A Ping Pong Intelligent Learning Method Based on Multi-Source Sensor Fusion
By using multi-source sensor fusion technology and combining it with a multi-rigid-body dynamics model of the human body, the problem of quantitative correlation between plantar pressure and joint movement in table tennis motion analysis was solved. This enabled physical modeling of the complete mechanical transmission path from plantar force to racket and quantitative evaluation of the kinetic chain, improving training efficiency and motion quality diagnosis.
Patent Information
- Application Number
- CN202611246979.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-17
- Publication Date
- 2026-11-10
AI Technical Summary
In existing table tennis motion analysis technologies, plantar pressure analysis and motion capture analysis are independent of each other and lack quantitative correlation at the level of mechanical transmission. It is impossible to perform physical modeling and quantitative evaluation of the complete chain from plantar force to racket impact.
Data is collected by foot pressure sensor array, depth camera and racket built-in inertial sensor unit. Combined with human multi-rigid-body dynamics model, the net torque and instantaneous power of joints are calculated step by step to establish the causal mapping relationship between foot force and joint movement, so as to realize the quantitative evaluation and correction of the kinetic chain.
It achieves physical modeling of the complete force transmission path from the foot to the racket, provides quantitative assessment of kinetic chain transmission efficiency and automatic identification of the dominant force-generating joints, and improves training efficiency and accurate diagnosis of movement quality.
Smart Images

Figure CN123069265A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of sports training analysis technology, and in particular to a table tennis intelligent learning method based on multi-source sensor fusion. Background Technology
[0002] The execution quality of table tennis techniques essentially depends on the spatiotemporal coordination of various parts of the body, with the coupling relationship between foot power generation and joint movement being particularly crucial. Taking a forehand topspin as an example, a standard kinetic chain requires power to originate from the foot pushing off the ground, progressing through the ankle, knee, and hip, and finally reaching the racket via trunk rotation, which in turn drives the shoulder, elbow, and wrist. During this process, the forward and backward movement trajectory of the foot's center of pressure directly reflects the accuracy of weight transfer. The loading rate of different areas of the foot determines the timing of subsequent joint power generation. If the arrival time of the foot's center of pressure at a specific position is not synchronized with the initiation of hip joint rotation, power leakage will inevitably occur during transmission. In other words, foot power generation and joint movement are not two independent processes, but rather the beginning and end links of the same biomechanical chain. Their precise alignment on the timeline and the progressive mechanical transmission together determine the final quality of the shot. Therefore, to achieve a scientific and quantitative evaluation of table tennis movements, it is essential to simultaneously acquire information on foot power generation and joint movement, and establish a physical connection between them at the level of mechanical transmission.
[0003] Existing table tennis motion analysis technologies have developed independently in two directions: foot pressure detection and motion capture. In foot pressure detection, current technologies use thin-film pressure sensor arrays embedded in insoles or mats to collect dynamic pressure distribution across different areas of the athlete's foot, calculating the trajectory of the pressure center and the force ratio between the left and right feet to assess the athlete's footwork and center of gravity stability during the stroke. However, this approach is limited to analyzing the sole surface of the foot, outputting only low-level physical quantities such as pressure distribution maps and center of gravity sway trajectories. It cannot correlate foot force information with the motion state of the upper limb joints, nor can it answer the question of whether the foot force is effectively transmitted to the racket. In motion capture, existing solutions mainly fall into two categories: one based on multi-view high-speed cameras or depth cameras, using computer vision algorithms to reconstruct human posture and output the three-dimensional angles and motion trajectories of each joint; the other based on inertial sensors, collecting acceleration and angular velocity data of different limb segments by having the athlete wear inertial measurement units (IMUs) on each segment. Both approaches provide relatively rich joint kinematic data, but due to the lack of input from the foot's biomechanical boundaries, they can only describe how the body moves, not how force is transferred from the ground to the racket. More importantly, the foot pressure detection and motion capture systems are independent in their data acquisition, each using different sensors, recording at its own time reference, and outputting data in its own format. There is no quantitative correlation between the two at the biomechanical transmission level, and they cannot establish a complete biomechanical mapping relationship from the foot to the racket. Essentially, they are just two sets of observational data from the same trainee in different dimensions, rather than upstream and downstream links in the same kinetic chain. Summary of the Invention
[0004] The embodiments of the present invention provide a table tennis intelligent learning method based on multi-source sensor fusion, which aims to solve the problem that in the existing table tennis motion analysis technology, foot pressure analysis and motion capture analysis are independent of each other and lack quantitative correlation at the mechanical transmission level, which makes it impossible to physically model and quantitatively evaluate the motion quality from the complete chain of force generation from the foot to the racket hitting the ball.
[0005] To achieve the above objectives, this invention provides a table tennis intelligent learning method based on multi-source sensor fusion, comprising the following steps: The foot pressure distribution time sequence data of trainees performing table tennis techniques is collected by foot pressure sensor array, and the resultant pressure vector of the foot and the position of its point of application at each moment are calculated. The trainee's full-body motion image sequence during the execution of the table tennis technique is captured by a depth camera. The three-dimensional coordinates of key points of the human body are extracted based on a deep learning posture estimation algorithm, and the time-series data of rotation angle and angular velocity of each joint are calculated. At the same time, the time-series data of the spatial motion parameters of the racket are collected by a six-axis inertial sensing unit set inside the racket. Using the resultant pressure vector of the sole at each moment and the position of its point of application as the bottom mechanical boundary condition, and using the time series data of the rotation angle and the time series data of the joint angular velocity as the kinematic input, the net torque time series curve and instantaneous power time series curve of each joint in the path from the ankle joint to the wrist joint during the ball hitting phase are calculated by progressively escalating upwards based on the human multi-rigid-body dynamics model. Based on the net torque time-series curves and instantaneous power time-series curves of each joint, the dominant force-generating joints and power chain transmission efficiency during the execution of the table tennis technique are determined.
[0006] Furthermore, the power chain transmission efficiency during the execution of the table tennis technique is determined based on the instantaneous power timing curve of each joint, including: calculating the cumulative positive power and cumulative negative power of each joint during the hitting phase, and determining the power transmission efficiency between joints based on the ratio of the cumulative positive power between adjacent joints.
[0007] Furthermore, when the cumulative positive power of the proximal joint is greater than that of the distal joint, and the power decreases progressively from the proximal to the distal end, the kinetic chain transmission direction is determined to be correct; when the cumulative positive power of the distal joint is greater than that of the proximal joint, the kinetic chain is determined to have compensatory force.
[0008] Furthermore, it also includes: calculating the work contribution rate of each joint during the ball-hitting phase based on the net torque time-series curve of each joint, and determining the joint with the highest work contribution rate as the dominant force-generating joint in the execution process of the table tennis technique.
[0009] Furthermore, it also includes: determining whether the start-up sequence of each joint meets the requirement of sequential order from proximal to distal based on the peak occurrence order of the instantaneous power timing curves of each joint within a preset time window before and after the ball striking moment.
[0010] Furthermore, the human multi-rigid-body dynamics model includes a trunk segment, left and right upper arm segments, left and right forearm segments, left and right thigh segments, and left and right lower leg segments. The joint degrees of freedom between each limb segment include the flexion-extension and rotational degrees of freedom of the hip joint, the flexion-extension degrees of freedom of the knee joint, the flexion-extension degrees of freedom of the ankle joint, the flexion-extension and abduction degrees of freedom of the shoulder joint, and the flexion-extension degrees of freedom of the elbow joint.
[0011] Furthermore, the plantar pressure vector is obtained by multiplying the pressure values of each sensing point of the foot pressure sensing array by their corresponding pressure directions and then superimposing the vectors. The position of the point of application of the plantar pressure vector is calculated by weighted averaging of the pressure values of each sensing point with respect to their spatial position.
[0012] Furthermore, the net torque time-series curves and instantaneous power time-series curves of each joint are calculated by progressively ascending the calculation process. This includes: using the plantar pressure vector and its point of application as the external force input at the ankle joint, and combining the inertial parameters of the lower leg segment and the angular velocity of the ankle joint, the net torque at the ankle joint is calculated using the Newton-Euler equations; using the net torque at the ankle joint as the external force input at the knee joint, and combining the inertial parameters of the thigh segment and the angular velocity of the knee joint, the net torque at the knee joint is calculated; and this process is repeated progressively upwards until the wrist joint.
[0013] Furthermore, the instantaneous power time-series curve is calculated based on the product of the net torque of each joint and the corresponding joint angular velocity, and the instantaneous power value at each moment is the dot product of the net torque vector and the angular velocity vector at that moment.
[0014] Furthermore, it also includes: comparing the power chain transmission efficiency with a preset standard power chain efficiency model, identifying energy leakage points in the power chain transmission process of the trainee, and generating correction prompts for the energy leakage points.
[0015] The above technical solution has the following technical effects: This invention uses the plantar pressure vector and its point of application as the bottom mechanical boundary conditions, and joint rotation angles and angular velocities as kinematic inputs. Through a multi-rigid-body dynamics model of the human body, it recursively calculates the net torque and instantaneous power of each joint, achieving a physical model of the complete mechanical transmission path from the foot to the racket in table tennis training. Traditional methods can only output two independent sets of observational data: plantar pressure distribution and joint motion angles. These two sets lack a physical correlation and are essentially just two different dimensions describing the same trainee. This invention, through inverse dynamics, establishes a causal mapping relationship between plantar pressure and joint motion at the mechanical level, allowing trainees to see how much force is transmitted and consumed at each joint after being generated from the ground.
[0016] This invention achieves a quantitative assessment of kinetic chain transmission efficiency by calculating the cumulative positive power of each joint during the striking phase and the ratio between adjacent joints. Traditional methods evaluate the kinetic chain based on the coach's subjective experience, lacking objective quantitative indicators. This invention transforms the abstract concept of whether the kinetic chain is smooth into a calculable power transmission ratio. When the power of the proximal joints gradually decreases to the distal joints, the kinetic chain is considered correct; otherwise, compensatory force is identified, providing trainees with quantifiable technical diagnostic evidence.
[0017] This invention achieves automatic identification of the dominant force-generating joint and precise determination of the joint initiation sequence through time-series analysis of the net torque and instantaneous power of each joint. The dominant force-generating joint is determined based on the joint with the highest work contribution rate. The order in which power peaks occur determines whether each joint is initiated sequentially from proximal to distal. This allows trainees and coaches to accurately locate abnormal force patterns and incorrect initiation sequences in technical movements, overcoming the shortcomings of traditional methods that only provide posture angle deviations but cannot explain the origin and destination of force.
[0018] This invention compares the kinetic chain transmission efficiency with a standard model, identifies energy leakage points, and generates corrective prompts. While conventional kinetic chain analysis can only determine the overall kinetic chain condition, this invention can pinpoint the specific joint where energy leakage occurs, allowing trainees to specifically correct the movement details of those joints, thus improving training efficiency. Attached Figure Description
[0019] Figure 1 This is a flowchart illustrating an embodiment of the intelligent learning method for table tennis based on multi-source sensor fusion according to the present invention. Detailed Implementation
[0020] To further illustrate the various embodiments, the present invention provides accompanying drawings. These drawings are part of the disclosure of the present invention, primarily used to illustrate the embodiments and to explain the operating principles of the embodiments in conjunction with the relevant descriptions in the specification. With reference to these drawings, those skilled in the art should be able to understand other possible implementations and the advantages of the present invention. Components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.
[0021] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments.
[0022] Figure 1 This is a flowchart illustrating a table tennis intelligent learning method based on multi-source sensor fusion provided in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps: By collecting time-series data on the distribution of plantar pressure during the execution of table tennis techniques by trainees using a foot pressure sensor array, the resultant plantar pressure vector and its point of application at each moment are calculated.
[0023] The foot pressure sensor array is a thin-film pressure sensor array distributed inside the training insole. When the foot pressure sensor array is placed inside the training insole, thin-film pressure sensors are embedded in the forefoot area, arch area and heel area of each insole. Each insole contains no less than 8 sensing points, with one insole for each foot, for a total of more than 16 sensing points.
[0024] The core of plantar pressure distribution data lies in combining the discrete pressure values collected from each sensor point into a unified mechanical vector. The pressure direction at each sensor point is perpendicular to the plane of the plantar pressure sensor array and upwards. Therefore, the magnitude of the combined plantar pressure vector is equal to the algebraic sum of the pressure values at each sensor point, with the direction vertically upwards. The location of the point of application of the combined plantar pressure vector, i.e., the center of plantar pressure, is calculated by weighted averaging of the pressure values at each sensor point with respect to their spatial location. Specifically, the abscissa of the center of plantar pressure in the plane of the plantar pressure sensor array is equal to the sum of the products of the pressure values at each sensor point and their abscissas, divided by the total pressure value; the ordinate is equal to the sum of the products of the pressure values at each sensor point and their ordinates, divided by the total pressure value. The combined plantar pressure vector describes the magnitude, direction, and point of application of the total force of the interaction between the trainee's foot and the ground, and is the most crucial bottom mechanical boundary condition in subsequent inverse dynamics recursive calculations. The continuous change trajectory of the center of plantar pressure during the ball-hitting process directly reflects the direction and speed of the trainee's center of gravity movement, while the curve of the magnitude of the combined plantar pressure vector over time reflects the magnitude and timing of the trainee's push-off force.
[0025] It's important to note that the change in the position of the plantar pressure center not only reflects the geometric distribution of force on the foot but also implies the movement of the trainee's hip joint, pelvis, and body rotation axis. Since the foot is the only link in the human kinetic chain in contact with the ground, the continuous displacement trajectory of the plantar pressure center in the forward-backward and left-right directions is essentially a projection of the pelvis's movement pattern in three-dimensional space onto the foot's support surface. When the trainee completes trunk rotation during a shot, the pelvis, as the biomechanical hub between the trunk and lower limbs, transmits its rotation through the hip joint to the femur, tibia, and finally the foot, altering the load distribution on the inner and outer sides and the front and back of the foot, thus driving the plantar pressure center to move in a specific direction. Therefore, there is a one-to-one correspondence between the direction of movement of the plantar pressure center and the direction of pelvic rotation. Taking a forehand topspin as an example, during the backswing phase, the pelvis rotates to the right rear, and the pressure center moves towards the right heel; during the swing and shot phase, the pelvis rotates to the left front, and the pressure center moves rapidly towards the left forefoot. The amplitude and speed of the pressure center's movement directly reflect the amplitude and angular velocity of pelvic rotation. The time difference between the pressure center reaching a specific position and the moment the hip joint angle reaches its peak quantitatively characterizes the coordination between plantar force application and hip joint rotation. Similarly, the position of the body's rotation axis around its longitudinal axis can be inferred from the relative relationship between the pressure centers of the two feet: when the body's rotation axis is centered, the pressure distribution of the left and right feet is basically symmetrical; when the rotation axis shifts to the left, the proportion of pressure borne by the left foot increases significantly, and the distance between the pressure centers of the left and right feet also changes accordingly. This mechanical coupling relationship between plantar pressure distribution and the movement state of the hip joint and pelvis means that the data collected by the foot pressure sensor array is not only independent plantar information, but also the mechanical basis supporting subsequent inverse dynamics recursive calculations that progressively extrapolate the net torque of the hip joint, trunk, and even the upper limb joints from the ankle joint upwards.
[0026] The system uses a depth camera to capture a sequence of full-body motion images of the trainee performing the table tennis technique. Based on a deep learning posture estimation algorithm, it extracts the three-dimensional coordinates of key points on the human body and calculates the time-series data of rotation angles and angular velocities of each joint. At the same time, it uses a six-axis inertial sensing unit installed inside the racket to collect the time-series data of the racket's spatial motion parameters.
[0027] The depth camera uses an Intel RealSense D435 or equivalent performance, positioned 3 to 5 meters directly in front of the trainee, with the lens optical axis roughly perpendicular to the trainee's sagittal plane. The camera is approximately 1.2 to 1.5 meters above the ground to cover the trainee's full range of motion. The depth camera includes an RGB camera and an infrared depth sensor, simultaneously outputting color and depth images at a frame rate of at least 30fps and a depth image resolution of at least 640×480 pixels. After deployment, the Zhang Zhengyou calibration method is used to calibrate the intrinsic parameters of the depth camera, obtaining the camera's intrinsic parameter matrix and distortion coefficients. Simultaneously, a multi-view calibration board is used to calibrate the camera's extrinsic parameters in the world coordinate system, including rotation matrices and translation vectors. After calibration, the transformation relationship from the camera coordinate system to the world coordinate system is fixed, and all subsequent calculations are performed in the world coordinate system.
[0028] The sequence of full-body motion images captured by a depth camera is processed frame by frame. A deep learning-based pose estimation algorithm is used to perform human detection and keypoint localization on the RGB images. The pose estimation algorithm employs HRNet, which, through a parallel multi-resolution subnet structure, fuses multi-scale information while maintaining high-resolution feature maps, outputting the two-dimensional coordinates of the shoulder, elbow, wrist, hip, knee, and ankle joints, as well as key points of the torso, on the image plane. Simultaneously, the depth values corresponding to each pixel are read from the depth image. For any keypoint, its two-dimensional coordinates on the image plane, its corresponding depth value, and the camera intrinsic parameter matrix are back-projected into three-dimensional space. After traversing all keypoints, a three-dimensional coordinate sequence of human keypoints is obtained.
[0029] After obtaining the three-dimensional coordinate sequence of key human body points, the joints are connected according to the topology of the human skeleton to construct the direction vectors of each limb segment. Limb segments include the upper arm, forearm, thigh, lower leg, and trunk. Taking the upper arm segment as an example, its direction vector points from the three-dimensional coordinates of the shoulder joint to the three-dimensional coordinates of the elbow joint; the direction vector of the forearm segment points from the three-dimensional coordinates of the elbow joint to the three-dimensional coordinates of the wrist joint. The rotation angles of each joint are calculated based on the spatial angles between the direction vectors of adjacent limb segments. Rotation angles include shoulder flexion-extension angle, shoulder abduction-adduction angle, elbow flexion-extension angle, wrist flexion-extension angle, hip flexion-extension angle, knee flexion-extension angle, ankle flexion-extension angle, trunk forward tilt angle, and trunk lateral tilt angle. A local coordinate system is established with the trainee's pelvic center point as the origin. The pelvic center point is defined as the midpoint between the left and right hip key points. A right-handed coordinate system is formed by using the direction from the center of the pelvis to the key points of the torso as the positive Y-axis, the direction from the center of the pelvis to the key points of the left or right hip as the positive X-axis, and the cross product of the X and Y axes as the positive Z-axis. The rotation angles of each joint are expressed in the local coordinate system by transforming them from the world coordinate system to the local coordinate system. This transformation includes translation and rotation. The reason for establishing the local coordinate system is that the trainee may move in space during training. If the joint angles are expressed in the world coordinate system, changes in the trainee's overall position will affect the calculation results. The local coordinate system, based on the trainee's own pelvis, can eliminate the interference of overall displacement and orientation changes on the joint angle calculation.
[0030] The joint angular velocity time-series data is obtained by time-difference analysis of the rotation angle time-series data of each joint. For any joint, its angular velocity at time t is equal to the difference between the rotation angle at time t and time t-Δt, divided by the time interval Δt. The value of Δt depends on the frame rate of the depth camera. Taking a frame rate of 30fps as an example, Δt is approximately 0.033 seconds. After traversing all time points, the angular velocity time-series curves of each joint are obtained.
[0031] The spatial motion parameters of a racket include at least three core dimensions: racket speed, impulse, and finger pressure. Racket speed refers to the racket's velocity in three-dimensional space, obtained by integrating the measurements from the three-axis accelerometers in a six-axis inertial sensing unit over time. It includes the magnitude of the resultant velocity and its components along each coordinate axis. The rate of change of racket speed, i.e., acceleration, directly reflects the magnitude of the resultant force exerted on the racket by the trainee, while the direction of racket speed reflects the direction of the racket's trajectory at the moment of impact. In a forehand attack, racket speed reaches its peak at the moment of contact; in a forehand topspin, racket speed maintains a high tangential component after contact to generate spin.
[0032] Impulse refers to the magnitude and direction of the impulse at the moment of impact between the racket and the ping-pong ball. It is calculated based on the sudden changes in racket acceleration and angular velocity detected by a six-axis inertial sensing unit. When the racket collides with the ball, the racket experiences an instantaneous impact force from the ball, causing a sudden change in the racket's acceleration and angular velocity within a very short time. The magnitude of the impulse is equal to the change in momentum of the racket before and after the moment of contact, i.e., the absolute value of the racket's mass multiplied by the change in velocity; the direction of the impulse is the same as the direction of the change in velocity. The impulse parameter reflects the intensity of the racket-ball interaction at the moment of impact; a larger impulse value means greater hitting power and higher ball speed. Different types of hitting techniques correspond to different impulse characteristic curves: the impulse peak of a smash is the highest and the shortest in duration, the impulse peak of a loop is slightly lower but the duration is slightly longer, and the impulse of a chop is manifested as a short, low-amplitude pulse.
[0033] Finger pressure refers to the distribution and timing of pressure changes exerted by the trainee's fingers on the racket handle or racket face. Thin-film pressure sensors are embedded on the front, back, and sides of the racket handle, or multiple pressure sensing points are placed along the back edge of the racket face to collect data on the distribution of grip force exerted by the trainee's fingers and palm on the racket during the swing and impact. Finger pressure data is transmitted to a data processing module via wired or wireless means, with a sampling frequency of at least 500Hz. Changes in finger pressure signals reflect the trainee's fine control over their wrist and fingers. During the backswing phase before impact, finger pressure is in a relatively stable grip state; at the moment of impact, subtle adjustments in finger and wrist force cause instantaneous changes in finger pressure distribution. For example, during a forehand topspin, the pressure difference between the thumb and index finger increases to coordinate with wrist pronation, while during a backhand push, the pressure difference decreases to coordinate with wrist abduction. The time difference between the moment of finger pressure change and the moment of abrupt change in racket acceleration can be used to assess the accuracy of the trainee's wrist force application. Excessive finger pressure indicates that the trainee is gripping the racket too tightly, resulting in wrist stiffness and limiting racket speed; insufficient finger pressure indicates that the grip is too loose, leading to unstable racket posture during the shot and reducing the efficiency of impulse transfer.
[0034] Racquet speed, impulse, and finger pressure are three dimensions that describe the racket's motion state from three levels: macroscopic motion speed, collision mechanics, and microscopic hand control. All three are directly acquired or indirectly calculated through the racket's built-in sensor array, without the need for external markers or auxiliary equipment. The data format is time-series signals, which can be directly aligned and fused with joint kinematic data for time axis analysis.
[0035] Using the resultant pressure vector of the sole at each moment and the position of its point of application as the bottom mechanical boundary condition, and using the time series data of the rotation angle and the time series data of the joint angular velocity as the kinematic input, the net torque time series curve and instantaneous power time series curve of each joint in the path from the ankle joint to the wrist joint during the ball hitting phase are calculated by progressively escalating upwards based on the human multi-rigid-body dynamics model.
[0036] The human multi-rigid-body dynamics model is the foundation for inverse dynamics calculations in this invention. Based on the anatomical structure of the human skeleton, this model comprises nine limb segments: the trunk, left and right upper arm segments, left and right forearm segments, left and right thigh segments, and left and right lower leg segments. These limb segments are connected by joints with degrees of freedom, including flexion / extension and rotation of the hip joint, flexion / extension of the knee joint, flexion / extension of the ankle joint, flexion / extension and abduction of the shoulder joint, and flexion / extension of the elbow joint. Each limb segment is treated as a rigid body, and its elastic deformation during motion is ignored. For each limb segment, its mass distribution parameters, center of mass position parameters, and moment of inertia parameters are determined based on anthropometric statistical data. Anthropometric statistics can be obtained by referring to the human inertial parameter table provided in Winter's "Biomechanics and Motor Control of Human Movement", or by referring to the regression equation of inertial parameters for Chinese adults published by Zheng Xiuqin et al. Based on the height and weight of the trainees, the proportion of the mass of each limb segment to the total weight, the proportion of the center of mass located near the proximal end of the limb segment, and the moment of inertia around the center of mass can be obtained by referring to the table.
[0037] The core idea of inverse dynamics is to infer the forces and torques at each joint of an object from the known laws of motion and the external forces acting on it. In this invention, the known information includes: the resultant force vector of the foot and its point of application (external force at the foot), the rotation angle and angular velocity time series data of each joint (kinematic laws), and the mass and moment of inertia of each limb segment (inertial parameters). The unknown information is the net torque at each joint—that is, the net torque transmitted by the muscles through the joints. Inverse dynamics establishes a quantitative relationship between kinematic and dynamic quantities using the Newton-Euler equations, thereby inferring the unknown quantities from the known quantities.
[0038] The Newton-Euler equations consist of two fundamental equations. For any limb segment, its translational motion satisfies Newton's second law: the net external force acting on that limb segment is equal to the mass of that limb segment multiplied by the linear acceleration of its center of mass. Its rotational motion satisfies Euler's equations: the net external torque acting on that limb segment is equal to the moment of inertia of that limb segment about its center of mass multiplied by its angular acceleration. By considering the various limb segments of the human body as a rigid chain of hinged joints, and starting from the farthest or nearest end, isolating each limb segment one by one, and listing its Newton-Euler equations, the net torque at each joint can be derived step by step.
[0039] In this invention, the recursive calculation starts from the ankle joint at the bottom and proceeds upwards along the kinetic chain to the wrist joint. The specific recursive process is as follows.
[0040] The resultant force vector of the plantar force and its point of application are used as the external force input at the ankle joint. The resultant force vector of the plantar force is the total reaction force of the ground acting on the sole of the foot, which is transmitted to the ankle joint through the foot. Combining the inertial parameters of the lower leg segment, including its mass, center of mass position, and moment of inertia, as well as the time series data of the ankle joint's angular velocity, the Newton-Euler equations are formulated for the lower leg segment. The resultant external force on the lower leg segment includes: the ground reaction force transmitted from the sole of the foot (known), the joint reaction force at the ankle joint (unknown), and the weight of the lower leg segment itself (known). The resultant external torque on the lower leg segment includes: the torque generated by the ground reaction force on the ankle joint (which can be calculated from the point of application and the magnitude of the force), the net torque at the ankle joint (unknown), and the torque generated by the weight of the lower leg segment on its center of mass (known). According to Newton's second law, the linear acceleration of the lower leg segment's center of mass multiplied by the lower leg segment's mass equals the resultant external force; according to Euler's equations, the moment of inertia of the lower leg segment multiplied by the angular acceleration equals the resultant external torque. By substituting the known quantities into the two equations above, we can solve for the joint reaction force and net torque at the ankle joint.
[0041] The net torque at the ankle joint is used as the external force input at the knee joint. The net torque at the ankle joint is the torque transmitted from the ankle joint to the upper end of the lower leg, and its reaction force is the torque exerted by the upper end of the lower leg on the knee joint. Combining the inertial parameters of the thigh segment, including its mass, center of mass position, and moment of inertia, as well as the angular velocity time series data of the knee joint, the Newton-Euler equations are formulated for the thigh segment. The net external forces acting on the thigh segment include: the force transmitted from the ankle joint (known), the joint reaction force at the knee joint (unknown), and the weight of the thigh segment itself (known). The net external torques acting on the thigh segment include: the torque transmitted from the ankle joint (known), the net torque at the knee joint (unknown), and the torque generated by the weight of the thigh segment about its center of mass (known). Substituting the known quantities into the Newton-Euler equations, the joint reaction force and the net torque at the knee joint can be solved inversely.
[0042] The same method is used to calculate upwards step by step: the net torque at the knee joint is used as the external force input at the hip joint, and combined with the inertial parameters of the torso segment and the angular velocity of the hip joint, the net torque at the hip joint is calculated in reverse; the net torque at the hip joint is used as the external force input at the shoulder joint, and combined with the inertial parameters of the upper arm segment and the angular velocity of the shoulder joint, the net torque at the shoulder joint is calculated in reverse; the net torque at the shoulder joint is used as the external force input at the elbow joint, and combined with the inertial parameters of the forearm segment and the angular velocity of the elbow joint, the net torque at the elbow joint is calculated in reverse; finally, the calculation is extended to the wrist joint. This yields the time-series curves of the net torque of each joint along the path from the ankle to the wrist joint during the hitting phase.
[0043] Instantaneous power is a core physical quantity describing the rate of energy transfer. The instantaneous power time-series curve of each joint is calculated based on the product of the net torque of each joint and the corresponding angular velocity. For any joint, its instantaneous power value at time t is equal to the dot product of the net torque vector and the angular velocity vector at that time. The physical meaning of the dot product is the product of the component of the torque in the rotational direction and the angular velocity; when the torque direction is the same as the rotational direction, the power is positive, indicating that the joint outputs energy to the distal end (centripetal contraction); when the torque direction is opposite to the rotational direction, the power is negative, indicating that the joint absorbs energy (eccentric contraction). By traversing all moments during the striking phase, the instantaneous power time-series curve of each joint can be obtained.
[0044] The start and end times of the hitting phase are determined by the acceleration signals from the six-axis inertial sensing unit. When a sudden peak occurs in the racket acceleration signal, that moment is considered the hitting moment. The hitting phase is defined as a time window from 200 milliseconds before the hitting moment to 100 milliseconds after the hitting moment, covering the entire process from power preparation to the completion of the hit.
[0045] Based on the net torque time-series curves and instantaneous power time-series curves of each joint, the dominant force-generating joints and power chain transmission efficiency during the execution of the table tennis technique are determined.
[0046] The work contribution rate of each joint during the hitting phase is calculated as follows: For any joint, the instantaneous power value at each moment during the hitting phase is integrated over time to obtain the total work done by that joint during the hitting phase. The sign of the total work reflects whether the joint outputs or absorbs energy during the hitting phase. Using the sum of the positive total work of all joints (i.e., the output energy) as the denominator and the positive total work of each joint as the numerator, the proportion of the positive work done by each joint to the total positive work is calculated, which is the work contribution rate of that joint. The joint with the highest work contribution rate is identified as the dominant force-generating joint in the execution of table tennis techniques. Taking the forehand topspin as an example, if the hip joint has the highest work contribution rate, it indicates that the trainee mainly relies on trunk rotation for power generation; if the shoulder joint has the highest work contribution rate, it indicates that the trainee mainly relies on upper limb power generation. The latter often means that the kinetic chain transmission efficiency is low because the power is not fully utilized by the large muscle groups of the lower limbs and trunk.
[0047] The activation sequence of each joint is determined by the order in which the peak values of the instantaneous power sequence curves of each joint appear during the hitting phase. For each joint, the moment when its instantaneous power reaches its maximum value during the hitting phase is identified as the peak power moment of that joint. The peak power moments of each joint are compared in the order from proximal to distal (ankle, knee, hip, shoulder, elbow, wrist). If the peak power moments of each joint appear sequentially from proximal to distal, the activation sequence of each joint is considered to meet the requirement of proximal to distal order, i.e., the activation sequence of the kinetic chain is correct. If the peak power moment of the distal joint is earlier than that of the proximal joint, the activation sequence is considered abnormal, indicating that the trainee used the distal joint too early during force exertion, resulting in the failure of the proximal large muscle groups to effectively transmit the force to the distal end.
[0048] Quantitative assessment of kinetic chain transmission efficiency is achieved by calculating the cumulative positive power of each joint during the impact phase and its ratio to that of adjacent joints. The cumulative positive power of each joint is the integral of the power value of that joint at each moment during the impact phase when its instantaneous power is positive, over time. Physically, it represents the total energy output by that joint to the distal end during the impact phase. For adjacent joints, such as the hip and shoulder joints, the ratio of the cumulative positive power of the shoulder joint to that of the hip joint is calculated; this ratio represents the power transmission efficiency between the hip and shoulder joints. This ratio ranges from 0 to 1. The closer the ratio is to 1, the more completely the energy output from the hip joint is transferred to the shoulder joint; the lower the ratio, the more energy is lost during the hip-shoulder transmission. The power transmission efficiency between adjacent joints, such as shoulder-elbow and elbow-wrist, is calculated in the same way. When the cumulative positive power of the proximal joint is greater than that of the distal joint, and the power transfer efficiency between adjacent joints decreases progressively from proximal to distal (i.e., ankle-knee transfer efficiency is higher than knee-hip, knee-hip is higher than hip-shoulder, hip-shoulder is higher than shoulder-elbow, and shoulder-elbow is higher than elbow-wrist), the kinetic chain direction is considered correct. When the cumulative positive power of a distal joint is greater than that of its proximal joint, it is determined that the kinetic chain at that joint is compensating for insufficient force, meaning the proximal joint is not exerting sufficient force, and the distal joint is exerting excessive force to compensate for the lack of proximal force.
[0049] The kinetic chain transmission efficiency is compared with a pre-defined standard kinetic chain efficiency model to identify energy leakage points in the trainee's kinetic chain transmission process. The standard kinetic chain efficiency model is derived from statistical results of the same technical movements by multiple professional table tennis players. It calculates the average and standard deviation of the power transmission efficiency between adjacent joints under the same technical movement, forming a standard efficiency range. When the power transmission efficiency between a trainee's joints falls below the lower limit of the standard efficiency range, an energy leakage is identified at that joint, which is considered the energy leakage point. After identifying the energy leakage point, corrective prompts are generated for that point. These prompts include at least the name of the energy leakage point, the deviation of the current transmission efficiency from the standard value, and a suggested correction direction. For example, if the power transmission efficiency between the shoulder and elbow joints is low, the corrective prompt could be: "Shoulder to elbow joint power transmission efficiency 65%, lower than the standard value of 80%. It is recommended to keep the shoulder joint relaxed during the swing and avoid shoulder locking." Based on these prompts, the trainee can clearly understand which part of their kinetic chain is problematic and how to correct it accordingly.
[0050] The embodiments described in the above specific embodiments are preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A table tennis intelligent learning method based on multi-source sensor fusion, characterized in that, Includes the following steps: The foot pressure distribution time sequence data of trainees performing table tennis techniques is collected by foot pressure sensor array, and the resultant pressure vector of the foot and the position of its point of application at each moment are calculated. The trainee's full-body motion image sequence during the execution of the table tennis technique is captured by a depth camera. The three-dimensional coordinates of key points of the human body are extracted based on a deep learning posture estimation algorithm, and the time-series data of rotation angle and angular velocity of each joint are calculated. At the same time, the time-series data of the spatial motion parameters of the racket are collected by a six-axis inertial sensing unit set inside the racket. Using the resultant pressure vector of the sole at each moment and the position of its point of application as the bottom mechanical boundary condition, and using the time series data of the rotation angle and the time series data of the joint angular velocity as the kinematic input, the net torque time series curve and instantaneous power time series curve of each joint in the path from the ankle joint to the wrist joint during the ball hitting phase are calculated by progressively escalating upwards based on the human multi-rigid-body dynamics model. Based on the net torque time-series curves and instantaneous power time-series curves of each joint, the dominant force-generating joints and power chain transmission efficiency during the execution of the table tennis technique are determined.
2. The intelligent learning method for table tennis based on multi-source sensor fusion according to claim 1, characterized in that, The power chain transmission efficiency during the execution of the table tennis technique is determined based on the instantaneous power time-series curve of each joint, including: calculating the cumulative positive power and cumulative negative power of each joint during the hitting phase, and determining the power transmission efficiency between joints based on the ratio of the cumulative positive power between adjacent joints.
3. The intelligent learning method for table tennis based on multi-source sensor fusion according to claim 2, characterized in that, When the cumulative positive power of the proximal joint is greater than that of the distal joint, and the power decreases progressively from the proximal to the distal, the kinetic chain transmission direction is determined to be correct; when the cumulative positive power of the distal joint is greater than that of the proximal joint, the kinetic chain is determined to have compensatory force.
4. The intelligent learning method for table tennis based on multi-source sensor fusion according to claim 1, characterized in that, Also includes: The work contribution rate of each joint during the ball-hitting phase is calculated based on the net torque time-series curve of each joint, and the joint with the highest work contribution rate is identified as the dominant force-generating joint in the execution of the table tennis technique.
5. The intelligent learning method for table tennis based on multi-source sensor fusion according to claim 1, characterized in that, Also includes: Based on the order of peak occurrence of the instantaneous power timing curves of each joint within the preset time window before and after the ball is struck, it is determined whether the start-up timing of each joint meets the requirement of sequential order from proximal to distal.
6. The intelligent learning method for table tennis based on multi-source sensor fusion according to claim 1, characterized in that, The human multi-rigid-body dynamics model includes a trunk segment, left and right upper arm segments, left and right forearm segments, left and right thigh segments, and left and right lower leg segments. The joint degrees of freedom between each limb segment include the flexion-extension and rotational degrees of freedom of the hip joint, the flexion-extension degrees of freedom of the knee joint, the flexion-extension degrees of freedom of the ankle joint, the flexion-extension and abduction degrees of freedom of the shoulder joint, and the flexion-extension degrees of freedom of the elbow joint.
7. The intelligent learning method for table tennis based on multi-source sensor fusion according to claim 1, characterized in that, The plantar pressure vector is obtained by multiplying the pressure values of each sensing point of the foot pressure sensor array by their corresponding pressure directions and then superimposing the vectors. The position of the application point of the plantar pressure vector is calculated by weighted averaging of the pressure values of each sensing point with respect to their spatial position.
8. The intelligent learning method for table tennis based on multi-source sensor fusion according to claim 1, characterized in that, The net torque time-series curve and instantaneous power time-series curve of each joint are calculated by recursively calculating upwards step by step. This includes: using the resultant force vector of the foot and its point of application as the external force input at the ankle joint, and combining the inertial parameters of the lower leg segment and the angular velocity of the ankle joint, the net torque at the ankle joint is calculated by back-calculating using the Newton-Euler equation; using the net torque at the ankle joint as the external force input at the knee joint, and combining the inertial parameters of the thigh segment and the angular velocity of the knee joint, the net torque at the knee joint is calculated by back-calculating upwards step by step until the wrist joint.
9. The intelligent learning method for table tennis based on multi-source sensor fusion according to claim 8, characterized in that, The instantaneous power time-series curve is calculated based on the product of the net torque of each joint and the corresponding angular velocity of the joint. The instantaneous power value at each moment is the dot product of the net torque vector and the angular velocity vector at that moment.
10. The intelligent learning method for table tennis based on multi-source sensor fusion according to claim 1, characterized in that, Also includes: The kinetic chain transmission efficiency is compared with a preset standard kinetic chain efficiency model to identify energy leakage points in the trainee's kinetic chain transmission process, and corrective prompts are generated for the energy leakage points.
