify the third target image and the fourth target image, obtain a first skeleton structure and a second skeleton structure, wherein the skeleton structure includes a plurality of skeleton nodes, annotate key nodes in the skeleton nodes, and calculate the first distance between each skeleton node and the default camera based on a monocular ranging algorithm;

[0019] Based on the time series of the third target image and the fourth target image, the movement trend of the bone nodes is determined; based on the movement trend and the second bone structure, the third bone structure of the second target image is generated; based on the bone nodes in the third bone structure, a human body model is generated; if there is an overlapping area in the human body model, the bone nodes appearing in the overlapping area are defined as target nodes; if the target node includes the key node, and the first distance of the key node is greater than that of the remaining target nodes, it is determined that the target object is obstructed.

[0020] Furthermore, after storing the first posture in the first database, the following operations are also included:

[0021] The position information of each key node in each of the first postures is obtained, the rotation angle data corresponding to each of the first postures is generated, the second difference of the rotation angle data of different first postures is calculated, and the first postures whose second difference is less than a third threshold are eliminated.

[0022] The present invention also provides an intelligent auxiliary training system for badminton, which is used to implement the above-mentioned method, and the system mainly includes:

[0023] An acquisition module is provided, wherein a plurality of cameras are set, and one camera is selected as a default camera, and the remaining cameras simultaneously shoot supplementary video images, and a first target image shot by the default camera is acquired, wherein the first target image is a first frame image including a target object, and a target pixel of the target object is acquired based on the first target image;

[0024] a first generating module, forming a target area based on the target pixels, acquiring a first posture of the target object in the target area, establishing a first database, comparing the first posture with a second posture in the first database, acquiring the second posture having the highest similarity to the first posture, and storing the first posture of the first target image and the similarity in the first database;

[0025] A second generation module is configured to obtain a second target image, where the second target image is the Kth frame image taken by the default camera and after the first target image, locate the target area in the second target image based on the target pixel, and determine whether the target object is blocked; if blocked, obtain the supplementary video image, analyze the supplementary video image to obtain the first posture of the target object at this time; if not blocked, obtain the first posture from the second target image, and store the first posture in the first database;

[0026] A scoring module adds the similarities of all the first postures in the first database, calculates an average value, and uses the average value as the motion score corresponding to the target object.

[0027] The present invention also provides a device, which is used to implement the above method, comprising: a processor, a memory and a communication bus, wherein the processor and the memory communicate with each other through the communication bus;

[0028] The memory is used to store computer programs;

[0029] The processor is used to execute the program stored in the memory to implement the method described in any one of claims 1-6.

[0030] The present invention also provides a computer medium, which stores program instructions, wherein when the program instructions are executed, the device where the computer medium is located is controlled to execute the above-mentioned method.

[0031] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0032] Through the present invention, a target area is formed based on target pixels in the first target image under the default camera, and posture recognition is performed within the target area, so that the scope of image processing is clarified and the interference of background information on the recognition result is effectively reduced; by comparing the first posture with the second posture known in the database, the posture with the highest similarity to the target posture is found, thereby improving the accuracy of posture recognition and reducing the possibility of recognition errors; a camera under a multi-viewing angle is used to determine the first posture of the target object, and even if the target object is blocked, the posture information can be recognized in real time, thereby increasing the robustness of posture recognition; by storing the first posture and the similarity value in the sub-first database, the sports score corresponding to the target object is obtained, and a visual data presentation and quantitative information feedback are provided for athletes or trainers. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A flowchart of the steps of the intelligent auxiliary training method for badminton of the present invention;

[0034] Figure 2 The figure is a structural diagram of the intelligent auxiliary training system for badminton of the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0036] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0037] like Figure 1 As shown, an intelligent auxiliary training method for badminton includes:

[0038] S1: multiple cameras are set, and one camera is selected as a default camera. The remaining cameras simultaneously shoot supplementary video images to obtain a first target image shot by the default camera. The first target image is a first frame image including a target object. The target pixel of the target object is obtained based on the first target image.

[0039] Specifically, in this embodiment, in order to better identify the human posture of athletes or trainers during badminton training, multiple cameras at different viewing angles are set for video shooting, and a camera that can maximize the capture of the target object is selected as the default camera by comprehensively considering multiple factors such as viewing angle, image clarity, frame rate and stability, and lighting conditions. The first target image is the first frame image taken by the default camera. The image may include multiple moving objects, such as multiple athletes or trainers, badminton, badminton rackets, etc., and the target object is the athlete A designated for observation.

[0040] S2: forming a target area based on target pixels, obtaining a first posture of the target object in the target area, establishing a first database, comparing the first posture with a second posture in the first database, calculating a highest similarity of the first posture, and storing the first posture of the first target image and the highest similarity in the first database.

[0041] Specifically, in the present embodiment, firstly, the target area is determined according to the pixel ratio of the target pixel in the target object, the target area is a rectangular or square area, including the athlete A, but it may also include other sports objects, such as badminton, the boundary pixels in the target area are identified to form the first posture of the target object (follow-up description), the first posture refers to the posture contour of the target object in the first target image, the first posture is compared with the second posture stored in advance in the first database, the second posture is the standard posture of the athlete in each state during the badminton training process, the second posture with the highest similarity to the first posture is obtained, the similarity calculation method is explained later, the information of the maximum similarity and the first posture are stored in the first database, the first database is used to store the posture information of each target image (the first target image, the second target image, ..., the nth target image), and the image with slight changes can also be filtered according to the size of the posture change in the target image (follow-up description), so as to increase the storage space of the first database, and facilitate data processing. By determining the target area, the present invention reduces the processing range of the first target image to a certain extent, enhances the efficiency of the target object posture recognition, stores all posture information and similarity information in the first database, and facilitates the display and query of subsequent data.

[0042] S3: Acquire a second target image, which is the Kth frame image taken by the default camera and after the first target image. Locate the target area in the second target image based on the target pixel, and determine whether the target object is blocked. If blocked, acquire a supplementary video image, analyze the supplementary video image to obtain the first posture of the target object at this time. If not blocked, acquire the first posture from the second target image, and store the first posture in the first database.

[0043] Specifically, in this embodiment, during the training process, athlete A needs to frequently move, jump and hit the ball based on the sending route of the badminton. During this process, the default camera will capture multiple frames of images, wherein the second target image is the Kth frame image after the first target image, and the athlete A in the second target image is identified, and the target area in the second target graphic is located according to the pixel ratio of athlete A (described later). The athlete A in the target area may be partially obstructed by other athletes or moving objects such as badminton due to movement, so that the first posture of the athlete A cannot be fully seen only from the perspective of the default camera. In this case, it is necessary to obtain video images captured by cameras at other perspectives, and obtain the first posture of the target object from the video image. If no obstruction occurs, the first posture is directly obtained from the second target image, and the first posture is stored in the first database.

[0044] S4: adding up the highest similarities of all first postures in the first database, calculating an average value, and using the average value as the motion score corresponding to the target object.

[0045] Specifically, in this embodiment, the highest similarities corresponding to each first posture in the first database are added and averaged to obtain the sports score of athlete A in the entire badminton training process. The higher the average value, the higher the level of the athlete's badminton training. The first posture of athlete A in multiple motion states can also be targeted. The higher the similarity of the first posture, the more standard the movement of athlete A in this motion state. If the similarity is low, training adjustments can be made according to the second posture in the corresponding state.

[0046] Through the present invention, a target area is formed based on target pixels in the first target image under the default camera, and posture recognition is performed within the target area, so that the scope of image processing is clarified and the interference of background information on the recognition result is effectively reduced; by comparing the first posture with the second posture known in the database, the posture with the highest similarity to the target posture is found, thereby improving the accuracy of posture recognition and reducing the possibility of recognition errors; a camera under a multi-viewing angle is used to determine the first posture of the target object, and even if the target object is blocked, the posture information can be recognized in real time, thereby increasing the robustness of posture recognition; by storing the first posture and the similarity value in the sub-first database, the sports score corresponding to the target object is obtained, and a visual data presentation and quantitative information feedback are provided for athletes or trainers.

[0047] It is particularly noteworthy that, through the above technical solution, locating the target area in the second target image includes the following steps:

[0048] In a first target image, a size of a target area is set based on a target object, a color including pixels is obtained in the target area of ​​the first target image and defined as target pixels, a first number of target pixels of different colors in the target area is calculated, and in a second target image, an alternative position where the target pixels appear is located, an alternative area is generated at the alternative position, a size of the alternative area is the same as that of the target area, a second number of pixels of different colors in each alternative area is calculated, a difference in the number of pixels of different colors is calculated based on the first number and the second number, and the alternative area with the smallest sum of the differences is set as the target area.

[0049] Specifically, in the present embodiment, the pixel points of the target area in the first target image are defined as target pixel points, the target pixel points include the pixel points of athlete A, for example, include blue, black, white and red, the first numbers of the four target pixel points of athlete A are 30, 30, 20 and 10 respectively, the total pixel value is 90, the size of the target area in the first target image is set to 12*12 according to the total pixel value and the pixel position distribution, in the second target image, its target area is set to the same size of 12*12, due to the change of the posture of athlete A during the training process, the target area may not only include the four target pixel points of a certain athlete, and the second numbers of the four target pixel points are 29, 31, 20, 9, but also include the target pixel points of badminton B, the colors of the target pixel points of badminton B include black and white, the second number is 2 and 6, two alternative areas are generated based on the two moving objects of a certain athlete and badminton B, the difference between the number of pixels of athlete A in the first target image is calculated, and the target area where the athlete is located is the target area of ​​athlete A.

[0050] Acquiring the first posture of the target object in the target area includes the following steps:

[0051] A first pixel is obtained, where the first pixel is any pixel in the target area, and a first difference between the first pixel and each surrounding pixel is calculated. If the first difference is greater than or equal to a first threshold and less than a second threshold, the first pixel is defined as a boundary pixel of the target object, and the boundary pixels are connected in sequence to obtain a boundary contour, and the boundary contour is defined as a first posture of the target object.

[0052] Specifically, in this embodiment, within the target area, any pixel is defined as a first pixel, and the target pixel often exists as a partial whole. The first difference between any first pixel and the surrounding pixels is calculated. If the first difference satisfies a value greater than or equal to a first threshold and less than a second threshold, the first threshold and the second threshold can be set by obtaining the background pixels of the target area and the target pixels of athlete A. The purpose is to identify pixels where pixel value jumps occur and the range of the jump is within a preset range. Such pixels are defined as boundary pixels, and contour information of the first posture of the target object is formed based on the boundary pixels.

[0053] Calculating the highest similarity of the first posture includes the following steps:

[0054] A plurality of first detection points are set on the boundary area, a plurality of second detection points are set on the boundary area of ​​the second posture in the first database, a reference position is set, the first posture and the second posture are placed based on the reference position, and the similarity α between the first posture and each second posture is calculated based on the first formula, the first formula is: Where N is the number of the first detection points, q I is the three-dimensional spatial position coordinate of the first detection point, q J is the three-dimensional spatial position coordinate of the Jth second detection point, Q1 is the set of first detection points, Q2 is the set of second detection points, min is the minimum value function for obtaining the second detection point closest to the first detection point, ||q I -q J || 2 The first detection point q I With the second detection point q J The square of the distance between them is calculated, and the size of each similarity is compared to obtain the highest similarity of the first posture.

[0055] Specifically, in this embodiment, for example, 50 first detection points are set on the boundary area of ​​the first posture, and 50 second detection points are also set on the boundary area of ​​the second posture in the first database. A plurality of reference positions are set, such as the head, left shoulder, right shoulder, elbow, wrist, finger and waist, which are positions that are important for the posture. The first posture and the second posture are placed on the same spatial plane with the reference position as the reference point, and the first formula is used to calculate the first posture. Calculate the similarity α between the first posture and each second posture, where, according to min||q I -q J || 2 Get the distance between the second detection point closest to the first detection point and the first detection point, q I and q J The three-dimensional spatial position coordinates of the Ith first detection point and the three-dimensional spatial position coordinates of the Jth second detection point are respectively used. The reciprocal of the average value of the sum of the distances between all the first detection points and the second detection point in the closest second posture is taken as the similarity between the first posture and the second posture. The highest similarity of the first posture is obtained by comparing the size of each similarity and stored in the first database.

[0056] Determining whether the target object is blocked includes the following steps:

[0057] The K-2 and K-1 frames taken by the default camera are obtained, which are defined as the third target image and the fourth target image respectively. The third target image and the fourth target image are identified to obtain the first skeleton structure and the second skeleton structure. The skeleton structure includes multiple skeleton nodes, key nodes are marked in the skeleton nodes, and the first distance between each skeleton node and the default camera is calculated based on the monocular ranging algorithm.

[0058] Based on the time series of the third target image and the fourth target image, the movement trend of the bone nodes is determined; based on the movement trend and the second bone structure, a third bone structure of the second target image is generated; based on the bone nodes in the third bone structure, a human body model is generated; if there is an overlapping area in the human body model, the bone nodes appearing in the overlapping area are defined as target nodes; if the target nodes include key nodes, and the first distance of the key nodes is greater than that of the remaining target nodes, it is determined that the target object is obstructed.

[0059] Specifically, in this embodiment, according to the third target image and the fourth target image of athlete A in the K-2th and K-1th frame images taken by the default camera, the three-dimensional bone node information of athlete A in the target image is detected by OpenPose and other technologies, and a bone structure is formed according to the bone node information. The bone node information may include the head, neck, left and right shoulders, left and right elbows, left and right wrists, left and right hips, left and right knees, femoral head, left and right ankles, left and right feet, spine and pelvis, etc., and the bone node information that is more important for badminton training posture recognition is selected as the key nodes, such as the head, neck, left and right shoulders, left and right elbows, left and right wrists, left and right hips, left and right knees, left and right ankles, left and right feet, and left and right hands. The first distance between each bone node and the default camera is calculated based on the monocular ranging algorithm, wherein the monocular ranging algorithm is a computer vision technology that estimates the distance between an object and a camera through a single camera.

[0060] Then, the movement trend of the skeleton node is determined according to the difference in the positions of the skeleton node in the third target image and the fourth target image. For example, the positions of the skeleton node Q in the third target image and the fourth target image, i.e., the K-2 and K-1 frame images, are P and P, respectively. k-2 and P k-1 , you can set different weights W1 and W2 respectively, according to the formula P k =W1P k-1 +W2P k-2 , obtain the position information P of the skeleton node Q in the second target image k , and generating a third bone structure of a second target image according to the distance between each bone node in the second bone structure.

[0061] A human body model is generated based on the position and distance of the bone nodes in the third bone structure. Due to the different hitting postures of athlete A in the process of playing badminton, there may be overlapping areas in the human body model. For example, athlete A jumps in the process of catching the ball (assuming that athlete A hits the ball with his right hand). In the process of jumping, the left hand position overlaps with the left thigh position (excluding the left knee) to form an overlapping area. The left hand position is defined as the target node, and the left hand position is a key node set in advance. Then, the first distance between the left hand position and the default camera and the difference between the first distances of other target nodes in the overlapping area of ​​the left thigh position and the default camera are determined. Other target nodes include the femoral head, greater trochanter, lesser trochanter, etc. If the first distance of the key node is greater than that of the other target nodes, it means that the left hand position as the key node is blocked, and the specific position information cannot be recognized under the default camera, and posture recognition is required through cameras with other perspectives.

[0062] After storing the first posture in the first database, the following operations are also included:

[0063] The position information of each key node in each first posture is obtained, the rotation angle data corresponding to each first posture is generated, the second difference of the rotation angle data of different first postures is calculated, and the first postures whose second difference is less than the third threshold are eliminated.

[0064] Specifically, in this embodiment, according to the first posture under each target image in the first database, the position information of the key nodes in the first posture is obtained, and according to the distance between the bone nodes in the bone model, as well as the principles of inverse kinematics and forward kinematics, the rotation angle of each key node, that is, the joint angle and the joint distance, are obtained, and the second difference of the rotation angle at the same key node in multiple first postures is calculated, and the first postures with the second difference less than a third threshold value (which can be set to 1), that is, the first postures with slight changes, are eliminated, and only multiple first postures representing the motion state remain, for example: preparation state, serving state, receiving state, moving state (different screening can be performed for different actions), hitting state, jump kill state, etc., so as to reduce redundant data, optimize the storage space of the first database, improve the accuracy of badminton player posture analysis, and speed up data processing.

[0065] like Figure 2 As shown, the present invention also provides an intelligent auxiliary training system for badminton, which is used to implement the above method, and the system mainly includes:

[0066] An acquisition module is provided, wherein multiple cameras are set and one camera is selected as a default camera. The remaining cameras simultaneously shoot supplementary video images, and a first target image shot by the default camera is acquired. The first target image is a first frame image including a target object. The target pixel of the target object is acquired based on the first target image.

[0067] The first generation module forms a target area based on target pixels, obtains a first posture of the target object in the target area, establishes a first database, compares the first posture with a second posture in the first database, calculates the highest similarity of the first posture, and stores the first posture of the first target image and the highest similarity in the first database.

[0068] The second generation module obtains a second target image, which is the Kth frame image taken by the default camera and after the first target image. The target area in the second target image is located based on the target pixel, and it is determined whether the target object is blocked. If blocked, a supplementary video image is obtained, and the supplementary video image is analyzed to obtain a first posture of the target object at this time. If not blocked, the first posture is obtained from the second target image, and the first posture is stored in the first database.

[0069] The scoring module adds up the highest similarities of all first postures in the first database, calculates an average value, and uses the average value as the motion score corresponding to the target object.

[0070] The present invention also provides a device for implementing the above method, comprising: a processor, a memory and a communication bus, wherein the processor and the memory communicate with each other via the communication bus.

[0071] Memory, used to store computer programs.

[0072] A processor, configured to execute a program stored in a memory to implement the method of any one of claims 1 to 6.

[0073] The present invention also provides a computer medium, which stores program instructions, wherein when the program instructions are executed, the device where the computer medium is located is controlled to execute the above method.

[0074] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0075] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The above-mentioned program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0076] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0077] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0078] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent assisted training method for badminton, characterized in that: The method comprises the following steps: S1: multiple cameras are set, and one camera is selected as a default camera. The remaining cameras simultaneously shoot supplementary video images, and a first target image shot by the default camera is obtained. The first target image is a first frame image including a target object, and a target pixel of the target object is obtained based on the first target image. S2: forming a target area based on the target pixels, acquiring a first posture of the target object in the target area, comparing the first posture with a second posture in a database, calculating a highest similarity of the first posture, and storing the first posture of the first target image and the highest similarity in a database; Wherein, acquiring the first posture of the target object in the target area comprises the following steps: Acquire a first pixel, where the first pixel is any pixel in the target area, calculate a first difference between the first pixel and each surrounding pixel, and if the first difference is greater than or equal to a first threshold and less than a second threshold, define the first pixel as a boundary pixel of the target object, connect the boundary pixels in sequence to obtain a boundary contour, and define the boundary contour as the first posture of the target object; Calculating and obtaining the highest similarity of the first posture comprises the following steps: A plurality of first detection points are set on the boundary contour, a plurality of second detection points are set on the boundary contour of the second posture in the database, a reference position is set, the first posture and the second posture are placed based on the reference position, and the similarity between the first posture and each of the second postures is calculated based on a first formula , the first formula is: ,in, is the number of the first detection points, For the The three-dimensional spatial position coordinates of the first detection point, For the The three-dimensional spatial position coordinates of the second detection point, is the set of the first detection points, is the set of the second detection points, min is the minimum value function for obtaining the second detection point closest to the first detection point, The first detection point With the second detection point The square of the distance between the two positions is calculated, and the magnitude of each similarity is compared to obtain the highest similarity of the first posture; S3: Acquire a second target image, where the second target image is the Kth frame image taken by the default camera and after the first target image, locate the target area in the second target image based on the target pixel, and determine whether the target object is blocked. If blocked, acquire the supplementary video image, analyze the supplementary video image to acquire the first posture of the target object at this time; if not blocked, acquire the first posture from the second target image, and store the first posture in a database; Wherein, determining whether the target object is blocked comprises the following steps: Obtain the K-2 and K-1 frames of images taken by the default camera, which are defined as the third target image and the fourth target image respectively, identify the third target image and the fourth target image, obtain a first skeleton structure and a second skeleton structure, wherein the skeleton structure includes a plurality of skeleton nodes, annotate key nodes in the skeleton nodes, and calculate the first distance between each skeleton node and the default camera based on a monocular ranging algorithm; Determining the movement trend of the skeleton nodes based on the time series of the third target image and the fourth target image, generating a third skeleton structure of the second target image based on the movement trend and the second skeleton structure, generating a human body model based on the skeleton nodes in the third skeleton structure, and if the human body model has an overlapping area, defining the skeleton nodes appearing in the overlapping area as target nodes, and if the target node includes the key node, and the first distance of the key node is greater than that of the remaining target nodes, determining that the target object is blocked; After acquiring the first posture from the second target image, the following operations are also included: Obtaining position information of each key node in each of the first postures, generating rotation angle data corresponding to each of the first postures, calculating second differences of the rotation angle data of different first postures, and eliminating the first postures whose second differences are less than a third threshold value; S4: Adding the highest similarities of all the first postures in the database, calculating an average value, and using the average value as the motion score corresponding to the target object.

2. The method according to claim 1, characterized in that Locating the target area in the second target image comprises the following steps: In the first target image, the size of the target area is set based on the target object, the color of the pixel points in the target area of ​​the first target image is obtained and defined as the target pixel points, and a first number of the target pixel points of different colors in the target area is calculated; in the second target image, an alternative position where the target pixel points appear is located, and an alternative area is generated at the alternative position, the size of the alternative area is the same as the size of the target area, and a second number of the pixel points of different colors in each of the alternative areas is calculated, and the difference in the number of the pixel points of different colors is calculated based on the first number and the second number, and the alternative area where the sum of the differences is the smallest is set as the target area.

3. An intelligent auxiliary training system for badminton, used to implement the method according to any one of claims 1-2, characterized in that: The system includes the following modules: An acquisition module is provided, wherein a plurality of cameras are set, and one camera is selected as a default camera, and the remaining cameras simultaneously shoot supplementary video images, and a first target image shot by the default camera is acquired, wherein the first target image is a first frame image including a target object, and a target pixel of the target object is acquired based on the first target image; A first generation module forms a target area based on the target pixels, obtains a first posture of the target object in the target area, compares the first posture with the second posture in the database, and calculates the highest similarity of the first posture, wherein obtaining the first posture of the target object in the target area includes the following steps: obtaining a first pixel, the first pixel being any pixel in the target area, calculating a first difference between the first pixel and each surrounding pixel, if the first difference is greater than or equal to a first threshold and less than a second threshold, defining the first pixel as a boundary pixel of the target object, connecting the boundary pixels in sequence to obtain a boundary contour, and defining the boundary contour as the first posture of the target object; calculating the highest similarity of the first posture includes the following steps: setting a plurality of first detection points on the boundary contour, setting a plurality of second detection points on the boundary contour of the second posture in the database, setting a reference position, placing the first posture and the second posture based on the reference position, and calculating the similarity of the first posture with each of the second postures based on a first formula. , the first formula is: ,in, is the number of the first detection points, For the The three-dimensional spatial position coordinates of the first detection point, For the The three-dimensional spatial position coordinates of the second detection point, is the set of the first detection points, is the set of the second detection points, min is the minimum value function for obtaining the second detection point closest to the first detection point, The first detection point With the second detection point The square of the distance between the two positions is calculated, and the magnitude of each similarity is compared to obtain the highest similarity of the first posture; a second generation module, obtaining a second target image, where the second target image is the Kth frame image taken by the default camera and after the first target image, locating the target area in the second target image based on the target pixel, and judging whether the target object is blocked; if blocked, obtaining the supplementary video image, analyzing the supplementary video image to obtain the first posture of the target object at this time; if not blocked, obtaining the first posture from the second target image, wherein judging whether the target object is blocked includes the following steps: obtaining the K-2th and K-1th frame images taken by the default camera, defining them as the third target image and the fourth target image respectively, identifying the third target image and the fourth target image, obtaining a first skeletal structure and a second skeletal structure, wherein the skeletal structure includes a plurality of skeletal nodes, marking key nodes in the skeletal nodes, and calculating the relationship between each skeletal node and the skeletal structure based on a monocular ranging algorithm; The first distance of the default camera is used to determine the movement trend of the skeleton nodes based on the time series of the third target image and the fourth target image, generate the third skeleton structure of the second target image based on the movement trend and the second skeleton structure, and generate a human body model based on the skeleton nodes in the third skeleton structure. If there is an overlapping area in the human body model, the skeleton nodes appearing in the overlapping area are defined as target nodes. If the target node includes the key node and the first distance of the key node is greater than the remaining target nodes, it is determined that the target object is blocked. After obtaining the first posture from the second target image, the following operations are also included: obtaining the position information of each key node in each of the first postures, generating rotation angle data corresponding to each of the first postures, calculating the second difference of the rotation angle data of different first postures, and eliminating the first postures whose second difference is less than the third threshold value. The scoring module adds up the highest similarities of all the first postures, calculates an average value, and uses the average value as the motion score corresponding to the target object.

4. An intelligent auxiliary training device for badminton, characterized in that: The device comprises a processor and a memory, wherein the memory is used to store at least one computer program, and the at least one computer program is loaded by the processor to execute the method according to any one of claims 1 to 2.

5. A computer medium, characterized in that The computer medium stores program instructions, wherein when the program instructions are executed, the device where the computer medium is located is controlled to execute the method according to any one of claims 1 to 2.

Images