Sports action correction and monitoring system based on image recognition and big data

Abstract

The application provides a sports action correction and monitoring system based on image recognition and big data, and relates to the technical field of artificial intelligence.The data synchronization and fusion module uses big data processing technology to integrate multi-source data, eliminate space-time differences, and improve the accuracy and reliability of analysis results.The biomechanics solving module uses Lagrange equation to calculate the lower limb joint torque vector in real time, and realizes instant monitoring of the joint torque.The load optimization module uses Adam optimization algorithm to optimize the stride width and knee flexion angle through gradient descent method, minimizes the maximum joint torque, reduces the joint load of athletes, and prevents potential harm.The feedback display module transmits the optimized action parameters to wearable devices in real time, guides athletes to adjust actions in real time, and ensures the efficiency and safety of training.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a sports movement correction and monitoring system based on image recognition and big data. Background Technology

[0002] Real-time monitoring of athletes' movement parameters and biomechanical indicators provides coaches with scientific and objective training data, helping athletes optimize their movement patterns and improve their competitive level. Furthermore, the development of devices such as inertial measurement units (IMUs) and ground reaction force sensors has made it possible to measure key biomechanical parameters such as lower limb joint torques during training, further advancing sports science research.

[0003] In the existing technology, publication number CN111738044A, entitled "A Method for Assessing School Violence Based on Deep Learning Behavior Recognition," the method involves collecting surveillance video data distributed throughout the school, using surveillance cameras at different locations as classification labels, splitting the video stream data into different continuous frame groups, inputting them, training, and constructing a three-dimensional convolutional neural network (3D-CNN) school violence assessment model, and using cross-validation to test the model's generalization ability. Based on this, the method identifies the action category of newly input data for individuals, determines the safety status of their location, and issues alerts for abnormal behavior. In the context of the big data era, this not only ensures the scientific, efficient, and safe management but also provides an effective solution for preventing and controlling school violence.

[0004] In indoor training centers specializing in sprinting, coaches rely on the naked eye and single-camera video playback to correct stride length and landing point, but they cannot simultaneously know the instantaneous joint load of the athlete's lower limbs, making it difficult to detect potential injury points in time;

[0005] Existing systems often lack the ability to simultaneously acquire and fuse multi-source data, making it difficult to effectively integrate image data, inertial measurement data, and ground reaction force data, thus limiting the accuracy and reliability of the analysis results. Furthermore, traditional biomechanical analysis methods often suffer from computational delays when calculating complex parameters such as joint torques in real time, failing to provide athletes with immediate movement adjustment suggestions and consequently affecting the effectiveness of movement optimization.

[0006] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a sports movement correction and monitoring system based on image recognition and big data, in order to solve the problems mentioned in the background.

[0008] The objective of this invention is achieved as follows: a sports movement correction and monitoring system based on image recognition and big data, specifically comprising:

[0009] Data acquisition module: used to acquire multi-view synchronous image streams, three-dimensional ground reaction force data and six-axis inertial measurement data of the target athlete during sprinting during the monitoring period in real time;

[0010] 3D pose estimation module: used to receive multi-view synchronous image streams and extract the 3D coordinate sequence of multiple lower limb key points of the target athlete;

[0011] Data synchronization and fusion module: used to fuse three-dimensional coordinate sequences, three-dimensional ground reaction force data, and six-axis inertial measurement data to output initial value vectors for sports movement correction at multiple lower limb key points;

[0012] Biomechanics Solver Module: Used to calculate the lower limb joint torque vector of a target athlete during sprinting exercise based on the Lagrange equation;

[0013] Load optimization module: It is used to receive the lower limb joint torque vector of the target athlete during sprinting during the monitoring period, optimize the initial value vector of sports movement correction, and output the corrected sports movement parameter vector of each lower limb key point;

[0014] Feedback display module: Used to receive the corrected sports movement parameter vectors of each lower limb key point and output movement adjustment instructions to the wearable display terminal in real time.

[0015] Furthermore, the multi-view synchronized image stream achieves global shutter synchronization through a hardware synchronizer;

[0016] The three-dimensional pose estimation module uses the HRNet-W48 deep learning network as the backbone network and combines a differentiable triangulation algorithm to calculate the three-dimensional coordinates of the lower limb key points.

[0017] The 3D pose estimation module includes a camera calibration submodule, used to obtain the intrinsic parameter matrices of the three cameras. With extrinsic matrix ;

[0018] The three-dimensional pose estimation module further includes a two-dimensional keypoint detection submodule, used to output the two-dimensional pixel coordinates of six lower limb keypoints in each frame of image. ;in This represents six key points of the lower limbs, and c=1,2,3 represents three cameras;

[0019] The 3D pose estimation module includes a differentiable triangulation layer, used to calculate the 3D coordinates of six lower limb key points based on the camera projection matrix and the pixel coordinates of the 2D key points. .

[0020] Furthermore, the triaxial ground reaction force data specifically includes:

[0021] The triaxial force vector of the triaxial ground reaction force data is set as follows: ;

[0022] in, It is a three-dimensional ground reaction force vector; It is a horizontal component force that is directed forward and backward or left and right. It is vertical support force;

[0023] The six-axis inertial measurement data specifically includes:

[0024] The six-axis inertial measurement data are characterized as follows;

[0025]

[0026] in, , , These are the attitude quaternions, linear acceleration, and angular velocity of the six-axis inertial measurement data at time t, respectively. Used to characterize the direction or orientation of multiple lower limb key points of the target athlete in space, that is, the rotational state of the object relative to the defined coordinate system;

[0027] Posture Quaternion Used to represent the IMU attitude of multiple lower limb key points at time t;

[0028] Linear acceleration Used to represent the triaxial acceleration of multiple lower limb key points at time t;

[0029] angular velocity Used to represent the triaxial angular velocity of multiple lower limb key points at time t.

[0030] Furthermore, the data synchronization and fusion module uses a Kalman filter algorithm to integrate the three-dimensional coordinates of six lower limb key points. ,speed Triaxial ground reaction force and six-axis inertial measurement data Data fusion is performed to generate a unified initial value vector for sports movement correction. Specifically:

[0031] The data synchronization and fusion module sets a unified timestamp based on the GPS-PPS reference. ;

[0032] The data synchronization and fusion module uses a rigid body transformation matrix. Translation vector The three-dimensional coordinates of six key points of the lower limbs Mapped to the laboratory coordinate system, this achieves interaction with the triaxial ground reaction force. and six-axis inertial measurement data A unified spatial reference;

[0033] Establish the Kalman filter model: Define the state transition matrix of the Kalman filter. and observation matrix ;

[0034] Using the defined Kalman filter model, input the initial value vector for sports motion correction. and observation vector Perform prediction and update steps;

[0035] Set the updated state vector as the initial value vector for correcting the sports movement at the current moment. .

[0036] Furthermore, the biomechanical solution module is based on the Lagrange equation and incorporates the mass-inertia matrix. Coriolis term and centrifugal force term Gravity terms Calculate the torque vector of the lower limb joints ;

[0037] The biomechanical solution module establishes a 7-DOF lower limb biomechanical model, including a 3-DOF hip joint, a 1-DOF knee joint, and a 3-DOF ankle joint.

[0038] The biomechanical solution module sets the mass parameters for each segment of the lower limb. Among them, ankle mass quality of knee joint Hip joint quality Quality parameters In this context, "j" represents the ankle, knee, and hip joints.

[0039] The biomechanical solution module describes the dynamic equations of each segment of the lower limb using the Lagrange equation and the difference between kinetic energy T and potential energy V, L=TV.

[0040] Define the lower limb joint torque vector as:

[0041]

[0042] in, For hip joint torque; This refers to the knee joint torque. This refers to the ankle joint torque.

[0043] Furthermore, the specific logic for outputting the corrected sports movement parameter vectors for each lower limb key point includes:

[0044] The load optimization module reads the lower limb joint torque vector sequence within a preset time period and calculates the peak vector of each joint torque. ;

[0045] The load optimization module defines a cost function. Lower limb joint torque vector The function for finding the maximum absolute value is as follows:

[0046]

[0047] These are adjustable motion parameters, representing stride width and knee flexion angle, respectively.

[0048] The load optimization module employs the Adam optimization algorithm to adjust action parameters over multiple iterations. To minimize the cost function ; and the optimized motion parameter vector is denoted as .

[0049] Furthermore, the load optimization module calculates the difference in action parameters between the current time and the previous time, forming a parameter increment vector. ;

[0050] Calculate the difference between the optimized action parameter vector and the parameter vector at the previous time step:

[0051]

[0052] in, This is the optimized motion parameter vector; This is the action parameter vector from the previous moment;

[0053] The load optimization module will optimize the action parameters. and parameter increment vector The data is transmitted to the feedback display module for real-time motion correction.

[0054] Define the optimized action parameter vector for:

[0055]

[0056] in, This is the optimized stride width; The optimized knee flexion angle.

[0057] Furthermore, the feedback display module guides athletes to adjust their stride and knee flexion angle in real time through a head-up display (HUD) and vibration feedback, forming a closed-loop motion correction mechanism.

[0058] Increase stride Mapped to the length of the arrow on the head-up display (HUD), it indicates the direction and magnitude of the stride adjustment;

[0059] Change in knee flexion angle Mapped to an angle scale on the HUD, indicating the angle that the knee joint needs to be adjusted.

[0060] When stride increment At that time, the vibration motor fixed at the athlete's iliac crest is activated, and the vibration prompts the athlete to adjust their stride.

[0061] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention simultaneously acquires multi-view image streams, ground reaction force data and six-axis inertial measurement data through the data acquisition module, ensuring that all key parameters during the movement are fully recorded; the three-dimensional posture estimation module adopts advanced image recognition algorithms to accurately extract the three-dimensional coordinates of multiple key points of the lower limbs, providing a basis for detailed motion analysis;

[0062] The data synchronization and fusion module utilizes big data processing technology to integrate multi-source data, eliminate spatiotemporal differences, and improve the accuracy and reliability of analysis results; the biomechanical solution module is based on the Lagrange equation to calculate the lower limb joint torque vector in real time, enabling real-time monitoring of joint torque;

[0063] The load optimization module uses the Adam optimization algorithm to optimize stride width and knee flexion angle through gradient descent, minimizing the maximum joint torque, reducing the joint load on athletes and preventing potential injuries.

[0064] The feedback display module transmits the optimized motion parameters to the wearable device in real time, guiding athletes to adjust their movements immediately and ensuring the efficiency and safety of training. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0066] Figure 1 This is a schematic diagram of the overall system modules of the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] As shown in the figure

[0069] Example 1:

[0070] Please see Figure 1 The present invention provides a technical solution:

[0071] A sports movement correction and monitoring system based on image recognition and big data, specifically including:

[0072] Data acquisition module: used to acquire multi-view synchronous image streams, three-dimensional ground reaction force data and six-axis inertial measurement data of the target athlete during sprinting during the monitoring period in real time;

[0073] Further explanation: The multi-view synchronized image stream achieves global shutter synchronization through a hardware synchronizer; and the multi-view synchronized image stream is 240 frames per second;

[0074] The specific steps to achieve global shutter synchronization are as follows:

[0075] 1.1) Camera arrangement: Three Sony IMX273 cameras are arranged on the side of the sprint track, forming an equilateral triangle; a triangular baseline with a length of 4m is set along the running direction at the side edge of the track.

[0076] The optical axes of all three camera lenses point to the center line of the runway, with a horizontal angle of 120°.

[0077] The camera optical center height is 1.20m, and the tripod top plate leveling error is ≤0.2°;

[0078] The camera is connected to the GPU workstation via a 10GbE fiber optic cable, providing a dedicated data channel for synchronous configuration.

[0079] 1.2) Synchronization Configuration: Each camera is configured with a global shutter, which receives GPS-PPS pulses through a hardware synchronizer to achieve global shutter synchronization. The PPS (Pulse Per Second) pulse synchronization error is ≤0.1 milliseconds. The specific operation is as follows:

[0080] The GPS-PPS signal is sent to the trigger port of each camera via an 8-channel splitter;

[0081] Set the camera trigger mode to "external trigger priority" and the trigger edge to the rising edge;

[0082] Calibrate trigger delay: Read the trigger-exposure delay given by the manufacturer and write it to the register;

[0083] Record the synchronization error: The difference between the three exposure signals measured with an oscilloscope is ≤0.1ms;

[0084] After synchronization is complete, a 240fps global shutter image stream is output for use in the ground force table installation and alignment.

[0085] 1.3) Data transmission: The camera captures images in RAW 12-bit format at 240 frames per second and transmits them in real time to the GPU workstation via a 10GbE network interface. The data size of a single frame is 1.57MB.

[0086] Example 2:

[0087] 3D pose estimation module: used to receive multi-view synchronous image streams and extract the 3D coordinate sequence of multiple lower limb key points of the target athlete;

[0088] Further explanation: The three-dimensional pose estimation module uses the HRNet-W48 deep learning network as the backbone network and combines a differentiable triangulation algorithm to calculate the three-dimensional coordinates of the lower limb key points;

[0089] Sprinting involves high-speed movements and significant transient force exertion in the lower limbs, requiring precise reconstruction of key points of the lower limbs from two dimensions into three dimensions from the perspective of three simultaneous cameras.

[0090] HRNet-W48 can maintain spatial consistency of key points in multi-resolution fusion;

[0091] Differentiable triangulation layers enable end-to-end training of the entire network, facilitating joint optimization of camera parameters and 2D / 3D keypoints;

[0092] The Gauss-Newton method can converge within 0.3ms / point on the GPU with 10 iterations, meeting the real-time requirement of 240Hz.

[0093] The 3D pose estimation module includes a camera calibration submodule, used to obtain the intrinsic parameter matrices of the three cameras. With extrinsic matrix ;

[0094] The specific implementation steps are as follows:

[0095] A 9×6 checkerboard calibration board was used, with 10 different postures placed on the side of the track;

[0096] The intrinsic parameter matrices of each camera were calculated using the OpenCV-Zhang calibration procedure. Furthermore, panoramic stitching optimization was performed on the external parameters, resulting in an average residual of ≤0.12px for the three cameras;

[0097] internal parameter matrix With extrinsic matrix Stored as a floating-point matrix and permanently written to the HRNet-W48 inference script configuration file.

[0098] The three-dimensional pose estimation module further includes a two-dimensional keypoint detection submodule, used to output the two-dimensional pixel coordinates of six lower limb keypoints in each frame of image. ;in This represents six key points of the lower limbs, and c=1,2,3 represents three cameras;

[0099] The specific implementation steps are as follows:

[0100] Image preprocessing: The synchronous image streams acquired by the three cameras are channel normalized, scaled to the range of [0,1], and the image resolution is adjusted to 384×288 pixels to adapt to the input requirements of the HRNet-W48 model.

[0101] Key point detection: Load the weight file of the HRNet-W48 model trained on a 50TB sports action dataset, perform inference on each frame of the image, and extract the two-dimensional pixel coordinates of six lower limb key points.

[0102] Subpixel regression: Gaussian subpixel regression is applied to the output heatmap of the HRNet-W48 model to accurately locate the subpixel position of each keypoint. The quantization accuracy reaches 0.01 pixels;

[0103] Data encapsulation: The two-dimensional coordinates of the six lower limb key points and their corresponding confidence values ​​are encapsulated into a structured data structure Key2D[c][i] for subsequent triangulation layer processing.

[0104] The 3D pose estimation module includes a differentiable triangulation layer, used to calculate the 3D coordinates of six lower limb key points based on the camera projection matrix and the pixel coordinates of the 2D key points. .

[0105] The specific implementation steps are as follows:

[0106] Initialization: For each lower limb keypoint i, use the two-dimensional pixel coordinates of the three cameras. The initial 3D coordinate estimates are obtained by using the Direct Linear Transformation (DLT) method. ;

[0107] Error function construction:

[0108] Construct the reprojection error function for each lower limb key point at the current time t:

[0109]

[0110] in, The three-dimensional coordinates of key points of the lower limb;

[0111] Let be the perspective projection function of the c-th camera. and For the camera's focal length, and The coordinates of the camera's principal point;

[0112] Let be the two-dimensional pixel coordinates of the i-th lower limb key point captured by the c-th camera at time t;

[0113] The Gauss-Newton method was used to analyze the error function. Iterative optimization is performed, with the number of iterations set to 10. The estimated 3D coordinates are updated in each iteration. ;

[0114] Using inverse depth parameterization, the three-dimensional coordinates are represented as , where d is the depth distance, to enhance the stability of the optimization process;

[0115] Convergence criterion: In each iteration, if the coordinate update magnitude... If mm, the optimization process will be terminated prematurely.

[0116] After optimization, the final 3D coordinates are output. The unit is millimeters, and the gradient information is fed back to the two-dimensional key point detection submodule to support end-to-end network training.

[0117] The 3D pose estimation module includes a joint training and parameter optimization submodule, which is used to simultaneously optimize the 2D keypoint detection weights and the parameters of the differentiable triangulation layer.

[0118] The specific implementation steps are as follows:

[0119] The 50TB competitive motion dataset was divided into training, validation, and test sets in an 8:1:1 ratio. The training set contains 2 million frames of labeled lower limb keypoint data in COCO format.

[0120] The total loss function is defined as follows:

[0121]

[0122] in, The loss is the mean squared error (MSE) for 2D keypoint detection. L1 distance loss for 3D coordinate prediction; =1, =0.5, which is the weight coefficient of the loss function;

[0123] The Adam optimizer was used, with an initial learning rate set to 1×10⁻ 4 The batch size is 32, and the learning rate is adjusted using the Cosine Decay strategy.

[0124] The training process consists of 60 epochs. After each epoch, the model performance is evaluated using a validation set, and the mean joint error (MPJPE) is monitored.

[0125] If the MPJPE of the validation set does not improve within 5 consecutive training cycles, training will be terminated early.

[0126] The three-dimensional attitude estimation module outputs a three-dimensional coordinate sequence of six lower limb key points and synchronizes this sequence with other sensor data according to timestamps;

[0127] The specific implementation steps are as follows:

[0128] During the reasoning process, the calculated three-dimensional coordinates Write to the shared memory region.

[0129] A 64-bit timestamp from the GPS-PPS signal source is added to the beginning of each output data frame, with a timestamp accuracy of 0.05 milliseconds, to ensure data time synchronization.

[0130] The timestamped 3D coordinate data is pushed to the "data synchronization and fusion module" via the ZeroMQ protocol, and aligned with the ground reaction force (GRF) and inertial measurement unit (IMU) data under a unified time reference.

[0131] The beneficial effects of this embodiment are:

[0132] 1. High-precision camera calibration: By using a 9×6 grid calibration board and the OpenCV-Zhang calibration algorithm, high-precision intrinsic and extrinsic parameter calibration of three cameras was achieved, with reprojection error residual ≤0.12 pixels, significantly improving the accuracy of 3D coordinate calculation.

[0133] 2. Differentiable triangulation algorithm: By introducing a differentiable triangulation layer, the 3D pose estimation process can be jointly optimized end-to-end. The backpropagation algorithm is used to automatically adjust camera parameters and key point positions, thereby achieving higher accuracy of 3D coordinates.

[0134] 3. Large-scale data training: The model is trained using a self-built 50TB competitive motion dataset, which includes 2 million frames of labeled lower limb keypoint data to ensure the model's generalization ability and robustness in real-world motion scenarios, with a mean joint point error (MPJPE) of less than 0.9 mm.

[0135] 4. Real-time performance guarantee: The Gauss-Newton method is used for fast iterative optimization on the GPU (10 iterations, convergence time ≤ 0.3 milliseconds per point) to ensure that the system can achieve real-time 3D pose estimation at a high frame rate of 240Hz.

[0136] 5. Multi-source data synchronization: High-precision time synchronization of 3D coordinate data with GRF and IMU data is achieved through GPS-PPS signals (time alignment error ≤ 0.05 milliseconds), which improves the accuracy of multi-source data fusion analysis, unlike existing technologies that only perform analysis of a single data source.

[0137] Example 2:

[0138] Further explanation: The aforementioned triaxial ground reaction force data specifically includes:

[0139] The sampling frequency of the triaxial ground reaction force data is 1kHz;

[0140] The triaxial force vector of the triaxial ground reaction force data is set as follows: ;

[0141] in, It is a three-dimensional ground reaction force vector; It is a horizontal component force that is directed forward and backward or left and right. It is vertical support force;

[0142] The specific operating steps are as follows:

[0143] An AMTI OR6-7 force measurement platform was selected, with a range of ±10kN and a sensitivity of 2µV / N.

[0144] The force measuring platform is buried in the starting area of ​​the runway, with the platform surface flush with the runway surface, and the installation error is controlled within ±0.5mm;

[0145] A four-point bolt fastening platform is used to ensure that its verticality error is less than 0.05°;

[0146] Configure the platform data acquisition system and set the sampling frequency to 1kHz;

[0147] A fourth-order Butterworth low-pass filter with a cutoff frequency of 100Hz is used to filter the original data, outputting a triaxial force vector. (Unit: Newton);

[0148] The collected data packets are stored in a structured manner, with a sampling timestamp appended to each data frame, which can then be used by the data fusion module.

[0149] Connect the GPS-PPS pulse to the SYNC_IN port of the force platform controller via a shielded cable;

[0150] By using GPS-PPS signals, the platform's data frames and the image acquisition module are time-aligned, ensuring that the time alignment error is less than 0.05ms.

[0151] A unified timestamp is written into the data stream to record the data frame acquisition time for subsequent multi-module data fusion.

[0152] The ground reaction force acquisition module provides mechanical data synchronized with the output of the three-dimensional attitude estimation module. Through the data synchronization and fusion module, the GRF data is accurately spatiotemporally correlated with the three-dimensional coordinate sequence of lower limb key points, providing the necessary mechanical information for the biomechanical solution module.

[0153] Further explanation: The six-axis inertial measurement data specifically includes:

[0154] The six-axis inertial measurement data are characterized as follows;

[0155]

[0156] in, , , These are the attitude quaternions, linear acceleration, and angular velocity of the six-axis inertial measurement data at time t, respectively. Used to characterize the direction or orientation of multiple lower limb key points of the target athlete in space, that is, the rotational state of the object relative to the defined coordinate system;

[0157] Posture Quaternion Used to represent the IMU attitude of multiple lower limb key points at time t;

[0158] Linear acceleration Used to represent the triaxial acceleration (m / s²) of multiple lower limb key points at time t.

[0159] angular velocity Used to represent the triaxial angular velocity (degrees / second) of multiple lower limb key points at time t.

[0160] The specific steps are as follows:

[0161] 2.1) Layout and installation: Bosch BMI270 IMUs were selected, each configured with a measuring range of ±16g and ±2000° / s;

[0162] The IMU is fixed to the athlete's waist (near the L4 vertebral level area) and the anterior edge of the left and right tibias. It is fixed with a 3D printed shell (weighing <25g) and medical-grade double-sided tape + Velcro combination to ensure vibration resistance with an anti-vibration frequency greater than 200Hz.

[0163] Check the installation angle of each IMU to ensure that the installation direction of each device is consistent with the human body coordinates, and the installation error does not exceed 1°.

[0164] 2.2) Attitude Determination and Data Acquisition:

[0165] Set the IMU to SPI slave mode and configure the external clock input, using a phase-locked GPS-PPS pulse (1kHz) as the external clock;

[0166] Using a Mahony filter with parameter β=0.05, real-time attitude determination is performed on the original acceleration and angular velocity signals, and the output quaternion is calculated. ;

[0167] Simultaneously collect linear acceleration and angular velocity Numerical values ​​are encapsulated into a data frame structure. ;

[0168] Each data frame contains a precise timestamp, determined by the device's internal counter and the synchronization results with an external GPS-PPS, with the sampling rate maintained at 1kHz.

[0169] 2.3) Time alignment:

[0170] The Firefly synchronization pulse signal is input to each IMU through the GPIO interface to achieve phase-locking with the GPS-PPS signal and ensure that the internal and external clocks of all IMUs are aligned.

[0171] Use an oscilloscope to detect the time drift between IMU data frames, ensuring that the time drift is less than 0.2 ms / hour;

[0172] The timestamps in each IMU data frame are unified with the image and force platform, and subsequent spatiotemporal correlation processing is performed through the data synchronization and fusion module.

[0173] The six-axis inertial measurement data acquired by the inertial attitude acquisition module is precisely aligned in time and spatially with the data from the three-dimensional attitude estimation module and the ground reaction force acquisition module through the data synchronization and fusion module, providing complete lower limb motion dynamic information for the biomechanical solution module.

[0174] Example 3:

[0175] Data synchronization and fusion module: used to fuse three-dimensional coordinate sequences, three-dimensional ground reaction force data, and six-axis inertial measurement data to output initial value vectors for sports movement correction at multiple lower limb key points;

[0176] Further explanation: The data synchronization and fusion module uses the Kalman filter algorithm to integrate the three-dimensional coordinates of six lower limb key points. ,speed Triaxial ground reaction force and six-axis inertial measurement data Data fusion is performed to generate a unified initial value vector for sports movement correction. Specifically:

[0177] The data synchronization and fusion module sets a unified timestamp based on the GPS-PPS reference. And the three-dimensional coordinates of six key points of the lower limbs Interpolate to 1kHz to ensure time alignment error is less than 0.05 milliseconds; specifically:

[0178] Using GPS-PPS (Global Positioning System - Pulse Per Second) signals as a unified time reference, a unified timestamp is set for the entire system. .

[0179] Three-dimensional coordinate sequence of six key points of the lower limb Original timestamp and unified timestamp Perform alignment to ensure that the time base of all data sources is consistent.

[0180] right The data was calculated using the five-point difference method and interpolated to a sampling rate of 1 kHz. The specific calculation formula is as follows:

[0181]

[0182] in, Let be the three-dimensional coordinates of the i-th lower limb key point at time t; Let be the velocity vector of the i-th lower limb key point; The sampling interval is seconds;

[0183] Using high-precision timing equipment, the time alignment error after interpolation is measured to ensure the error value is accurate. millisecond;

[0184] The data synchronization and fusion module uses a rigid body transformation matrix. Translation vector The three-dimensional coordinates of six key points of the lower limbs Mapped to the laboratory coordinate system, this achieves interaction with the triaxial ground reaction force. and six-axis inertial measurement data A unified spatial reference;

[0185] The specific operating steps are as follows:

[0186] Obtain rigid body transformation parameters: Determine the rigid body transformation matrix based on the camera calibration results. Translation vector The details are as follows:

[0187]

[0188] in, Let be the three-dimensional coordinates of the i-th lower limb key point in the laboratory coordinate system; It is a rotation matrix; It is a translation vector.

[0189] Coordinate transformation: The three-dimensional coordinates of the six interpolated lower limb key points are transformed. By mapping the rigid body transformation described above to the laboratory coordinate system, we obtain three-dimensional coordinates under a unified spatial reference. ;

[0190] Spatial alignment verification: Verification using known spatial calibration points in the laboratory. Spatial alignment accuracy to ensure spatial error millimeters;

[0191] Set the initial value vector for sports movement correction as follows:

[0192]

[0193] in, Let be the three-dimensional coordinates of the i-th lower limb key point at time t;

[0194] Let i be the velocity of the i-th lower limb key point;

[0195] This is the three-dimensional ground reaction force vector;

[0196] This is a six-axis inertial measurement data vector.

[0197] Establish the Kalman filter model: Define the state transition matrix of the Kalman filter. and observation matrix The details are as follows:

[0198]

[0199] in, This is the original observation vector; This is the state transition matrix, used to predict the current state; This is the observation matrix, used to combine the current observation data; The noise is assumed to follow a zero-mean Gaussian distribution.

[0200] Kalman filtering: Using a defined Kalman filter model, the initial value vector for correcting the input sports motion is processed. and observation vector Perform the prediction and update steps; specifically as follows:

[0201]

[0202]

[0203] in, The predicted state vector; The error covariance matrix is ​​the prediction result. Let be the process noise covariance matrix.

[0204] The update steps are as follows:

[0205]

[0206]

[0207]

[0208] in, Kalman gain; The observation matrix; To observe the noise covariance matrix; It is the identity matrix;

[0209] Generate initial value vector: update the state vector Set as the initial value vector for sports movement correction at the current moment. ;

[0210] Correct the initial value vector of sports movements It is written to a circular buffer and transmitted to the subsequent processing module through a high-speed data channel.

[0211] The data synchronization and fusion module outputs a unified motion state vector. This vector contains the initial value vector for sports movement correction. The sampling frequency is 1kHz, and the data is written to a circular buffer for subsequent modules to use.

[0212] The specific operating steps are as follows:

[0213] Let the unified motion state vector be:

[0214]

[0215] Will Data is written sequentially to a circular buffer with a length of 2 seconds to ensure data continuity and real-time performance.

[0216] Configure the data interface to enable the subsequent biomechanical solution module to read data from the circular buffer in real time. This completes the subsequent processing and analysis of the data.

[0217] The beneficial effects of this embodiment are as follows:

[0218] High-precision time alignment: A unified timestamp is set using a GPS-PPS reference, and interpolation to 1kHz is performed using a five-point differential method to align errors in real time, thereby improving the accuracy of multi-source data synchronization.

[0219] Unified spatial reference transformation: Through rigid body transformation matrix and translation vector, the three-dimensional coordinates of six lower limb key points are accurately mapped to the laboratory coordinate system, ensuring spatial consistency and distinguishing it from the problem of inconsistent spatial references in multi-source data in existing technologies.

[0220] Integrated Kalman Filter Algorithm: By introducing the Kalman filter algorithm and combining it with the state transition matrix and observation matrix, efficient fusion of coordinate, velocity, ground reaction force and inertial measurement data is achieved, generating a comprehensive initial value vector for sports movement correction, thus improving the accuracy and real-time performance of data fusion.

[0221] Example 4:

[0222] Biomechanics Solver Module: Used to calculate the lower limb joint torque vector of a target athlete during sprinting exercise based on the Lagrange equation;

[0223] The biomechanical solution module is based on the Lagrange equation and incorporates the mass-inertia matrix. Coriolis term and centrifugal force term Gravity terms Calculate the torque vector of the lower limb joints ;

[0224] Further explanation: 3.1) The biomechanical solution module establishes a 7-DOF lower limb biomechanical model, including a 3-DOF hip joint, a 1-DOF knee joint, and a 3-DOF ankle joint;

[0225] A 7-DOF lower limb biomechanical model was selected to describe the motion and biomechanical characteristics of the hip, knee, and ankle joints. Specifically, this includes:

[0226] Hip joint: 3 degrees of freedom, representing flexion and extension, abduction and adduction, and rotation;

[0227] Knee joint: 1 degree of freedom, representing flexion and extension;

[0228] Ankle joint: 3 degrees of freedom, representing flexion and extension, inversion and eversion, and rotation;

[0229] Define a local coordinate system for each joint to ensure that the motion relationships of each part of the model are consistent with the actual human anatomy; the origin of the coordinate system is set at the center of the joint, and the coordinate axis directions conform to the right-hand rule.

[0230] Set the connection parameters between each joint, including joint length, angle limits, and range of motion, to simulate the dynamic characteristics of the lower limbs of a real athlete.

[0231] 3.2) The biomechanical solution module sets the mass parameters for each segment of the lower limb. Among them, ankle mass quality of the knee joint Hip joint quality Quality parameters In this context, "j" represents the ankle, knee, and hip joints.

[0232] The specific operating steps are as follows: In this embodiment, based on the Dempster proportion table and combined with the test subject's weight of 75kg, the following is obtained: , and ;

[0233] Calculate the mass-inertia matrix based on the mass parameters of each segment. This is used to describe the inertial characteristics of the lower limb joint system; the specific formula for calculating the inertial matrix is ​​as follows:

[0234]

[0235] in, Represents the system in generalized coordinates The inertial coupling term under;

[0236] The mass-inertia matrix was verified through simulation tests. The accuracy of this information is ensured to accurately reflect the dynamic characteristics of the lower limb joint system.

[0237] 3.3) The biomechanical solution module describes the dynamic equations of each segment of the lower limb by using the Lagrange equation and the difference between kinetic energy T and potential energy V, L=TV.

[0238]

[0239] The specific operating steps are as follows:

[0240] Define kinetic energy T and potential energy V: Kinetic energy Potential energy ;

[0241] Using the Lagrange equation L=TV, the dynamic equations of the lower limb joint system are established:

[0242]

[0243] in, Generalized coordinates for lower limb joints, 7 dimensions (hip × 3, knee × 1, ankle × 3);

[0244] For generalized velocity of lower limb joints; For generalized acceleration of the lower limb joints;

[0245] This is the torque vector of the lower limb joints, with units of Newton-meter (N·m).

[0246] The Jacobian matrix for the foot describes the relationship between the foot and the joints;

[0247] This is the vector of the three-dimensional ground reaction force.

[0248] Mass-inertia matrix Coriolis term and centrifugal force term Gravity terms Integrate into the system of equations:

[0249]

[0250] Simulation tests were conducted to verify the accuracy of the Lagrange equations, and the equation parameters were fine-tuned based on actual motion data to ensure the authenticity and accuracy of the dynamic model.

[0251] The biomechanical solution module employs the improved Newmark-β method, with parameters set... , Perform numerical integration, step size Milliseconds, discretely solving a system of equations;

[0252] Based on the initial value vector for sports movement correction provided by the data synchronization and fusion module Define the generalized coordinates of the lower limb joints ,speed and acceleration ;

[0253] The dynamic equations are discretized, and the improved Newmark-β method is applied to iteratively solve for the lower limb joint torques. :

[0254]

[0255]

[0256] Among them, iterative updates are performed step by step. , and Continue until the system of equations reaches the convergence condition.

[0257] In each iteration, if N·m, then the joint torque at the current moment is considered to be Convergence has been achieved; iteration has stopped.

[0258] The biomechanical solver module outputs the lower limb joint torque vector. The unit is Newton-meter (N·m), and it is transmitted to the load optimization module;

[0259] Define the lower limb joint torque vector as:

[0260]

[0261] in, For hip joint torque; This refers to the knee joint torque. This refers to the ankle joint torque.

[0262] The calculated joint torque vector is transmitted via a high-speed data interface. Transmitted to the load optimization module to ensure transmission delay millisecond.

[0263] Joint torque vector Store data in real time in a structured data format (such as CSV or binary format) for easy subsequent analysis and processing.

[0264] The beneficial effects of this embodiment are as follows:

[0265] High-degree-of-freedom biomechanical model: The lower limb biomechanical model adopts a 7-degree-of-freedom model, covering the multi-degree-of-freedom movement of the three major joints of hip, knee and ankle, which is more accurate and comprehensive than the existing single-degree-of-freedom model.

[0266] Precise mass parameter setting: Mass parameters for each segment of the lower limb are set based on the Dempster scale and combined with the actual weight of the test subject to ensure the authenticity of the model's mass distribution and improve the accuracy of the dynamic equation calculation.

[0267] Example 5:

[0268] Load optimization module: It is used to receive the lower limb joint torque vector of the target athlete during sprinting during the monitoring period, optimize the initial value vector of sports movement correction, and output the corrected sports movement parameter vector of each lower limb key point;

[0269] Further explanation: The specific logic for outputting the corrected sports movement parameter vectors for each lower limb key point includes:

[0270] 4.1) The load optimization module reads the lower limb joint torque vector sequence within a preset time period and calculates the peak vector of each joint torque. ;

[0271] The preset time period in this embodiment is within 200 milliseconds;

[0272] 4.1) The specific operating steps are as follows:

[0273] Set the scrolling window length to 200 milliseconds, the sampling frequency to 1 kHz, and the window size to 200 time frames.

[0274] Receive the lower limb joint torque vector sequence from the biomechanics solution module. .

[0275] At the current time t, extract the lower limb joint torque vector sequence with a time range from t-200 milliseconds to t:

[0276]

[0277] Within the extracted scrolling window, calculate the maximum absolute values ​​of the torques at the hip, knee, and ankle joints respectively:

[0278]

[0279] in, , , ;

[0280] The calculated peak vector Stored in a memory buffer for use in subsequent steps; ensures that the scroll window slides for 1 millisecond at a time, enabling real-time processing.

[0281] 4.2) The load optimization module defines the cost function. Lower limb joint torque vector The function for finding the maximum absolute value is as follows:

[0282]

[0283] The specific operating steps are as follows:

[0284] These are adjustable motion parameters, representing stride width and knee flexion angle, respectively.

[0285] The adjustable motion parameter range is set as follows in this embodiment:

[0286] stride width The range is [0.5, 1.5] meters; knee flexion angle The range is [0°, 120°]; set the initial motion parameter vector. This serves as the starting point for the optimization algorithm.

[0287] 4.3) The load optimization module uses the Adam optimization algorithm to adjust the action parameters over multiple iterations. To minimize the cost function ; and the optimized motion parameter vector is denoted as ;

[0288] Further explanation: The learning rate of the Adam optimization algorithm in this embodiment. First-order moment parameters

[0289] The finite difference method is chosen as the cost function. Action parameters The gradient estimation method is as follows:

[0290]

[0291] in, and They were calculated using the finite difference method, respectively.

[0292] Perform 40 optimization iterations, with each iteration executing the following steps:

[0293] Forward propagation: Action parameters based on the current time t Adjust stride width and knee flexion angle, and re-acquire and calculate lower limb joint moment vectors. ;

[0294] Calculate the cost function under the current action parameters. .

[0295] Gradient estimation using the finite difference method .

[0296] Apply the Adam optimizer to update the action parameters based on the gradient:

[0297]

[0298] like If the iteration is terminated prematurely, then the iteration will be terminated early.

[0299] Record the optimized action parameters after each iteration. and the corresponding cost function value This is used for subsequent analysis and verification.

[0300] 4.4) The load optimization module calculates the difference in action parameters between the current time and the previous time, forming a parameter increment vector. ;

[0301] The specific calculation steps are as follows:

[0302] Calculate the difference between the optimized action parameter vector and the parameter vector at the previous time step:

[0303]

[0304] in, This is the optimized motion parameter vector; This is the action parameter vector from the previous moment;

[0305] Set parameter increment vector threshold To ensure the smoothness of motion parameter adjustment; this embodiment ;

[0306]

[0307] The calculated parameter increment vector It is stored in a memory buffer for use in subsequent steps. It should be noted that the threshold in this embodiment... The determination was made by an expert panel using methods such as fuzzy hierarchical analysis (FAHP).

[0308] 4.5) The load optimization module will optimize the action parameters. and parameter increment vector The data is transmitted to the feedback display module for real-time motion correction; this embodiment ensures that the calculation delay does not exceed 4 milliseconds.

[0309] Define the optimized action parameter vector for:

[0310]

[0311] in, This is the optimized stride width; This is the optimized knee flexion angle.

[0312] The optimized motion parameter vector is transmitted via a high-speed data interface. and parameter increment vector The data is transmitted to the feedback display module, ensuring a transmission delay of no more than 4 milliseconds.

[0313] The beneficial effects of this embodiment are as follows:

[0314] The innovativeness of the cost function definition: The maximum absolute value of the lower limb joint torque vector is used as the cost function, which directly optimizes the most unfavorable joint load situation, ensuring that the overall joint torque is reduced, thus enhancing the pertinence and effectiveness of the optimization effect.

[0315] Real-time optimization and low-latency transmission: By strictly controlling the latency of optimization calculation and data transmission, the latency of the entire process is ensured to be no more than 4 milliseconds, realizing real-time action correction. Compared with the higher latency in existing technologies, this significantly improves the real-time performance and response speed of the system.

[0316] Rolling window and peak extraction mechanism: A 200-millisecond rolling window processing and peak vector calculation method are introduced to ensure that the optimization target is based on the recent motion state, thereby enhancing the dynamic adaptability and accuracy of the optimization process.

[0317] Smooth control of parameter increment vector: By setting a parameter increment threshold, excessive parameter adjustments are avoided, ensuring the smoothness of the motion optimization process and the naturalness of human movements, which is different from the abrupt optimization problem that may exist in existing technologies.

[0318] The following description is provided for this embodiment:

[0319] The sprint load optimization experiment was conducted in a standard track and field training field, selecting six national-level youth sprinters as experimental subjects (numbered: Zhang San, Li Si, Wang Wu, Zhao Liu, Sun Qi, and Zhou Ba). Each athlete had completed a 30-meter warm-up sprint at maximum effort. Before the experiment, the height, weight, and lower limb length of each athlete were measured, with an average weight of 75 kg and a lower limb length of 0.92 m. The instruments used in the experiment included: a three-dimensional motion capture system (sampling rate 1 kHz, positioning error <0.5 mm), a ground reaction force sensing platform (sampling rate 1 kHz, error <5 N), a six-axis inertial measurement unit (sampling rate 1 kHz, drift <0.1° / s), and a computing platform (Intel i7, 16 GB RAM). The experiment was conducted at a temperature of 22°C and a humidity of 50%, with uniform lighting.

[0320] The experimental steps are as follows:

[0321] First, the data synchronization and fusion module is activated to obtain the three-dimensional coordinate sequence of lower limb key points for each athlete during the sprint. Joint velocity Ground reaction force and inertial data And generate the sports movement correction initial value vector through Kalman filtering. Then will Input the biomechanical solver module to calculate the lower limb joint moment vector using the Lagrange method. Record the initial maximum torque of the joints: hip joint 120–140 N·m, knee joint 220–245 N·m, ankle joint 105–120 N·m.

[0322] Based on this, the load optimization module is invoked, and its learning rate is... First-order moment parameters Read and calculate the peak vector within the scrolling window. .

[0323] Optimize objective function The maximum value of the three joint peaks is taken, and the Adam algorithm is used for 40 iterations. In each iteration, the sprint state is re-simulated and the new state is recalculated. With stride width and knee flexion angle This is an adjustable parameter. The gradient is estimated using the finite difference method, with the convergence threshold set to [value missing]. ;

[0324] In initial parameters Based on this, iterate until the optimal action parameters are obtained. Stride width was adjusted to the range of 0.92–1.05m, and knee flexion angle was adjusted to the range of 38–52°. Correspondingly, the maximum joint torque decreased significantly after optimization: hip joint to 102–118 N·m, knee joint to 190–215 N·m, and ankle joint to 90–105 N·m. The entire optimization process had a computational latency of ≤4ms per athlete, with real-time output of parameter increment vectors. The results are then transmitted wirelessly to the coach's terminal and the athlete's wearable display.

[0325] The above implementation process fully verifies the stability and real-time performance of the load optimization module in real sprint scenarios. Quantitative data shows that the maximum joint torque is reduced by an average of 15.8%, and the optimized gait parameters are more in line with human biomechanical characteristics, thereby effectively reducing the risk of sports injuries. It has significant innovation and practical value.

[0326] The data table for the example is as follows:

[0327]

[0328] The data in the table above clearly demonstrates the optimization effect of the load optimization module in actual sprinting: stride length and flexion angle parameters are reasonably adjusted, and the peak torque of the three joints is significantly reduced, with an average reduction rate of about 15%.

[0329] Example 6:

[0330] Feedback display module: Used to receive the corrected sports movement parameter vectors of each lower limb key point and output movement adjustment instructions to the wearable display terminal in real time.

[0331] Further explanation: The feedback display module guides athletes to adjust their stride and knee flexion angle in real time through a head-up display (HUD) and vibration feedback, forming a closed-loop motion correction mechanism;

[0332] Increase stride Mapped to the length of the arrow on the head-up display (HUD), it indicates the direction and magnitude of the stride adjustment;

[0333] Change in knee flexion angle Mapped to an angle scale on the HUD, indicating the angle that the knee joint needs to be adjusted.

[0334] Parameter increment vector via UDP protocol Send the data to the HUD glasses worn by the athlete to ensure real-time and stable data transmission, and maintain a rendering frequency of 90Hz;

[0335] When stride increment At this time, a vibration motor fixed to the athlete's iliac crest is activated, prompting the athlete to adjust their stride through vibration; in this embodiment, the vibration motor is 100Hz; the number of trigger pulses... Furthermore, in this embodiment... The expert group makes adaptive adjustments based on actual experimental needs, using methods such as fuzzy hierarchical analysis (FAHP), without imposing limitations.

[0336] Subscribe to HUD acknowledgment signals to monitor the sending and receiving status of data packets in real time.

[0337] If a data packet loss is detected, the system will automatically retransmit it within 2 frames to ensure the integrity of the feedback command.

[0338] The overall closed-loop delay of the system is controlled within 42 milliseconds, which meets the requirements for real-time action correction.

[0339] The feedback display module displays the incremental vector of motion parameters generated by the load optimization module through a HUD and vibration feedback device. The feedback is transmitted to athletes in real time. Through dual visual and tactile feedback, athletes can instantly understand and adjust stride width and knee flexion angle, achieving precise correction and optimization of movement, forming a complete closed-loop system.

[0340] It should be noted that all calculation formulas in this application employ regression analysis, including but not limited to machine learning algorithms, to deeply analyze the collected parameters and identify their natural trends and interrelationships. Specialized software, such as Python's Scikit-learn library or the R language, is used to automatically generate mathematical models that match the data. Then, cross-validation and other methods are used to objectively evaluate the model performance, and continuous feedback and optimization are combined to ensure that the created formulas truly reflect the inherent laws of the data, thereby guaranteeing their effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula undergo dimensionless processing within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless processing techniques include, but are not limited to, Min-Max Normalization and Z-Score standardization.

[0341] The technical solution of this invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of this invention.

[0342] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0343] The above description of the embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A sports movement correction and monitoring system based on image recognition and big data, characterized in that, Specifically, it includes: Data acquisition module: used to acquire multi-view synchronous image streams, three-dimensional ground reaction force data and six-axis inertial measurement data of the target athlete during sprinting during the monitoring period in real time; 3D pose estimation module: used to receive multi-view synchronous image streams and extract the 3D coordinate sequence of multiple lower limb key points of the target athlete; Data synchronization and fusion module: used to fuse three-dimensional coordinate sequences, three-dimensional ground reaction force data, and six-axis inertial measurement data to output initial value vectors for sports movement correction at multiple lower limb key points; Biomechanics Solver Module: Used to calculate the lower limb joint torque vector of a target athlete during sprinting exercise based on the Lagrange equation; Load optimization module: It is used to receive the lower limb joint torque vector of the target athlete during sprinting during the monitoring period, optimize the initial value vector of sports movement correction, and output the corrected sports movement parameter vector of each lower limb key point; The specific logic for outputting the corrected sports movement parameter vectors for each lower limb key point includes: The load optimization module reads the lower limb joint torque vector sequence within a preset time period and calculates the peak vector of each joint torque. ; The load optimization module defines a cost function. Lower limb joint torque vector The function for finding the maximum absolute value is as follows: ； These are adjustable motion parameters, representing stride width and knee flexion angle, respectively. The load optimization module employs the Adam optimization algorithm to adjust action parameters over multiple iterations. To minimize the cost function ; and the optimized motion parameter vector is denoted as ; Feedback display module: Used to receive the corrected sports movement parameter vectors of each lower limb key point and output movement adjustment instructions to the wearable display terminal in real time.

2. The sports motion correction and monitoring system based on image recognition and big data according to claim 1, characterized in that: The multi-view synchronized image stream achieves global shutter synchronization through a hardware synchronizer; The three-dimensional pose estimation module uses the HRNet-W48 deep learning network as the backbone network and combines a differentiable triangulation algorithm to calculate the three-dimensional coordinates of the lower limb key points. The 3D pose estimation module includes a camera calibration submodule, used to obtain the intrinsic parameter matrices of the three cameras. With extrinsic matrix ; The three-dimensional pose estimation module further includes a two-dimensional keypoint detection submodule, used to output the two-dimensional pixel coordinates of six lower limb keypoints in each frame of image. ;in This represents six key points of the lower limbs, and c=1,2,3 represents three cameras; The 3D pose estimation module includes a differentiable triangulation layer, used to calculate the 3D coordinates of six lower limb key points based on the camera projection matrix and the pixel coordinates of the 2D key points. .

3. The sports movement correction and monitoring system based on image recognition and big data according to claim 2, characterized in that: The triaxial ground reaction force data specifically includes: The triaxial force vector of the triaxial ground reaction force data is set as follows: ; in, It is a three-dimensional ground reaction force vector; It is a horizontal component force that is directed forward and backward or left and right. It is vertical support force; The six-axis inertial measurement data specifically includes: The six-axis inertial measurement data are characterized as follows; ； in, , , These are the attitude quaternions, linear acceleration, and angular velocity of the six-axis inertial measurement data at time t, respectively. Used to characterize the direction or orientation of multiple lower limb key points of the target athlete in space, that is, the rotational state of the object relative to the defined coordinate system; Posture Quaternion Used to represent the IMU attitude of multiple lower limb key points at time t; Linear acceleration Used to represent the triaxial acceleration of multiple lower limb key points at time t; angular velocity Used to represent the triaxial angular velocity of multiple lower limb key points at time t.

4. The sports movement correction and monitoring system based on image recognition and big data according to claim 3, characterized in that: The data synchronization and fusion module uses a Kalman filter algorithm to integrate the three-dimensional coordinates of six key points on the lower limbs. ,speed Triaxial ground reaction force and six-axis inertial measurement data Data fusion is performed to generate a unified initial value vector for sports movement correction. Specifically: The data synchronization and fusion module sets a unified timestamp based on the GPS-PPS reference. ; The data synchronization and fusion module uses a rigid body transformation matrix. Translation vector The three-dimensional coordinates of six key points of the lower limbs Mapped to the laboratory coordinate system, this achieves interaction with the triaxial ground reaction force. and six-axis inertial measurement data A unified spatial reference; Establish the Kalman filter model: Define the state transition matrix of the Kalman filter. and observation matrix ; Using the defined Kalman filter model, input the initial value vector for sports motion correction. and observation vector Perform prediction and update steps; Set the updated state vector as the initial value vector for correcting the sports movement at the current moment. .

5. The sports movement correction and monitoring system based on image recognition and big data according to claim 4, characterized in that: The biomechanical solution module is based on the Lagrange equation and incorporates the mass-inertia matrix. Coriolis term and centrifugal force term Gravity terms Calculate the torque vector of the lower limb joints ; The biomechanical solution module establishes a 7-DOF lower limb biomechanical model, including a 3-DOF hip joint, a 1-DOF knee joint, and a 3-DOF ankle joint. The biomechanical solution module sets the mass parameters for each segment of the lower limb. Among them, ankle mass quality of the knee joint Hip joint quality ; Quality parameters In this context, "j" represents the ankle, knee, and hip joints. The biomechanical solution module describes the dynamic equations of each segment of the lower limb using the Lagrange equation and the difference between kinetic energy T and potential energy V, L=TV. Define the lower limb joint torque vector as: ； in, For hip joint torque; This refers to the knee joint torque. This refers to the ankle joint torque.

6. The sports movement correction and monitoring system based on image recognition and big data according to claim 5, characterized in that: The load optimization module calculates the difference in action parameters between the current time and the previous time, forming a parameter increment vector. ; Calculate the difference between the optimized action parameter vector and the parameter vector at the previous time step: ； in, This is the optimized motion parameter vector; This is the action parameter vector from the previous moment; The load optimization module will optimize the action parameters. and parameter increment vector The data is transmitted to the feedback display module for real-time motion correction. Define the optimized action parameter vector for: ； in, This is the optimized stride width; This is the optimized knee flexion angle.

7. The sports movement correction and monitoring system based on image recognition and big data according to claim 6, characterized in that: The feedback display module guides athletes to adjust their stride and knee flexion angle in real time through a head-up display (HUD) and vibration feedback, forming a closed-loop motion correction mechanism. Increase stride Mapped to the length of the arrow on the head-up display (HUD), it indicates the direction and magnitude of the stride adjustment; Change in knee flexion angle Mapped to an angle scale on the HUD, indicating the angle that the knee joint needs to be adjusted. When stride increment At that time, the vibration motor fixed at the athlete's iliac crest is activated, and the vibration prompts the athlete to adjust their stride.

Images