CN116271757A - Auxiliary system and method for basketball practice based on AI technology - Google Patents

Abstract

The invention discloses an auxiliary system and method for basketball practice based on AI technology, including an AI data acquisition unit, a display query terminal and an analysis and processing unit arranged in the background; the AI data acquisition unit collects player facial information and motion video data; The recognition processing server is used to analyze and process sports video data to generate 3D skeleton data of athletes; the virtual simulation server is used to model stadiums and roles, bind athletes 3D skeleton data and role modeling, and generate rendering models for query; venue display The screen is used to display the query rendering model; the present invention generates a rendering model for query through the binding of the three-dimensional skeleton data of the athlete and the modeling of the virtual character; the athlete can look back at his own training actions from a three-dimensional perspective, reducing the intensity of repeated teaching by the coach; the athlete can Visually view the training situation, find movement defects, and take targeted measures to improve training efficiency.

Description

An auxiliary system and method for basketball practice based on AI technology

technical field

The invention relates to the technical field of basketball auxiliary practice equipment, in particular to an auxiliary system and method for basketball practice based on AI technology.

Background technique

Basketball exercises mainly train their own operational skills by imitating the form of specific actions. Athletes need to frequently imitate, observe and correct actions. The current basketball teaching mostly learns basketball skills by observing the coach's demonstration actions. After the first sports training demonstration, the accuracy of the teaching movements will decrease due to the large amount of physical exertion, which will affect the basketball training effect of the students. At the same time, the athletes cannot review their own sports images through the display device in real time, so that they can see their feet intuitively. The point of focus, take-off height, take-off angle, the size of the instantaneous explosive force during the ball control movement, whether the body movements during the ball control movement are kept as technical movements, etc., to further understand the shortcomings of your movements, compare them with standard movements, and find the gaps in movements Therefore, the coaches cannot visually grasp the training progress of the basketball players.

The patent document whose document number is CN110052002A discloses a basketball shooting training auxiliary system, including a controller, a wireless signal receiving unit, a backboard, a bracelet and a first camera, a basket is installed on the backboard, and a An inclination sensor, an acceleration sensor and a wireless signal sending unit, the acceleration sensor and the inclination sensor are connected to the wireless signal sending unit circuit, the wireless signal sending unit is connected to the controller, the board surface of the backboard is made of light-transmitting material, A horizontal LED light strip and a vertical LED light strip are arranged on the back of the backboard.

According to the patent, the basketball shooting training auxiliary system can quickly cultivate a player's shooting feel and improve shooting accuracy. However, athletes cannot look back at their own training actions from a three-dimensional perspective; they cannot automatically generate training progress analysis reports and individual athlete movement analysis reports for targeted training.

The patent document whose document number is CN111724414A discloses a basketball motion analysis method based on 3D posture estimation, and discloses a basketball training auxiliary system method based on posture estimation and action recognition pair analysis.

The system analyzes basketball games and training videos. The coaches find weaknesses in the game through the system analysis results, improve the level of competition, and make targeted preparations for future games in advance, arrange players and game strategies according to the opponent's characteristics, and The results of the analysis are fed back to the athletes to help them reduce the risk of injury. The system adopts the multi-view stereo posture detection method, obtains the 3D posture and position information of the players on the court through multiple video cameras with depth information, recognizes the players' movements and tracks the player's trajectory, and finally predicts and analyzes the players' movements , to build the player's motion posture and trajectory model, and the system will recognize the motion posture on the field.

But it has the following problems: 1. Pay attention to player trajectory tracking, predict and analyze the player's movements, which is suitable for basketball tactics teaching, but not very applicable to basketball training; 2. There is no three-dimensional dynamic output without image, and athletes cannot visualize three-dimensional Look back at your own training actions from an angle; 3. Progress analysis report on unassisted basketball training.

Contents of the invention

In response to the above problems, the purpose of the present invention is to provide an auxiliary system and method for basketball practice based on AI technology, which can assist coaches to reduce the intensity of repeated teaching; help athletes to look back at their own training actions from a three-dimensional angle, find action defects, and take targeted actions. Measures to improve training efficiency.

The object of the present invention can be achieved through the following technical solutions: an auxiliary system for basketball practice based on AI technology, including an AI data collection unit arranged on a physical basketball court, a display query terminal and an analysis and processing unit arranged at the background;

The AI data collection unit includes facial recognition collection equipment and motion video collection equipment for collecting athlete facial information and motion video data;

The analysis and processing unit includes an AI recognition processing server, a virtual simulation server and a data storage server;

The display query terminal includes a venue display screen;

The AI recognition processing server is used to analyze and process motion video data to generate three-dimensional skeleton data of athletes;

The virtual simulation server is used to bind and generate rendering models for venue modeling and role modeling, athlete three-dimensional skeleton data and role modeling for query;

The site display screen is used to display the query rendering model.

An auxiliary method for basketball practice based on AI technology, comprising the following steps:

S1. The virtual simulation server models the basketball venue and the player's role, realizing the binding of the digitally simulated basketball court and the real basketball court;

S2. Facial recognition collection equipment collects athlete information and establishes athlete training files;

S3. The motion video collection device collects the athlete's motion video and transmits it to the AI recognition processing server;

S4. The AI recognition processing server recognizes and analyzes sports videos, and generates three-dimensional skeleton data of athletes;

S5. The virtual simulation server realizes the binding of the athlete's three-dimensional skeleton data and role modeling, generates rendering model data, and transmits the rendering model data to the data storage server and the display query terminal;

S6. The display query terminal displays the query dynamic rendering model.

Further: in the step S4, the AI recognition processing server recognizes and analyzes the motion video to generate three-dimensional skeleton data of the athlete, including the following steps:

S41. Establishing a data set;

S42. Extract two-dimensional skeleton data, compare the data sets, and obtain the athlete's two-dimensional posture coordinates;

S43. Convert the two-dimensional attitude coordinates into three-dimensional attitude coordinates to obtain three-dimensional skeleton data.

A kind of auxiliary method of basketball practice based on AI technology according to claim 3, it is characterized in that: extracting two-dimensional skeleton data in the described S42 comprises the following steps:

S421. Normalize the input image, establish a detection frame, and detect the position of the human body in the detection frame;

S422. Input the normalized image into the pose estimation module, generate a pose estimation suggestion, compare the pose estimation suggestion with the actual pose of the data set, and generate a pose estimation result;

S423. Generate two-dimensional skeleton data for assigning a pose estimation result.

Further: in the step S5, the virtual simulation server realizes the binding of the athlete's three-dimensional skeleton data and role modeling, generates rendering model data, and transmits the rendering model data to the display query terminal, including the following steps:

S51, stadium modeling, constructing a virtual basketball court;

S52. Character modeling, constructing a virtual character, determining the key skeleton points of the virtual character, and placing the movement trajectory of the key skeleton points on the constructed virtual character;

S53. Binding the three-dimensional skeleton data of the athlete and the key skeleton points of the virtual character;

S54. Render the skeleton and skin of the avatar to form a rendered model; obtain an animation of the athlete's rendered model.

Further: in the step S54, the skeleton and skin of the avatar are rendered, wherein the face skin rendering of the avatar adopts facial information data collected by a facial recognition collection device.

Further: the S5 virtual simulation platform server also generates a training progress analysis report and an athlete's individual movement analysis report.

Further: the display query terminal also includes a mobile handheld terminal, and the handheld terminal is used for querying rendering model animations, training progress analysis reports, and athlete individual movement analysis reports.

Beneficial effects of the present invention:

1. The present invention generates a rendering model for query through the binding of the three-dimensional skeleton data of the athlete and the modeling of the virtual character; the athlete can look back at his own training actions from a three-dimensional angle, compare the standard actions, find insufficient actions, and reduce the intensity of repeated teaching by the coaches; You can visually check the focus point, take-off height, and take-off angle of your feet during training, find out the lack of coordination of body movements in ball-handling sports, find out movement defects, and take targeted measures to improve training efficiency.

2. By generating training progress analysis reports and athletes' individual movement analysis reports, it helps coaches to grasp the athletes' training situation and training progress, and adopts targeted training receipts, which will help improve the pertinence of basketball training and improve training efficiency.

3. Through the mobile handheld terminal, it assists the coaches in teaching, flexibly checks the training situation of the athletes, checks the exercise progress and exercise reports, and also facilitates the athletes to view the rendering model animation.

4. By rendering the virtual character's face skin and using the facial information data collected by the facial recognition acquisition device, the facial features of the virtual character correspond to the athletes. When the athletes watch the playback, they can feel their training process more vividly and intuitively, and can view themselves from a three-dimensional perspective Actions can make the restoration of actions more realistic.

Description of drawings

Fig. 1 is a kind of structural representation of the assistant system of the basketball practice based on AI technology of the present invention;

Fig. 2 adopts RMPE algorithm to generate two-dimensional skeleton data flowchart for the present invention;

Fig. 3 is that the present invention adopts VideoPose3D to generate three-dimensional skeleton data flowchart;

Fig. 4 is a flow chart of binding the athlete's three-dimensional skeleton data and the key skeleton points of the avatar according to the present invention.

100. AI data collection department; 110. Facial recognition collection equipment; 120. Sports video collection equipment;

200. Display query terminal; 210. Site display screen;

300. Analysis and processing unit; 310. AI recognition processing server; 320. Virtual simulation server; 330. Data storage server.

Detailed ways

Embodiments of the present invention are described in detail below, examples of which are shown in the drawings, wherein the same or similar symbols designate the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the figures are exemplary only for explaining the present invention and should not be understood as limiting the present invention.

As shown in Figures 1-4, the present invention discloses an auxiliary system for basketball practice based on AI technology, including an auxiliary system for basketball practice based on AI technology, which is characterized in that it includes an AI data collection unit set on a physical basketball court 100. Display the query terminal 200 and the analysis processing unit 300 arranged in the background; the AI data collection unit 100 includes a facial recognition collection device 110 and a motion video collection device 120 for collecting athlete facial information and motion video data; the analysis processing portion 300 includes AI recognition processing server 310, virtual simulation server 320 and data storage server 330; including AI recognition processing server 310 for analyzing and processing motion video data to generate three-dimensional skeleton data of athletes; virtual simulation server 320 for venue modeling and role modeling, The player's three-dimensional skeleton data and role modeling are bound and a rendering model is generated for query; the field display screen 210 is used to display the query rendering model.

An auxiliary method for basketball practice based on AI technology, comprising the following steps:

S1. The virtual simulation server 320 models the basketball venue and the player's role, realizing the binding of the digitally simulated basketball court and the real basketball court;

S2. The facial recognition collection device 110 collects athlete information, and establishes athlete training files;

S3. The sports video collection device 120 collects the athlete's sports video, and transmits it to the AI recognition processing server 310;

S4, the AI recognition processing server 310 recognizes and analyzes the sports video, and generates three-dimensional skeleton data of the athlete;

S5. The virtual simulation server 320 realizes the binding of the athlete's three-dimensional skeleton data and the character modeling, generates rendering model data, and transmits the rendering model data to the display query terminal 200;

S6. The display query terminal 200 displays and queries the dynamic rendering model.

The facial recognition collection device 110 is used to collect facial images of athletes, authenticate athlete information, and call database athlete information files. After authentication, athletes and coaches can view athlete training data, complete basic information collection and comparative analysis of players, and the files include multiple data , such as athlete name, date of birth, height and weight, pre-training movement analysis report, etc.

The sports video collection device 120 collects the sports video of the physical basketball court, and transmits it to the AI recognition processing server 310 in real time. The sports video collection device 120 adopts a high-definition camera, such as a Hikvision (DS-2CD5A24EFWD-IZS) camera, and the resolution of the collected video meets High-definition requirements, high-definition video collection can be carried out, and multi-dimensional collection and recording of players' movements, running distances, and position videos can be recorded, which is convenient for AI recognition processing server 310 to analyze players' movements, speeds, angles and other information to establish an accurate 3D skeleton Information, one or more cameras can be used.

The AI recognition processing server 310 is set in the backstage of the venue, and it uses wired network communication, wireless network communication or 5G communication module to connect with motion video collection equipment, facial recognition collection equipment 110, virtual simulation server 320 and storage server.

The AI recognition processing server 310 first establishes a data set, including basketball training standard videos, and the data set includes basic training actions of single and multiple people, such as: shooting, layup, dribbling in situ, running dribble, dunk, running without the ball Single-player actions such as dribble breakthrough and defense, etc., the data set is imported through an external device, and the existing data set can be used or self-made and imported as standard comparison data.

The AI recognition processing server 310 analyzes and processes the collected video, and extracts the athlete's two-dimensional skeleton data using a two-dimensional pose estimation algorithm. The two-dimensional pose estimation algorithm adopted can be OpenPose, RMPE and other algorithms.

Taking the RMPE algorithm as an example, the process is shown in Figure 2. First, use the YOLOv3 algorithm to normalize the input image, divide the image size into several grids, establish a detection frame, and detect the position of the human body in the detection frame.

Then the human body frame is input into the SymmetricSTN+SPPE module. The RMPE algorithm uses symmetric STN and parallel SPPE, where SPPE trains a single person image. Through parallel SPPE, the STN is trained to move the human body to the center of the extraction area, and then through SPPE for accurate Pose estimation, which automatically generates pose proposals; compares the pose estimation proposals with the actual poses of the dataset to generate pose estimation results.

The PoseNMS module is used to assign values to the pose estimation results to generate two-dimensional skeleton data.

The RMPE algorithm and the YOLOv3 algorithm in the RMPE algorithm, the SymmetricSTN+SPPE module and the PoseNMS module are all existing technologies, which can be queried through open literature.

To convert two-dimensional skeleton data into 3D skeleton pose data, the VideoPose3D algorithm can be used. As shown in Figure 3, the VideoPose3D algorithm uses the full-product network to process the key point sequence of the input two-dimensional skeleton data in the time domain to obtain a three-dimensional skeleton data.

Both the VideoPose3D algorithm and the full product network are existing technologies, which can be queried through open literature.

The three-dimensional skeleton data corresponds to each part of the athlete, such as: head, torso, upper arm, lower arm, thigh and calf, shoulder, elbow, wrist, knee, ankle, etc., and the three-dimensional skeleton data is combined with the key skeleton of the virtual character point to bind to realize the binding of virtual characters and athletes.

For the construction of virtual characters, PoseStudio software can be used to effectively fix the specific functions of the virtual character’s bone structure and some joints, etc., and then use Maya software to build a virtual basketball stadium, import the virtual character model generated by PoseStudio software, and adjust the virtual character. Technical action virtual simulation. Place the movement trajectory of the key bones on the virtual character constructed, and control the training process of the virtual athlete by controlling the movement trajectory of these key bones, so as to facilitate the grasp of the technical characteristics of the basketball player.

Using Maya software to build a virtual basketball stadium, the primary task is to restore the building to the greatest possible extent according to the scale of the stadium, reduce various invisible surfaces in the model, and whether the virtual character model can be accurately displayed in 3D animation It is the evaluation standard for judging the quality of venue model construction.

As shown in Figure 4, the Maya software is used to model the basketball player's task skeleton and skin, and a smooth virtual character model is obtained by using UV texture mapping, simplifying the number of model faces, and merging organizations. The key skeleton points of the avatar are bound to realize the binding between the avatar and the athlete.

Use Mava software to process the characters, environment and sound effects in the later stage, and get the 3D animation of the athletes' basketball training.

When using Maya software to model and render basketball characters, use the facial information data collected by the facial recognition collection device 110 to render the virtual character's face skin, so that the facial features of the virtual character correspond to the athletes, and the athletes can be more vivid and intuitive when watching the playback Feel your own training process, and you can view your own movements from a three-dimensional perspective, which can make the restoration of movements more realistic.

Further, the display query terminal 200 also includes a mobile handheld terminal. The handheld terminal 220 is connected to the data storage server 330 through the network. The handheld terminal 220 can be used for querying rendering model animations, training progress analysis reports, and athlete individual movement analysis reports, which is convenient to use.

The handheld terminal 220 can be used as a sports auxiliary device, and can use a mobile phone, a tablet computer or other portable electronic terminal equipment. The handheld terminal 220 can assist the coach to grasp the athlete's situation, and can check the exercise progress and exercise report, which is convenient for the athlete to view the rendering model animation.

Further, through the data analysis of the server of the virtual simulation platform, the athlete's exercise time, jump height, action angle, and the main difference points from the standard action can be obtained, and a sports report can be generated for athletes and coaches to view and analyze.

Sports reports include training progress analysis reports, athlete individual movement analysis reports and many other aspects. The following is an example of a training progress report:

The subjects of basketball training are 5.8mX5 turn-back running and full-court 3/4 acceleration running. On September 1, 2022, October 1, 2022, November 1, 2022 and December 1, 2022, the basketball During the training, 15 athletes were tested for speed, and the results are shown in Table 1.

Table 1 Speed training test results

From the sports report in Table 1, it can be known that before September 1, the average running time of the athletes’ 5.8mX5 turn-back runs and the 3/4 acceleration running of the whole field were 11.2s and 4.52s respectively, and the overall time was longer; Tests will be conducted on Sunday, November 1, and December 1 respectively. The test data and result data can be automatically generated. Athletes and coaches can grasp the progress of basketball training, which is more conducive to improving the pertinence of basketball training and taking training measures.

The above is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto, any person familiar with the technical field within the technical scope disclosed in the present invention, according to the technical solution of the present invention Any equivalent replacement or change of the inventive concepts thereof shall fall within the protection scope of the present invention.

Claims (8)

1. An auxiliary system for basketball practice based on AI technology, characterized in that: comprising an AI data acquisition unit (100) arranged on a physical basketball court, a display query terminal (200) and an analysis and processing unit (300) arranged at the background ; The AI data collection unit (100) includes a facial recognition collection device (110) and a sports video collection device (120), which are used to collect player facial information and sports video data; The analysis processing unit (300) includes an AI recognition processing server (310), a virtual simulation server (320) and a data storage server (330); The display query terminal (200) includes a site display screen (210); The AI recognition processing server (310) is used to analyze and process motion video data to generate three-dimensional skeleton data of athletes; The virtual simulation server (320) is used to bind and generate rendering models for venue modeling and role modeling, athlete three-dimensional skeleton data and role modeling for query; The field display screen (210) is used to display the query rendering model. 2. An auxiliary method for basketball practice based on AI technology, characterized in that: comprising the following steps: S1, the virtual simulation server (320) models the basketball stadium and the player's role, and realizes the binding of the digitally simulated basketball court and the real basketball court; S2. The facial recognition acquisition device (110) collects athlete information and establishes athlete training files; S3. The sports video collection device (120) collects the athlete's sports video, and transmits it to the AI recognition processing server (310); S4, AI recognition processing server (310) recognizes and analyzes motion video, generates three-dimensional skeleton data of athlete; S5, the virtual simulation server (320) realizes the binding of the athlete's three-dimensional skeleton data and role modeling, generates the rendering model data, and transmits the rendering model data to the data storage server (330) and the display query terminal (200); S6. The display query terminal (200) displays and queries the dynamic rendering model. 3. the auxiliary method of a kind of basketball practice based on AI technology according to claim 2, it is characterized in that: in described step S4, AI recognition processing server (310) recognizes and analyzes motion video, generates three-dimensional skeleton data of athlete, comprises The following steps: S41. Establishing a data set; S42. Extract two-dimensional skeleton data, compare the data sets, and obtain the athlete's two-dimensional posture coordinates; S43. Convert the two-dimensional attitude coordinates into three-dimensional attitude coordinates to obtain three-dimensional skeleton data. 4. the auxiliary method of a kind of basketball practice based on AI technology according to claim 3, is characterized in that: extracting two-dimensional skeleton data comprises the following steps in the described S42: S421. Normalize the input image, establish a detection frame, and detect the position of the human body in the detection frame; S422. Input the normalized image into the pose estimation module, generate a pose estimation suggestion, compare the pose estimation suggestion with the actual pose of the data set, and generate a pose estimation result; S423. Generate two-dimensional skeleton data for assigning a pose estimation result. 5. the auxiliary method of a kind of basketball practice based on AI technology according to claim 2, it is characterized in that: in described step S5, virtual simulation server (320) realizes the binding of athlete's three-dimensional skeleton data and character modeling, Generate rendering model data, and transmit the rendering model data to the display query terminal (200), including the following steps: S51, stadium modeling, constructing a virtual basketball court; S52. Character modeling, constructing a virtual character, determining the key skeleton points of the virtual character, and placing the movement trajectory of the key skeleton points on the constructed virtual character; S53. Binding the three-dimensional skeleton data of the athlete and the key skeleton points of the virtual character; S54. Render the skeleton and skin of the avatar to form a rendered model; obtain an animation of the athlete's rendered model. 6. A kind of auxiliary method for basketball practice based on AI technology according to claim 5, characterized in that: in said S54, the skeleton and skin of the virtual character are rendered, wherein the facial skin rendering of the virtual character adopts Facial information data collected by the facial recognition collection device (110). 7. The auxiliary method of basketball practice based on AI technology according to claim 2, characterized in that: the S5 virtual simulation platform server also generates training progress analysis reports and athlete individual action analysis reports. 8. A smart playground training method based on AI technology according to claim 2, characterized in that: the display query terminal (200) also includes a mobile handheld terminal, and the handheld terminal (220) is used to query the rendering model Animation, training progress analysis report and athlete individual movement analysis report.

Images