A badminton shuttlecock serving machine control method and system based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making.

Abstract

This invention provides a badminton serving machine control method and system based on visual perception, posture recognition, trajectory prediction, and adaptive decision-making, belonging to the field of serving control. It addresses the problems of limited serving patterns, poor equipment perception capabilities, and ineffective training due to a lack of human-like game strategy. This invention predicts the badminton flight trajectory and landing point based on a rigid body aerodynamics model; extracts skeletal joints based on a geometric constraint loss function, calculates the athlete's posture state, and constructs a dynamic defensive potential energy field representing the range of attention coverage; generates adaptive control decisions including the target landing point and serving pattern based on expert knowledge rule-driven and potential energy field extremum optimization methods; and drives the actuator to complete the serve through inverse kinematics. This invention can perceive the athlete and ball state in real time, simulate real-person game thinking, and adaptively serve to target the athlete's defensive weaknesses, significantly improving the intelligence level and practical effectiveness of badminton training.

Description

Technical Field

[0001] This invention relates to the field of badminton serving machine control technology, and more specifically, to a badminton serving machine control method and system based on visual perception, posture recognition, trajectory prediction, and adaptive decision-making. Background Technology

[0002] With the rapid development of artificial intelligence and machine vision technology, sports training is evolving towards "precision" and "intelligence." In the field of badminton, cutting-edge research has begun to explore how to endow machines with "eyes" and "brains," enabling them to observe, understand, and adapt to their environment. Badminton demands extremely high levels of reaction speed, footwork, and hitting skills from athletes. Traditional multi-ball training heavily relies on coaches' manual serves, and its intensity, consistency, and diversity of training modes are limited by the coach's physical strength and experience. Therefore, mechanical ball-serving machines have emerged, providing stable and repetitive serves and becoming an important tool for auxiliary training. However, most existing ball-serving machines are in the "open-loop" control stage, meaning they launch according to preset fixed programs (such as speed, angle, and frequency), unable to sense the trainee's real-time state and provide interactive responses. This results in a rigid training mode that struggles to simulate the rapidly changing trajectories of the ball in real matches.

[0003] As crucial equipment for automating athlete training, current technological development revolves around three core directions: vision-based athlete posture perception and analysis; trajectory tracking and landing point prediction for high-speed badminton shuttlecocks; and intelligent badminton robot systems integrating perception and control. However, while existing badminton serving machine technology has addressed the basic function of "serving" to some extent, significant technological limitations remain in terms of intelligence, anthropomorphism, and interactivity, making it difficult to meet the needs of advanced tactical training.

[0004] Currently, most badminton serving machines on the market use open-loop control. These devices primarily control the speed of the serving motor and the angle of the gimbal through a preset fixed program, thereby achieving fixed-point serves or simple random-landing serves. The first major drawback of existing technology is the lack of environmental perception and closed-loop feedback capabilities. It cannot sense whether the served ball flies along a predetermined trajectory, nor can it sense the state of the receiving player. The second major drawback of existing technology is the limited dimension and insufficient real-time capability of visual perception. In actual combat, the formulation of serving strategies often depends on the opponent's positioning, center of gravity movement, and racket holding posture. Existing technologies generally lack the ability to recognize the high-precision posture of the athlete's skeletal joints, and cannot analyze the athlete's movement intentions in real time. This causes the serving machine to mechanically execute the program, unable to dynamically adjust the serving strategy based on the student's positioning flaws or fatigue state like a human coach, lacking the intelligent game closed loop of "observation-judgment-decision". Furthermore, in terms of trajectory prediction and control decision-making, existing technologies mainly adopt simplified parabolic models, which fail to effectively solve the adaptive control problem in unstructured environments. When performing path planning, existing ball machine systems often ignore the nonlinear relationship between the aerodynamic drag coefficient and the angle of attack, resulting in extremely low accuracy in predicting landing points when simulating complex ball trajectories such as high clears and smashes.

[0005] To address the aforementioned technical challenges, this invention provides a control system and method that integrates visual perception, pose recognition, trajectory prediction, and adaptive decision-making. This enables the ball-launching machine not only to "see" the movement of the ball and the player, but also to "understand" the training scenario and make intelligent ball-returning decisions that simulate real competitive intentions. This achieves a fundamental leap from "fixed program launch" to "adaptive interactive training." Summary of the Invention

[0006] The technical problem to be solved by this invention is:

[0007] To address the problems of existing badminton serving equipment, such as its limited serving modes, poor equipment sensing capabilities, and lack of human-like game strategies leading to ineffective training.

[0008] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0009] This invention provides a badminton serve machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making, comprising the following steps:

[0010] S100: Model the key elements within the court; when the badminton serving machine is in standby mode, collect image data information through image acquisition devices deployed at the side of the court, establish a mapping relationship matrix between the world coordinate system and the camera coordinate system, divide the court into areas, and lock the target athletes and current environmental status within the court in real time.

[0011] S200. When the badminton serving machine is in the ready-to-return state, the spatial position of the badminton target is calculated, and combined with the visual information at continuous moments, the three-dimensional pose parameters of the badminton are estimated. The three-dimensional pose parameters include spatial coordinates, direction of motion, head orientation, and velocity information. Based on the rigid body aerodynamic model, the trajectory of the shuttlecock is calculated. Combined with the three-dimensional pose parameters of the badminton and its historical motion state data, its flight trajectory is predicted to obtain the corresponding flight path characteristics or expected landing area, and the serving machine is moved to the corresponding landing area.

[0012] S300: When the badminton serving machine is in the return judgment state, the acquired athlete image data is input into the pre-trained human posture estimation neural network to extract the skeletal joint data of the target athlete; based on the extracted skeletal joints, the athlete's motion parameters are calculated, including trunk posture calculation, left and right foot position topology and target movement direction, so as to establish a dynamic human state model that can reflect the athlete's attention distribution, reaction area and defensive tendency, which is used to determine the athlete's attention concentration area.

[0013] S400. When the badminton serving machine is in the return decision state, based on the trajectory prediction result of step S200, and combined with the preset training objectives, athlete posture state or training strategy requirements, the current training state is analyzed, a set of decision parameters for serving control is constructed, and a return ball path decision-maker is established.

[0014] S500: When the badminton serving machine is in the return execution state, the adaptive serving control decision is converted into a specific control command. The preset ball path is subjected to inverse kinematic solution to calculate the control parameters of the serving machine and send them to the execution mechanism of the badminton serving machine to control the serving mechanism to complete the corresponding serving action. A closed-loop optimization mechanism is formed through real-time feedback.

[0015] Further, in step S100, the camera coordinate system is transformed to the world coordinate system:

[0016]

[0017] in, The transformation moments from the camera coordinate system to the world coordinate system are... For rotation matrix, The translation vector that describes the camera position. and These represent the point coordinates in camera coordinates and world coordinates, respectively.

[0018] Furthermore, in step S200, based on aerodynamic theory, the shuttlecock is defined as a six-degree-of-freedom rigid body, subjected to gravity. air resistance and air lift The effects of the three vector forces are represented as follows:

[0019]

[0020] in, Let g be the total mass of the shuttlecock, and g be the acceleration due to gravity. Let be the instantaneous velocity vector of the badminton shuttlecock, with a direction angle of . , air density, The feature projection area of ​​a badminton shuttlecock. The attitude angle is the axis of symmetry of the badminton shuttlecock. The angle of attack is defined as the angle between the velocity direction and the ball axis. , The coefficient of air resistance experienced by the shuttlecock is denoted by , which represents the angle of attack. With Reynolds number The function, The coefficient of air lift applied to the shuttlecock; The coefficient of lift is the air pressure on the shuttlecock, and the angle of attack is... The function, A unit vector perpendicular to the direction of velocity;

[0021] Therefore, the complete equation of motion for the center of mass is obtained as follows:

[0022]

[0023] When attack angle At this time, aerodynamics will generate a restoring torque. It attempts to push the ball forward; the magnitude of the restoring torque depends on the center of mass. To the pressure center Lever arm length According to Euler's formula, the equation of motion for a rigid body rotation is:

[0024]

[0025] in, For rotational inertia, To restore the torque coefficient, is the rotational damping coefficient.

[0026] Further, in step S300, the following are included:

[0027] S310. Skeletal joint extraction and coordinate mapping: Utilizing a lightweight convolutional neural network to regress each skeletal joint of the human skeleton from an image. Heat map During network training, the mean squared error loss function is used to compare the predicted heatmap with the actual heatmap; to address the limb distortion problem that occurs in badminton movements, a limb geometric constraint loss function L is introduced. total for:

[0028]

[0029] in, For the first Predicted heatmap of individual skeletal joints This is a truth heatmap with manual annotations, where N is the total number of skeletal joints; For the network prediction of the first Euclidean distance of the root limb, The prior value of the athlete's anatomical standard limb length. To constrain the weights, M represents the total number of limbs;

[0030] Using the transformation matrix from the camera coordinate system to the world coordinate system established in step S100, the pixel coordinates are mapped back to the world coordinate system.

[0031] S320, Calculation of motion state, including trunk posture calculation, left and right foot position topology and target motion direction determination;

[0032] S330, Determination of Athlete's Area of ​​Concentration: The area of ​​concentration is modeled as a projection of the athlete's center of mass. The region is a Gaussian elliptical distribution centered on the athlete; the shape and offset of this region are influenced by the athlete's velocity vector. and torso normal vector Co-modulation; the attention center is defined as , It is a time constant;

[0033] Therefore, site coordinates Attention intensity at the location for:

[0034]

[0035] in, torso normal vector The determined rotation angle, Extending along the direction of movement, it represents the athlete's extreme attention level in the current direction of movement. Perpendicular to the direction of movement, representing lateral interception capability; x c and y c These are the horizontal and vertical coordinates of the athlete's center of mass projected onto the ground.

[0036] Further, in step S320, the following is included:

[0037] S321, Trunk pose calculation based on extraction A set of skeletal joints Construct a local torso plane, with the left shoulder as an example. right shoulder Left hip Right hip Given the three-dimensional coordinates, then the shoulder vector Spinal vector and torso normal vector for:

[0038]

[0039] If the torso normal vector If the direction is towards the net, the athlete is in a head-on confrontation state; if the direction is to the side, the athlete is in a sideways stance, preparing for a powerful attack.

[0040] S322, Topology of left and right foot positions, given the left foot Right foot The three-dimensional coordinates are defined by the vector connecting the two feet. Calculate the relationship between this vector and the normal vector of the net plane. The included angle :

[0041]

[0042] when When the feet are parallel, the athlete is considered to be in a parallel defensive stance, and it is assumed that their attention is focused on receiving the smash; when At that time, the athlete was judged to be in a straddle position, and it was believed that his attention was focused on the initiation of forward and backward movement;

[0043] S323. Determining the direction of target movement and the athlete's speed. Achieved through motion estimation between adjacent frames:

[0044]

[0045] in, It is the change in the three-dimensional position coordinates of the athlete's center of mass. It is the time interval between frames.

[0046] Further, in step S400, the following is included:

[0047] S410. Generate benchmark tactical parameters based on expert knowledge rules. Pre-set expert tactical knowledge base or strategy comparison table. Use the currently acquired training objectives, athlete position and status categories and preset training strategy requirements as multi-dimensional index inputs. Through table lookup matching or rule reasoning, calculate the corresponding serving pattern, serving intensity level and basic ball path type to form a subset of expert decision parameters.

[0048] S420: Determine the optimal tactical landing point based on potential energy field extremum optimization. Based on the dynamic defensive potential energy field, perform numerical analysis and gradient search on the potential energy distribution data within the field area to identify potential energy values. The local minimum points or low potential energy areas with values ​​below the preset safety threshold are identified as the athlete's current defensive weak point or blind spot, and the geometric center or extreme point coordinates of the area are locked as the target landing point parameters for the return ball.

[0049] A badminton serving machine control system based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making. The system has program modules corresponding to the above steps, and executes the steps in the above-mentioned badminton serving machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making during operation.

[0050] A computer-readable storage medium storing a computer program configured to, when invoked by a processor, implement steps of a badminton serving machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making.

[0051] Compared with the prior art, the beneficial effects of the present invention are:

[0052] ① Intelligent Decision-Making Capability Based on Human Posture Perception: This invention inputs acquired athlete image data into a pre-trained human posture estimation network, enabling real-time extraction of multi-joint skeletal joint information of the athlete. Based on the spatial relationships and temporal changes between key points, it calculates the athlete's facing direction, movement direction, and left and right foot positioning parameters. This method allows the ball-serving machine to proactively perceive the athlete's return preparation state and area of ​​focus, significantly improving the intelligence and targeting of serve decisions compared to traditional control methods that rely on fixed parameters or single positional information.

[0053] ② Adaptive Serving Mechanism Integrating Posture Perception and Trajectory Prediction: This invention combines human posture analysis results with badminton flight trajectory prediction. Through comprehensive analysis of the athlete's dynamic behavioral characteristics and the trajectory of the incoming shuttlecock, it achieves real-time adjustment of the serving strategy. This invention can adaptively adjust the serving direction, landing area, and serving pattern based on changes in the athlete's stance, movement trends, and focus of attention, thereby improving the interactivity and effectiveness of the training process and enhancing the training system's adaptability to athletes of different skill levels.

[0054] ③ Robust visual perception capability for high-speed, small-target scenarios: Addressing the characteristics of badminton shuttlecocks—small size, high flight speed, and complex trajectory—this invention improves stability and reliability in high-speed dynamic training scenarios by combining continuous visual perception with temporal attitude analysis. Through comprehensive processing of multi-frame attitude information, the impact of single-frame recognition errors, short-term occlusion, and noise interference on system judgment results is effectively reduced, thus improving overall control accuracy.

[0055] ④ Non-contact human-computer interaction and engineering feasibility: This invention achieves athlete status analysis based on visual perception, eliminating the need for athletes to wear additional sensors or markers, thus avoiding interference with the training process and improving user comfort and naturalness. Furthermore, the system has a clear structure and well-defined implementation path, demonstrating good engineering feasibility and potential for widespread application.

[0056] In summary, this invention, by constructing an adaptive decision-making mechanism that combines visual perception-based human pose recognition with trajectory prediction, enables the badminton serving machine to understand and respond to the dynamic state of athletes, significantly improving the overall performance of the system in terms of intelligence level, training targeting, interactive capabilities, and adaptability to complex scenarios. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of a badminton serving machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making in an embodiment of the present invention.

[0058] Figure 2 This is a schematic diagram of a badminton serving machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making in an embodiment of the present invention.

[0059] Figure 3 This is a flowchart of a badminton serving machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making in an embodiment of the present invention.

[0060] Figure 4 This is a flowchart illustrating the pre-return ball state in an embodiment of the present invention;

[0061] Figure 5 This is a flowchart illustrating the process of determining the return ball status in an embodiment of the present invention. Detailed Implementation

[0062] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0063] Specific Implementation Plan 1: Combining Figures 1 to 5 As shown, this invention provides a badminton serve machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making, comprising the following steps:

[0064] S100, the environment initialization stage, models the key elements in the field; when the badminton serving machine is in standby mode, image data information is collected by the image acquisition device deployed on the side of the court, and a mapping relationship matrix between the world coordinate system and the camera coordinate system is established to lock the target athlete and the current environmental status in the field in real time.

[0065] The image acquisition device is a binocular camera or a multi-view camera. During the initialization phase, it acquires the camera's intrinsic and extrinsic parameter matrices, including focal length, principal point coordinates, distortion parameters, and the camera's pose in the world coordinate system. By solving the acquired data from the checkerboard or calibration board, a reliable coordinate transformation matrix is ​​obtained, that is, the camera coordinate system is transformed to the world coordinate system.

[0066]

[0067] in, The transformation moments from the camera coordinate system to the world coordinate system are... For rotation matrix, The translation vector that describes the camera position. and These represent the point coordinates in camera coordinates and world coordinates, respectively.

[0068] By dividing the training field in the world coordinate system into grids, the system constructs multiple functional areas, including the front court area, the back court area, and the sideline area, and marks the current position of the ball machine and the standing area of ​​the athletes, providing scene structure constraints for subsequent return decisions;

[0069] S200. When the badminton serving machine is in the ready-to-return state, the spatial position of the badminton target is calculated, and combined with the visual information at continuous moments, the three-dimensional pose parameters of the badminton are estimated. The pose parameters include spatial coordinates, direction of movement, head orientation, and velocity information. Based on the rigid body aerodynamic model, the trajectory of the shuttlecock is calculated. Combined with the badminton pose information and its historical motion state data, its flight trajectory in the future period is predicted to obtain the corresponding flight path characteristics or expected landing area, and the serving machine is moved to the corresponding landing area.

[0070] The flight motion of a badminton shuttlecock is a strong drag motion under a non-conservative force field, characterized by a gentle ascent and a steep, almost vertical descent. Based on the shuttlecock's shape—its 16 feathers are inserted at a specific angle, creating a turbine effect as airflow passes over them—and the high-speed rotating rigid body's fixed-axis property, the shuttlecock maintains a forward-facing posture even at low speeds near its highest point, preventing it from drifting aimlessly. This analysis suggests that the shuttlecock's motion can be fixed within a plane. The shuttlecock possesses extremely light mass, a high drag coefficient, a significantly asymmetrical structure, and unique self-stabilizing flipping characteristics. Its flight attitude and trajectory are often influenced by a combination of factors, including air resistance, lift, attitude recovery torque, and velocity decay. Therefore, traditional parabolic prediction methods cannot meet the high-precision requirements of real-time training scenarios. This invention achieves stable prediction of the shuttlecock's future flight path in complex training scenarios by introducing three-dimensional pose estimation, multi-frame velocity fusion, and a simplified aerodynamic model. Specifically, it includes the following steps:

[0071] In the course of badminton, it is defined as a six-degree-of-freedom rigid body. According to aerodynamic theory, it is subjected to gravity during its movement in the air. ), air resistance ( (opposite to the velocity vector) and air lift ( The effects of the three vector forces (perpendicular to the velocity vector) are expressed as follows:

[0072]

[0073] in, Let g be the total mass of the shuttlecock, and g be the acceleration due to gravity. Let be the instantaneous velocity vector of the badminton shuttlecock, with a direction angle of . , air density, The feature projection area of ​​a badminton shuttlecock depends on its shape. The attitude angle is the axis of symmetry of the badminton shuttlecock. The angle of attack is defined as the angle between the velocity direction and the ball axis. , The coefficient of air resistance experienced by the shuttlecock is denoted by , which represents the angle of attack. With Reynolds number The function, The coefficient of lift is the air pressure on the shuttlecock, and the angle of attack is... The function, A unit vector perpendicular to the direction of velocity;

[0074] Therefore, the complete equation of motion for the center of mass is obtained as follows:

[0075]

[0076] When attack angle At this time, aerodynamics will generate a restoring torque ( ), attempting to push the ball forward; the magnitude of the restoring torque depends on the center of mass ( ) to the center of pressure ( ) lever arm length Therefore, the equation of motion for a rigid body can be obtained from Euler's formula:

[0077]

[0078] in, For rotational inertia, To restore the torque coefficient, The rotational damping coefficient is used to prevent the shuttlecock from swinging indefinitely;

[0079] In the trajectory prediction stage, this invention constructs the flight trend based on the position and attitude sequences of the badminton shuttlecock in the past several frames. It eliminates the trajectory deviation caused by single-frame noise through a data-driven multi-frame prediction model, making the prediction results closer to the real flight state. On this basis, this invention uses the constraints of the court boundary, net height and current service task to perform a second verification of the prediction results, ensuring that the predicted landing area is physically reachable and conforms to the training logic.

[0080] S300, combined Figure 4As shown, when the badminton serving machine is in the return judgment state, the acquired athlete image data is input into a pre-trained human posture estimation neural network to extract the skeletal joint data of the target athlete. The skeletal joints include the shoulder, elbow, wrist, hip, knee, and ankle. Based on the extracted skeletal joints, the athlete's torso posture, left and right foot stance topology, and target movement direction are calculated to establish a dynamic human state model that reflects the athlete's attention distribution, reaction area, and defensive tendency, which is used to determine the athlete's attention concentration area. This process is not only used to determine whether the athlete is currently in a state of preparing to receive the ball, but also to provide constraints for subsequent adaptive serving strategies, enabling the serving machine to make intelligent adjustments according to the athlete's response ability, reaction direction, and stance, thereby achieving a more realistic competitive training experience.

[0081] Specifically, including,

[0082] S310, Skeletal Joint Extraction and Coordinate Mapping: When inputting the acquired images into the pre-trained network, not only single-frame images are input, but temporal information is also introduced. This utilizes a lightweight convolutional neural network (i.e., a human pose estimation neural network, such as an improved version). Regress each joint of the human skeleton from the image. Heat map When training the network, the mean squared error loss function is usually used to compare the predicted heatmap with the actual heatmap. To address the limb distortion problem caused by "large strides" or "backhand shots" commonly seen in badminton, a limb geometric constraint loss function is introduced, defining the loss function L. total for:

[0083]

[0084] The first term is the location regression loss. For the first Predicted heatmap of individual skeletal joints The first term is a manually annotated ground truth heatmap, where N is the total number of skeletal joints; the second term represents the limb geometric constraint loss. For the network prediction of the first Euclidean distance of the root limb This is the prior value of the athlete's anatomically standard limb length. To constrain the weights, M represents the total number of limbs;

[0085] Using the transformation matrix from the camera coordinate system to the world coordinate system established in step S100, the pixel coordinates are mapped back to the world coordinate system.

[0086] S320, Motion state calculation, including trunk posture calculation, left and right foot position topology, and target motion direction determination; specifically including,

[0087] S321, Trunk pose calculation based on extraction A set of skeletal joints The system constructs a normal vector of the torso plane through skeletal joints to characterize the athlete's torso orientation in order to better judge the likelihood of him exerting force.

[0088] Construct a local torso plane, with the left shoulder as an example. right shoulder Left hip Right hip Since the three-dimensional coordinates are known, the shoulder vector can be represented. Spinal vector and torso normal vector for:

[0089]

[0090] If the torso normal vector If the direction is towards the net, the athlete is in a head-on confrontation state; if the direction is to the side, the athlete is in a sideways stance, preparing for a powerful attack.

[0091] S322, Topology of left and right foot positions, given the left foot Right foot The three-dimensional coordinates are defined by the vector connecting the two feet. Calculate the relationship between this vector and the normal vector of the net plane. The included angle :

[0092]

[0093] when At that time, it was assumed that the athlete's feet were basically parallel, indicating a parallel defensive stance, and that their attention was focused on receiving the smash; when At that time, the athlete was judged to be in a straddle position, and it was believed that his attention was focused on the initiation of forward and backward movement;

[0094] S323. Determining the direction of target movement and the athlete's speed. This can be achieved through motion estimation between adjacent frames:

[0095]

[0096] in, It is the change in the three-dimensional position coordinates of the athlete's center of mass. It is the time interval between frames;

[0097] S330. Athlete's attention concentration area determination: Based on the parameters calculated above, a probability distribution model is constructed with the athlete's center of mass as the center, offset towards the direction of the movement trend, and weighted by the direction of the face. A dynamic defensive potential energy field is defined to characterize the athlete's current attention coverage area; the attention area is modeled as a projection of the athlete's center of mass. The region is a Gaussian elliptical distribution centered on the athlete; the shape and offset of this region are influenced by the athlete's velocity vector. and torso normal vector Co-modulation; the attention center is defined as , It is a time constant; therefore, the site coordinates Attention intensity at the location for:

[0098]

[0099] in, For the trunk normal vector The determined rotation angle, Extending along the direction of movement, it represents the athlete's extreme attention level in the current direction of movement. Perpendicular to the direction of movement, representing lateral interception capability; x c and y c These are the horizontal and vertical coordinates of the athlete's center of mass projected onto the ground;

[0100] S400, combined Figure 2 As shown, when the badminton serving machine is in the return decision state, based on the trajectory prediction results and combined with the preset training objectives, athlete posture, or training strategy requirements, the current training state is analyzed to construct a set of decision parameters for serve control and establish a return trajectory decision maker. The decision parameters are used to characterize the control requirements of the return, which include the target landing point, serve direction, serve intensity, serve mode, or preset trajectory. This generates corresponding serve control decisions, which can be adjusted according to dynamic changes during the training process, thereby achieving adaptive serve control.

[0101] Specifically, including,

[0102] S410. Based on expert knowledge rules, the system generates baseline tactical parameters. The system has a pre-set expert tactical knowledge base or strategy comparison table. The currently acquired training objectives (such as fixed-point practice and random confrontation), athlete position and state categories (such as forward leaning preparation, side-stepping preparation, and backward adjustment), and preset training strategy requirements (such as speed training, footwork training, and comprehensive confrontation training) are used as multi-dimensional index inputs. Through table lookup matching or rule reasoning, the corresponding serving patterns (including high clear, smash, drop shot, or flat drive), serving intensity level, and basic ball path type are calculated, forming a subset of expert decision parameters.

[0103] S420: Based on the optimization of the potential energy field extreme value, the optimal tactical landing point is determined. Based on the dynamic defensive potential energy field generated in the previous steps, the decision-maker performs numerical analysis and gradient search on the potential energy distribution data within the field area to identify the potential energy value. The presence of local minima or low-potential areas with values ​​below a preset safety threshold indicates that the athlete's attention coverage in that area is weak and their reaction time is longer. This area is identified as the athlete's current defensive weakness or blind spot, and the geometric center or extreme point coordinates of this area are locked as the target landing point parameters for the return ball to improve the targeting of training.

[0104] Random perturbations can also be constructed around low potential energy areas to simulate sudden changes in direction and rhythm in real matches, allowing athletes to constantly adapt to different ball conditions.

[0105] Based on the above information, the system constructs a multi-dimensional parameter space including landing point coordinates, ball speed, ball release angle, and spin parameters, and optimizes the parameter space using heuristic search or fast gradient evaluation methods to generate a set of optimal serving parameters. During the generation process, the system also considers the current motion state of the ball machine, such as the spin angle, base position, and maximum output range of the power mechanism, to ensure that the generated serving parameters can be accurately executed by the serving execution mechanism. In addition, this invention also supports dynamic updating of the strategy library based on the landing point distribution and success rate of the athlete's return ball, realizing continuous learning and training intensity adjustment, making the training process more intelligent, progressive, and personalized.

[0106] S500, combined Figure 3 As shown, when the badminton serving machine is in the return execution state, the decision-maker converts the adaptive serving control decision into specific control commands, performs inverse kinematics solution on the preset ball path, calculates the control parameters of the serving machine and sends them to the execution mechanism of the badminton serving machine to control the serving mechanism to complete the corresponding serving action, and forms a closed-loop optimization mechanism through real-time feedback;

[0107] The decision-maker integrates a subset of expert decision parameters with target landing parameters, and constructs a complete set of serve control decision parameters through rigid body dynamics numerical inversion calculation. It generates adaptive serve control commands that include specific serve direction, initial velocity vector and landing coordinates, thereby enabling dynamic training to exploit weaknesses in the athlete's real-time defensive state or specific tactical routines.

[0108] After receiving control parameters, the actuator of the badminton serving machine adjusts the serving action through servo drive, motor speed regulation or ejection mechanism. After the serve is completed, the present invention continues to monitor the actual flight path of the badminton shuttlecock using a visual sensor, and compares the predicted trajectory with the actual trajectory to determine the error of the strategy control.

[0109] Specific Implementation Scheme 2: The present invention provides a badminton serving machine control system based on visual perception, pose recognition, trajectory prediction and adaptive decision-making. The system has a program module corresponding to the above steps, and executes the steps in the above-mentioned badminton serving machine control method based on visual perception, pose recognition, trajectory prediction and adaptive decision-making during operation.

[0110] The other combinations and connections in this implementation scheme are the same as in Specific Implementation Scheme 1.

[0111] Specific Implementation Scheme 3: The present invention provides a computer-readable storage medium storing a computer program configured to implement, when called by a processor, the steps of a badminton serving machine control method based on visual perception for pose recognition, trajectory prediction, and adaptive decision-making.

[0112] The other combinations and connections in this implementation scheme are the same as in Specific Implementation Scheme 1.

[0113] It should be noted that the badminton serving machine described in this invention refers to an intelligent serving and returning device that integrates environmental perception, intelligent decision-making, and execution control. It includes at least: a multimodal sensor module for acquiring information about the badminton shuttlecock and the opponent's status within the court. This sensor module includes, but is not limited to, visual sensors, depth sensors, or other position detection units, used to identify the spatial position, velocity vector, and posture parameters of the badminton shuttlecock in real time, as well as the opponent's stance, posture, and movement trend; a high-performance computing and control unit electrically connected to the sensor module, which possesses real-time data processing and rapid decision-making capabilities, performing tasks including target detection, three-dimensional pose estimation, trajectory prediction, and serving strategy generation; and a mechanical execution mechanism connected to the computing and control unit. This mechanical execution mechanism includes a multi-degree-of-freedom motion structure for adjusting the launch direction and posture, and a power output component for achieving precise badminton shuttlecock launch. It can precisely control the launch angle, initial velocity, and launch timing according to control commands, thereby achieving high-precision directional launches to the target area. Through the aforementioned closed-loop structure of perception-computation-execution, the badminton serving machine (badminton robot) is equipped with real-time response capabilities to dynamic confrontation scenarios and human-like tactical execution capabilities.

[0114] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A badminton shuttlecock serving machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making, characterized in that, Includes the following steps: S100: Model the key elements within the court; when the badminton serving machine is in standby mode, collect image data information through image acquisition devices deployed at the side of the court, establish a mapping relationship matrix between the world coordinate system and the camera coordinate system, divide the court into areas, and lock the target athletes and current environmental status within the court in real time. S200. When the badminton serving machine is in the ready-to-return state, the spatial position of the badminton target is calculated, and combined with the visual information at continuous moments, the three-dimensional pose parameters of the badminton are estimated. The three-dimensional pose parameters include spatial coordinates, direction of motion, head orientation, and velocity information. Based on the rigid body aerodynamic model, the trajectory of the shuttlecock is calculated. Combined with the three-dimensional pose parameters of the badminton and its historical motion state data, its flight trajectory is predicted to obtain the corresponding flight path characteristics or expected landing area, and the serving machine is moved to the corresponding landing area. S300: When the badminton serving machine is in the return judgment state, the acquired athlete image data is input into the pre-trained human posture estimation neural network to extract the skeletal joint data of the target athlete; based on the extracted skeletal joints, the athlete's motion parameters are calculated, including trunk posture calculation, left and right foot position topology and target movement direction, so as to establish a dynamic human state model that can reflect the athlete's attention distribution, reaction area and defensive tendency, which is used to determine the athlete's attention concentration area. S400. When the badminton serving machine is in the return decision state, based on the trajectory prediction result of step S200, and combined with the preset training objectives, athlete posture state or training strategy requirements, the current training state is analyzed, a set of decision parameters for serving control is constructed, and a return ball path decision-maker is established. S500: When the badminton serving machine is in the return execution state, the adaptive serving control decision is converted into a specific control command. The preset ball path is subjected to inverse kinematic solution to calculate the control parameters of the serving machine and send them to the execution mechanism of the badminton serving machine to control the serving mechanism to complete the corresponding serving action. A closed-loop optimization mechanism is formed through real-time feedback.

2. The badminton serving machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making according to claim 1, characterized in that: In step S100, the camera coordinate system is transformed to the world coordinate system: in, The transformation moments from the camera coordinate system to the world coordinate system are... For rotation matrix, The translation vector that describes the camera position. and These represent the point coordinates in camera coordinates and world coordinates, respectively.

3. The badminton shuttlecock serving machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making according to claim 2, characterized in that: In step S200, based on aerodynamic theory, the shuttlecock is defined as a six-degree-of-freedom rigid body, subjected to gravity. air resistance and air lift The effects of the three vector forces are represented as follows: in, Let g be the total mass of the shuttlecock, and g be the acceleration due to gravity. Let be the instantaneous velocity vector of the badminton shuttlecock, with a direction angle of . , air density, The feature projection area of ​​a badminton shuttlecock. The attitude angle is the axis of symmetry of the badminton shuttlecock. The angle of attack is defined as the angle between the velocity direction and the ball axis. , The coefficient of air resistance experienced by the shuttlecock is denoted by , which represents the angle of attack. With Reynolds number The function, The coefficient of lift is the air pressure on the shuttlecock, and the angle of attack is... The function, A unit vector perpendicular to the direction of velocity; Therefore, the complete equation of motion for the center of mass is obtained as follows: When attack angle At this time, aerodynamics will generate a restoring torque. It attempts to push the ball forward; the magnitude of the restoring torque depends on the center of mass. To the pressure center Lever arm length According to Euler's formula, the equation of motion for a rigid body rotation is: in, For rotational inertia, To restore the torque coefficient, is the rotational damping coefficient.

4. The badminton serving machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making according to claim 3, characterized in that: Step S300 includes, S310. Skeletal joint extraction and coordinate mapping: Utilizing a lightweight convolutional neural network to regress each skeletal joint of the human skeleton from an image. Heat map During network training, the mean squared error loss function is used to compare the predicted heatmap with the actual heatmap; to address the limb distortion problem that occurs in badminton movements, a limb geometric constraint loss function L is introduced. total for: in, For the first Predicted heatmap of individual skeletal joints This is a truth heatmap with manual annotations, where N is the total number of skeletal joints; For the network prediction of the first Euclidean distance of the root limb, This is the prior value for the athlete's anatomically standard limb length. To constrain the weights, M represents the total number of limbs; Using the transformation matrix from the camera coordinate system to the world coordinate system established in step S100, the pixel coordinates are mapped back to the world coordinate system. S320, Calculation of motion state, including trunk posture calculation, left and right foot position topology and target motion direction determination; S330, Determination of Athlete's Area of ​​Concentration: The area of ​​concentration is modeled as a projection of the athlete's center of mass. The region is a Gaussian elliptical distribution centered on the athlete; the shape and offset of this region are influenced by the athlete's velocity vector. and torso normal vector Co-modulation; the attention center is defined as , It is a time constant; Therefore, site coordinates Attention intensity at the location for: in, For the trunk normal vector The determined rotation angle, Extending along the direction of movement, it represents the athlete's extreme attention level in the current direction of movement. Perpendicular to the direction of movement, representing lateral interception capability; x c and y c These are the horizontal and vertical coordinates of the athlete's center of mass projected onto the ground.

5. The badminton serving machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making according to claim 4, characterized in that: Step S320 includes, S321, Trunk pose calculation based on extraction A set of skeletal joints Construct a local torso plane, with the left shoulder as an example. right shoulder Left hip Right hip Given the three-dimensional coordinates, then the shoulder vector Spinal vector and torso normal vector for: If the torso normal vector If the direction is towards the net, the athlete is in a head-on confrontation state; if the direction is to the side, the athlete is in a sideways stance, preparing for a powerful attack. S322, Topology of left and right foot positions, given the left foot Right foot The three-dimensional coordinates are defined by the vector connecting the two feet. Calculate the relationship between this vector and the normal vector of the net plane. The included angle : when When the feet are parallel, the athlete is considered to be in a parallel defensive stance, and it is assumed that their attention is focused on receiving the smash; when At that time, the athlete was judged to be in a straddle position, and it was believed that his attention was focused on the initiation of forward and backward movement; S323. Determining the direction of target movement and the athlete's speed. Achieved through motion estimation between adjacent frames: in, It is the change in the three-dimensional position coordinates of the athlete's center of mass. It is the time interval between frames.

6. The badminton shuttlecock serving machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making according to claim 5, characterized in that: In step S400, the following are included: S410. Generate benchmark tactical parameters based on expert knowledge rules. Pre-set expert tactical knowledge base or strategy comparison table. Use the currently acquired training objectives, athlete position and status categories and preset training strategy requirements as multi-dimensional index inputs. Through table lookup matching or rule reasoning, calculate the corresponding serving pattern, serving intensity level and basic ball path type to form a subset of expert decision parameters. S420: Determine the optimal tactical landing point based on potential energy field extremum optimization. Based on the dynamic defensive potential energy field, perform numerical analysis and gradient search on the potential energy distribution data within the field area to identify potential energy values. The local minimum points or low potential energy areas with values ​​below the preset safety threshold are identified as the athlete's current defensive weak point or blind spot, and the geometric center or extreme point coordinates of the area are locked as the target landing point parameters for the return ball.

7. A badminton shuttlecock serving machine control system based on visual perception for pose recognition, trajectory prediction, and adaptive decision-making, characterized in that: The system has a program module corresponding to the steps of any one of the claims 1-6 above, and executes the steps in the above-described badminton serving machine control method based on visual perception, pose recognition, trajectory prediction and adaptive decision-making when running.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program configured to, when invoked by a processor, implement the steps of the badminton serving machine control method based on visual perception, pose recognition, trajectory prediction, and adaptive decision-making as described in any one of claims 1-6.

Images