Action data analysis method and system based on smart sports

Abstract

The invention discloses an action data analysis method and system based on smart sports, and relates to the technical field of action data analysis. The motion data analysis method based on smart sports comprises the following steps: acquiring video stream data of a to-be-analyzed motion, inputting the video stream data into a pre-trained key point identification model for prediction analysis to obtain a key point set, performing division to obtain a period key point set of each motion period of the to-be-analyzed motion, and performing analysis to obtain an initial motion integrating degree index; the method comprises the following steps: acquiring a to-be-analyzed action, acquiring action correction data, performing analysis to obtain an action correction index of the to-be-analyzed action, and performing comprehensive analysis in combination with an initial action integrating degree index to obtain a comprehensive action integrating degree index. In this way, the joint angle, the motion track and the periodic change of the athlete can be accurately captured, the performance of the athlete in each motion period can be comprehensively evaluated, and then the motion analysis accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of motion data analysis, and particularly to a motion data analysis method and system based on intelligent sports. Background Art

[0002] Intelligent sports refers to the application of modern information technologies, especially technologies such as the Internet of Things, big data, cloud computing, and artificial intelligence, to the sports field to improve athlete performance, management, etc., and to improve aspects such as sports health management. Motion data analysis is a key application area, which mainly involves aspects such as motion capture, motion evaluation, motion correction, and performance optimization. By accurately collecting the motion data of athletes, such as the motion data during the long-distance running training process of long-distance runners, and analyzing it, it can help coaches and sports medicine experts formulate personalized training plans for athletes and timely adjust the training content and intensity of athletes.

[0003] In the prior art, traditional motion data analysis relies on traditional two-dimensional video analysis or simple sensor data, which easily leads to incomplete and inaccurate motion evaluation and is difficult to comprehensively consider the accuracy of motions.

[0004] The limitations of the prior art at least include the following problems. The prior art is difficult to provide sufficient depth information, resulting in the difficulty of accurately quantifying the changes of athletes' joints in different depth spaces. For example, in track and field training, there are many repetitive motions, and the periodic characteristics of each motion of an athlete are crucial for the final performance. The prior art is difficult to dynamically capture the subtle changes of an athlete within each cycle, thus making it difficult to deeply analyze the key point sets in each motion cycle, which in turn easily leads to low accuracy of joint angles and motion trajectories, and then makes it difficult to comprehensively evaluate complex motions or periodic motions. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a motion data analysis method and system based on intelligent sports, which solves the problems of the prior art lacking depth information, being difficult to accurately quantify the joint changes in periodic motions, resulting in low motion accuracy, and being difficult to comprehensively evaluate.

[0006] To achieve the above object, the present invention is realized through the following technical solutions: A method for analyzing motion data based on intelligent sports, comprising the following steps: obtaining video stream data of the motion to be analyzed, the video stream data including a plurality of frame motion image data, that is, the pixel value, two-dimensional coordinate and corresponding depth information value of each pixel point in the motion image; inputting the video stream data of the motion to be analyzed into a pre-trained key point recognition model for prediction analysis to obtain the key point set of each frame of the motion image of the motion to be analyzed, and performing division processing to obtain the periodic key point set of each motion cycle of the motion to be analyzed; and performing data analysis on the periodic key point set of each motion cycle of the motion to be analyzed to obtain the initial motion fitting degree index of the motion to be analyzed; obtaining the motion correction data of the motion to be analyzed, analyzing to obtain the motion correction index of the motion to be analyzed, and combining with the initial motion fitting degree index for comprehensive analysis to obtain the comprehensive motion fitting degree index of the motion to be analyzed, and the specific formula is as follows: Wherein, ZhQ is the comprehensive motion fitting degree index of the motion to be analyzed, CsQ is the initial motion fitting degree index of the motion to be analyzed, α1 is the initial coefficient stored in the database, β1 is the initial adjustment coefficient stored in the database, DzX is the motion correction index of the motion to be analyzed, α2 is the correction coefficient stored in the database, β2 is the correction adjustment coefficient stored in the database, β3 is the superposition coefficient stored in the database, and α1 + α2 = 1; and training management is performed based on the comprehensive motion fitting degree index of the motion to be analyzed.

[0007] Further, the key point recognition model is specifically a pose estimation network (High-Resolution Network), and the pose estimation network includes an input layer for receiving video stream data, an initial convolutional layer, a multi-resolution stream layer, a feature fusion layer, a bottleneck layer, a key point detection layer, and an output layer.

[0008] Further, the specific steps for obtaining the key point set of each frame of action image to be analyzed are as follows: In the input layer of the pose estimation network, several frames of action image data of the action to be analyzed are received and preprocessed; in the initial convolutional layer of the pose estimation network, feature extraction processing is performed on each frame of action image data of the action to be analyzed after preprocessing to obtain the feature map of each frame of action image of the action to be analyzed; in the multi-resolution flow layer of the pose estimation network, parallel processing is performed on the feature map of each frame of action image of the action to be analyzed to obtain the multi-scale feature map of each frame of action image of the action to be analyzed; in the feature fusion layer of the pose estimation network, fusion processing is performed on the multi-scale feature map of each frame of action image of the action to be analyzed to obtain the multi-scale fusion feature map of each frame of action image of the action to be analyzed; in the bottleneck layer of the pose estimation network, optimization extraction processing is performed on the multi-scale fusion feature map of each frame of action image of the action to be analyzed to obtain the optimized deep feature map of each frame of action image of the action to be analyzed; in the key point detection layer of the pose estimation network, detection processing is performed on the optimized deep feature map of each frame of action image of the action to be analyzed to obtain the key point heat map set of each frame of action image of the action to be analyzed; in the output layer of the pose estimation network, prediction processing is performed on the key point heat map set of each frame of action image of the action to be analyzed to obtain the key point set of each frame of action image of the action to be analyzed, that is, the two-dimensional coordinates of several key points.

[0009] Further, the periodic key point set is specifically the two-dimensional coordinates of each key point of each frame of action image. The specific steps for obtaining the initial action fitness index of the action to be analyzed are as follows: Read the two-dimensional coordinates of each key point of each frame of action image of each action cycle of the action to be analyzed, and perform matching coordinate conversion processing in combination with the two-dimensional coordinates of each pixel point and the corresponding depth information value to obtain the three-dimensional coordinates of each key point of each frame of action image of each action cycle of the action to be analyzed; comprehensively analyze the three-dimensional coordinates of each key point of each frame of action image of each action cycle of the action to be analyzed and the reference three-dimensional coordinates of each reference key point of each frame of reference action image of each reference action cycle stored in the standard database to obtain the joint trajectory consistency index of the action to be analyzed; and perform interpolation correspondence processing on the three-dimensional coordinates of each key point of each frame of action image of each action cycle of the action to be analyzed and the reference three-dimensional coordinates of each reference key point of each frame of reference action image stored in the standard database to obtain the three-dimensional coordinates of each corresponding key point of each frame of corresponding action image of each corresponding action cycle stored in the standard database for the action to be analyzed; and comprehensively analyze the three-dimensional coordinates of each corresponding key point of each frame of corresponding action image of each corresponding action cycle stored in the standard database for the action to be analyzed to obtain the action angle deviation index of the action to be analyzed, and comprehensively analyze it in combination with the joint trajectory consistency index to obtain the initial action fitness index of the action to be analyzed.

[0010] Furthermore, the specific formula for calculating the initial action fitness index of the action to be analyzed is as follows: Among them, CsQ is the initial action fitness index of the action to be analyzed, GyZ is the joint trajectory consistency index of the action to be analyzed, μ1 is the joint trajectory adjustment coefficient stored in the database, DzP is the action angle deviation index of the action to be analyzed, μ2 is the action angle adjustment coefficient stored in the database, and μ3 is the sensitive adjustment coefficient stored in the database.

[0011] Furthermore, the action correction data includes arm length value, leg length value, environmental factor, muscle tension distribution index, and training load index. The specific steps to obtain the action correction index of the action to be analyzed are as follows: Obtain the reference arm length value, reference leg length value, reference environmental factor, reference muscle tension distribution index, and reference training load index stored in the standard database; comprehensively analyze the arm length value, leg length value, environmental factor, muscle tension distribution index, and training load index of the action to be analyzed with the reference arm length value, reference leg length value, reference environmental factor, reference muscle tension distribution index, and reference training load index stored in the standard database to obtain the action correction index of the action to be analyzed.

[0012] Furthermore, the specific formula for calculating the action correction index of the action to be analyzed is as follows: Among them, DzX is the action correction index of the action to be analyzed, is the interaction coefficient stored in the database, CzH is the arm length value of the action to be analyzed, and HcZ is the reference arm length value stored in the standard database, is the arm length adjustment coefficient stored in the database, TcZ is the leg length value of the action to be analyzed, and ZtC is the reference leg length value stored in the standard database, is the leg length adjustment coefficient stored in the database, HjY is the environmental factor of the action to be analyzed, and JhY is the reference environmental factor stored in the standard database, is the environmental adjustment coefficient stored in the database, JfZ is the muscle tension distribution index of the action to be analyzed, and CjF is the reference muscle tension distribution index stored in the standard database, is the muscle tension adjustment coefficient stored in the database, XhL is the training load index of the action to be analyzed, and ChX is the reference training load index stored in the standard database, is the training load adjustment coefficient stored in the database.

[0013] Further, the specific steps for training management based on the comprehensive action fitness index of the action to be analyzed are as follows: Compare and analyze the comprehensive action fitness index of the action to be analyzed with the preset comprehensive action fitness index threshold; if the comprehensive action fitness index of the action to be analyzed is lower than the preset comprehensive action fitness index threshold, take the first training management measure; if the comprehensive action fitness index of the action to be analyzed is not lower than the preset comprehensive action fitness index threshold, take the second training management measure.

[0014] The action data analysis system based on intelligent sports includes: a data acquisition module for acquiring video stream data of the action to be analyzed, where the video stream data includes several frames of action image data, that is, the pixel value, two-dimensional coordinates, and corresponding depth information value of each pixel point in the action image; a prediction and division module for inputting the video stream data of the action to be analyzed into a pre-trained key point recognition model for prediction analysis to obtain the key point set of each frame of action image of the action to be analyzed, and performing division processing to obtain the periodic key point set of each action cycle of the action to be analyzed; a data analysis module for performing data analysis on the periodic key point set of each action cycle of the action to be analyzed to obtain the initial action fitness index of the action to be analyzed; a comprehensive correction module for obtaining the action correction data of the action to be analyzed, analyzing to obtain the action correction index of the action to be analyzed, and performing comprehensive analysis in combination with the initial action fitness index to obtain the comprehensive action fitness index of the action to be analyzed; a training management module for performing training management based on the comprehensive action fitness index of the action to be analyzed.

[0015] The present invention has the following beneficial effects:

[0016] (1) The action data analysis method based on intelligent sports can accurately capture the joint angles, action trajectories, and periodic changes of athletes by acquiring video stream data and inputting it into a pre-trained key point recognition model, so as to comprehensively evaluate the performance of athletes in each action cycle. Furthermore, the subtle joint angle deviations and inconsistencies in action trajectories can be dynamically analyzed, thereby improving the accuracy of action analysis, helping coaches monitor the technical execution of athletes in real time, and providing improvement suggestions in a timely manner.

[0017] (2) The action data analysis method based on intelligent sports can provide immediate feedback to athletes by analyzing each action cycle during training in real time, through key point set division and action correction index, and provide precise correction suggestions based on the generated comprehensive action fitness index, thereby accurately guiding action correction, avoiding long-term training errors caused by action fixation of athletes, and providing personalized training optimization plans to help athletes gradually improve their action technical level and training efficiency in continuous training.

[0018] (3) The method for analyzing motion data based on intelligent sports can provide data-driven training management for athletes through the calculation of the comprehensive motion fitness index. The training progress and motion performance of athletes will be quantified in real time, and combined with the motion correction data of athletes, the training content will be dynamically adjusted, so as to accurately manage the training load of athletes, avoid overtraining or undertraining, and then ensure that athletes are always training in the best state, and then effectively reduce the risk of injuries.

[0019] (4) The system for analyzing motion data based on intelligent sports can achieve real-time and high-precision analysis of athletes' motions through accurate video stream data acquisition and key point recognition technology, and carefully capture the key point sets of each motion cycle for scientific classification, accurately analyze each stage of the motion, provide data on joint trajectory consistency and angle deviation, obtain the initial motion fitness index, and combine the correction data to optimize the motion in real time, so as to dynamically adjust the training plan, thereby improving the accuracy of athletes' motions, and then avoiding the risk of sports injuries caused by improper training, and then continuously adjusting the plan during long-term training, so as to ensure the continuous progress of athletes.

[0020] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flowchart of the method for analyzing motion data based on intelligent sports of the present invention.

[0022] Figure 2 It is a specific step flowchart of obtaining the cycle key point set of each motion cycle of the motion to be analyzed in the method for analyzing motion data based on intelligent sports of the present invention.

[0023] Figure 3 It is a block diagram of the system for analyzing motion data based on intelligent sports of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] Please refer to Figure 1, an embodiment of the present invention provides a technical solution: a method for analyzing motion data based on intelligent sports, including the following steps: obtaining video stream data of the motion to be analyzed (when the athlete is training), the video stream data includes a number of frame motion image data, that is, the pixel value, two-dimensional coordinates and corresponding depth information value of each pixel point in the motion image; inputting the video stream data of the motion to be analyzed into a pre-trained key point recognition model for prediction analysis to obtain the key point set of each frame of the motion image of the motion to be analyzed, and performing division processing to obtain the periodic key point set of each motion cycle of the motion to be analyzed; and performing data analysis on the periodic key point set of each motion cycle of the motion to be analyzed to obtain the initial motion fitness index of the motion to be analyzed; obtaining the motion correction data of the motion to be analyzed, analyzing to obtain the motion correction index of the motion to be analyzed, and combining with the initial motion fitness index for comprehensive analysis to obtain the comprehensive motion fitness index of the motion to be analyzed, and its specific formula is as follows: Where ZhQ is the comprehensive motion fitness index of the motion to be analyzed, CsQ is the initial motion fitness index of the motion to be analyzed, α1 is the initial coefficient stored in the database, β1 is the initial adjustment coefficient stored in the database, DzX is the motion correction index of the motion to be analyzed, α2 is the correction coefficient stored in the database, β2 is the correction adjustment coefficient stored in the database, β3 is the superposition coefficient stored in the database, and α1 + α2 = 1; and perform training management based on the comprehensive motion fitness index of the motion to be analyzed.

[0025] It should be explained that the term exp(-β3*CsQ*DxZ) in the formula is used to adjust the superposition effect of the initial motion fitness index and the motion correction index to avoid the comprehensive motion fitness index being too high or too low.

[0026] α1 and α2 can be obtained through the following steps: read the initial motion fitness index and motion correction index of the motion to be analyzed, and perform summation analysis to obtain the comprehensive motion fitness sum value, and perform ratio analysis on the initial motion fitness index and motion correction index of the motion to be analyzed respectively with the comprehensive motion fitness sum value, and use the ratio analysis result as the corresponding coefficient.

[0027] β1, β2, and β3 can be obtained through the following steps: based on historical data, the initial motion fitness index, and the motion correction index, perform statistical regression analysis to quantify the specific impact of each factor on the comprehensive motion fitness index, so as to fit the initial weight value. Secondly, use the sensitivity analysis method to adjust the value range of each coefficient and observe its impact on the comprehensive motion fitness evaluation result to ensure the stability and rationality of the model.

[0028] Specifically, the key point recognition model is specifically a pose estimation network (High-Resolution Network), and the pose estimation network includes an input layer for receiving video stream data, an initial convolution layer, a multi-resolution stream layer, a feature fusion layer, a bottleneck layer, a key point detection layer, and an output layer.

[0029] Among them, the input layer is used to receive the original video stream data and perform some basic preprocessing (such as normalization, cropping, scaling, etc.), pass the input image to the network, and prepare for subsequent feature extraction.

[0030] The initial convolution layer is used to extract low-level features from the original input image and generate feature maps.

[0031] The multi-resolution stream layer is used for multiple parallel convolution streams to process feature maps of different resolutions. In this layer, feature streams of different resolutions are retained, and different levels of features in the image are processed from low resolution to high resolution.

[0032] The feature fusion layer is used to fuse the features from different resolution streams, enabling the network to simultaneously process the global information of low resolution and the detailed information of high resolution.

[0033] The bottleneck layer is used to further extract features and reduce the computational complexity.

[0034] The key point detection layer is used to predict the positions of each key point in the image (for example, the head, shoulders, knees of the human body, etc.).

[0035] The output layer is the final prediction output, which is used to return the coordinates of each key point of the human body. For the pose estimation task, the output layer will return the two-dimensional coordinates of each joint point.

[0036] The specific steps to obtain the key point set of each frame of action image of the action to be analyzed are as follows:

[0037] In the input layer of the pose estimation network, several frame action image data of the action to be analyzed are received and preprocessed; in the initial convolutional layer of the pose estimation network, feature extraction processing is performed on each frame of action image data of the action to be analyzed after preprocessing (by sliding a convolutional kernel or filter over the input image and performing a dot product calculation with the local region of the image to extract local features), obtaining a feature map of each frame of action image of the action to be analyzed; in the multi-resolution flow layer of the pose estimation network, parallel processing is performed on the feature map of each frame of action image of the action to be analyzed (processing the input feature map through convolutional kernels of different sizes respectively to generate feature streams of different resolutions, these streams respectively contain the detailed information of the image, i.e., through the high-resolution feature stream, and the global information, i.e., through the low-resolution feature stream, and the feature streams of multiple resolutions are processed in parallel in the network to ensure that features at each scale can be extracted, the lower-resolution stream usually captures the large-scale global structure, while the higher-resolution stream focuses on the details of the image), obtaining a multi-scale feature map of each frame of action image of the action to be analyzed; in the feature fusion layer of the pose estimation network, fusion processing is performed on the multi-scale feature map of each frame of action image of the action to be analyzed (fusing feature maps of different resolutions, usually including the low-resolution global information and the high-resolution detailed information, i.e., through upsampling and convolutional layers for fusion to ensure that the sizes of the feature maps are consistent and the information is complementary, and the fused feature map passes through the ReLU activation function to increase the non-linear transformation), obtaining a multi-scale fusion feature map of each frame of action image of the action to be analyzed; in the bottleneck layer of the pose estimation network, optimized extraction processing is performed on the multi-scale fusion feature map of each frame of action image of the action to be analyzed (processing the input multi-scale fusion feature map through convolutional operations to further extract high-level features, learning the complex patterns in the image, then performing residual connection to avoid the problem of gradient disappearance, enhancing the transmission of information, then performing feature compression and expansion, adjusting the number of channels through 1x1 convolution to optimize the computing and representation capabilities, and finally performing the activation function and batch normalization to introduce non-linearity and enhance the expression ability of the network), obtaining an optimized deep feature map of each frame of action image of the action to be analyzed;In the key point detection layer of the pose estimation network, detection processing is performed on the optimized deep feature map of each frame of the action image to be analyzed (for each joint, such as the head, shoulders, knees, etc., the network generates a separate heatmap. The heatmap of each key point is usually generated by a convolutional operation. The convolutional kernel is responsible for processing the feature map and generating a new heatmap. The pixel value of each heatmap represents the probability that the position is the key point. Usually, the convolutional kernel is a 1x1 convolutional kernel used to generate the response map of each key point. The heatmap of each key point represents the probability distribution of the possible position of the key point in the image, that is, the larger the pixel value of each position, the more likely the position is the position of the key point. The heatmap is usually a two-dimensional matrix, and each pixel in it corresponds to a certain position in the image. The value of the heatmap represents the probability that the position is a certain key point. For example: heat; Figure 1 : represents the position of the head, and each pixel point in the image represents the probability that the head appears at that position), obtaining the key point heatmap set of each frame of the action image to be analyzed (including multiple key point heatmaps, and each key point will have a corresponding heatmap); in the output layer of the pose estimation network, prediction processing is performed on the key point heatmap set of each frame of the action image to be analyzed (using argmax to find the coordinates of the maximum response point in the heatmap, that is, returning the position index of the maximum value in the heatmap. This position index is the predicted key point position, and further improving the coordinate accuracy through interpolation or fitting techniques, and combining non-maximum suppression to remove redundant key point predictions to ensure the uniqueness of each key point position, and finally outputting the precise coordinates of each key point), obtaining the key point set of each frame of the action image to be analyzed, that is, the two-dimensional coordinates of several key points (key points include but are not limited to the left shoulder, left shoulder, left elbow, left wrist, left hip, left knee, left ankle, left toe, right shoulder, right shoulder, right elbow, right wrist, right hip, right left knee, right ankle, right toe, etc.).

[0038] Among them, the pre-training process of the pose estimation network is as follows:

[0039] Obtain an annotated image set, including several annotated images, and each image is annotated with the positions of key points (such as: head, shoulders, elbows, knees, ankles, etc.), and divide the annotated image set into a key point training set and a key point validation set.

[0040] And train based on the key point training set: Set the number of training loops, and in each training loop, update the weights based on the backpropagation algorithm.

[0041] The processing steps of each training loop:

[0042] Input training images: Randomly select a batch of images from the key point training set for training, and each time the input for training is the image and the two-dimensional coordinates of its corresponding key points.

[0043] Forward propagation: Input the training images into the network and perform calculations through each layer of the network to finally generate the heatmaps of the key points.

[0044] Calculate the loss: Calculate the loss (e.g., mean squared error) between the predicted heatmap and the ground truth heatmap.

[0045] Backward propagation: Update the weights of the network according to the loss through the backpropagation algorithm. The backpropagation algorithm is based on the gradient descent method to optimize the network parameters.

[0046] Optimization algorithms: Commonly used optimization algorithms include Adam, SGD (Stochastic Gradient Descent), etc. The Adam optimizer can usually better handle the sparse gradient problem.

[0047] After each training loop, perform evaluation and analysis based on the key point validation set: Validation metrics: Commonly used evaluation metrics include PCK (Percentage of Correct Keypoints), AP (Average Precision), etc., which are used to measure the prediction accuracy of the network for the positions of the key points, and perform forward propagation using the images in the validation set to calculate the loss value between the predicted heatmap and the ground truth heatmap.

[0048] Evaluation metrics: Calculate the loss function value and accuracy of the model on the validation set to evaluate the performance of the model.

[0049] Parameter adjustment: According to the evaluation results, if the loss on the validation set does not decrease or the accuracy does not improve, it may be necessary to adjust the hyperparameters of the network (such as the learning rate, network structure, etc.).

[0050] Adjust the model parameters based on the evaluation and analysis:

[0051] If the performance on the validation set is poor, hyperparameters such as the learning rate, batch size, and number of convolutional layers can be adjusted to improve the training effect.

[0052] Use the early stopping technique. If the loss on the validation set does not improve significantly within several training epochs, stop the training to prevent overfitting.

[0053] Adjust the network structure: According to the training and validation results, the structure of the model such as the number of layers and the size of the convolutional kernels can be adjusted to further optimize the contour extraction effect.

[0054] When a given input image is provided, the network will output the two-dimensional coordinates of each key point, and the training process ends to obtain a trained network model.

[0055] In this implementation scheme, by adopting a multi-resolution flow layer and a key-point detection layer, different levels of features, including global information and detail information, can be accurately extracted from images. As a result, when the model processes complex motions and subtle actions, it can better locate each key point and provide more accurate motion analysis. Secondly, the pose estimation network processes image features of different scales through the multi-resolution flow layer, ensuring that in different motion cycles, whether it is a large-scale global motion or a subtle joint action, it can be fully captured and processed, thereby enhancing the robustness of the model and enabling it to perform excellently in dynamic and complex motions. Finally, in the key-point detection stage, the heatmaps generated by the network can intuitively reflect the probability distribution of each joint point. Then, the exact position of each key point is obtained through argmax, and the coordinate accuracy is improved using interpolation methods, significantly enhancing the real-time performance and accuracy of the prediction, thus making it applicable to motion analysis scenarios that require quick feedback.

[0056] Specifically, as Figure 2 shown, the periodic key-point set is specifically the two-dimensional coordinates of each key point in each frame of the action image. The specific steps to obtain the periodic key-point set of each action cycle of the action to be analyzed are as follows: Read the two-dimensional coordinates of each key point in the key-point set of each frame of the action image to be analyzed and perform comprehensive analysis (i.e., calculate the distance between corresponding key points of adjacent-frame action images based on the Euclidean distance), obtaining the matching similarity indices of several groups of adjacent-frame action images of the action to be analyzed; and perform judgment and matching processing based on the matching similarity indices of each group of adjacent-frame action images of the action to be analyzed, obtaining the two-dimensional coordinates of each key point of several frames of action images of each action cycle of the action to be analyzed, that is, the periodic key-point set.

[0057] Among them, the specific steps of the judgment and matching processing are as follows: Respectively perform judgment and analysis on the matching similarity indices of each group of adjacent-frame action images of the action to be analyzed with a preset matching similarity index; if the matching similarity index of each group of adjacent-frame action images of the action to be analyzed is higher than the preset matching similarity index, mark it as the end of the prediction cycle, and if the matching similarity index of the next group of adjacent-frame action images of this group of adjacent-frame action images is lower than the preset matching similarity index, then mark the common frame in these two groups as the end frame of the first cycle, and mark the next frame of the end frame of the cycle as the start frame of the next cycle (the first frame is the start frame of the first cycle), then mark the start frame of the first cycle, the end frame of the first cycle, and the frames between them as the first action cycle, and so on, obtaining several action cycles.

[0058] In this implementation, by calculating the Euclidean distance of key points between adjacent frames, it is possible to accurately judge the similarity between each frame of action image and the previous frame, and then accurately divide the cycle of the action. As a result, the definition of the action cycle is more scientific, and the detailed changes within each action cycle can be captured, avoiding errors and data loss. Secondly, by matching the similarity index to judge the end and start of the cycle, the cycle division can be dynamically adjusted according to the actual performance of the action. In this way, for actions of different types or complexities, the appropriate cycle start and end points can be flexibly identified, thus adapting to various training and analysis scenarios and avoiding the inaccuracy caused by fixed cycle division. Finally, by marking the start frame and end frame of each cycle and dividing the action data into multiple small cycles accordingly, it helps to analyze each stage of the action more meticulously, and the key point data of each cycle can be analyzed separately to obtain higher accuracy. Especially in complex actions or rapidly changing movements, the changes and abnormalities in action details can be accurately identified. At the same time, by analyzing the matching similarity index of adjacent frames, highly consistent action stages can be effectively screened out, reducing the errors caused by action frames with low similarity, thus accurately judging the boundaries of the action cycle and improving the overall accuracy of action analysis, which is especially suitable for detailed motion assessment and technical improvement.

[0059] Specifically, the specific steps to obtain the initial action matching degree index of the action to be analyzed are as follows: Read the two-dimensional coordinates of each key point of each frame of the action image of each action cycle of the action to be analyzed, and perform matching coordinate conversion processing in combination with the two-dimensional coordinates of each pixel point and the corresponding depth information value to obtain the three-dimensional coordinates of each key point of each frame of the action image of each action cycle of the action to be analyzed; comprehensively analyze the three-dimensional coordinates of each key point of each frame of the action image of each action cycle of the action to be analyzed and the reference three-dimensional coordinates of each reference key point of each frame of the reference action image of each reference action cycle stored in the standard database to obtain the joint trajectory consistency index of the action to be analyzed; and perform interpolation correspondence processing on the three-dimensional coordinates of each key point of each frame of the action image of each action cycle of the action to be analyzed and the reference three-dimensional coordinates of each reference key point of each frame of the reference action image of each reference action cycle stored in the standard database (that is, determine the timestamps, frame numbers, and time intervals of each frame in the action to be analyzed and the standard database are inconsistent, so time alignment is required. For each time step, that is, each frame, use the interpolation method to calculate the three-dimensional coordinates of each key point in the analyzed action and the standard database. Even if their frame numbers are different, after interpolation, the three-dimensional coordinates of each key point in the action to be analyzed and each frame in the standard database will be aligned to ensure that they can be compared), to obtain the three-dimensional coordinates of each corresponding key point of each frame of the corresponding action image of each corresponding action cycle stored in the standard database for the action to be analyzed; and comprehensively analyze the three-dimensional coordinates of each corresponding key point of each frame of the corresponding action image of each corresponding action cycle stored in the standard database for the action to be analyzed to obtain the action angle deviation index of the action to be analyzed, and comprehensively analyze it in combination with the joint trajectory consistency index to obtain the initial action matching degree index of the action to be analyzed.

[0060] Among them, the reference three-dimensional coordinates of each reference key point of each frame of the reference action image of each reference action cycle stored in the standard database are obtained through the standard video stream data stored in the standard database (the same logic as obtaining the three-dimensional coordinates of each key point of each frame of the action image of each action cycle).

[0061] The specific matching coordinate conversion processing is as follows: Based on the two-dimensional coordinates of each key point of each frame of the action image of each action cycle of the action to be analyzed and the two-dimensional coordinates of each pixel point, perform matching processing to obtain the pixel points corresponding to each key point of each frame of the action image of each action cycle of the action to be analyzed, and perform coordinate conversion processing in combination with the depth information value corresponding to the pixel point (use the internal parameters of the acquisition device, such as focal length, principal point coordinates, etc. to convert the two-dimensional coordinates of the key point to the acquisition coordinate system, that is, the coordinate system of the acquisition device, and through the corresponding depth information value, convert the coordinates in the acquisition coordinate system to the world coordinate system to obtain the three-dimensional coordinates), to obtain the three-dimensional coordinates of each key point of each frame of the action image of each action cycle of the action to be analyzed.

[0062] The specific steps for the joint trajectory consistency index are as follows: comprehensively analyze the three-dimensional coordinates of each key point in each frame of the action image for each action cycle of the action to be analyzed (calculate the distance between adjacent key points based on the Euclidean distance formula and perform summation processing), obtain the trajectory distance value of the action to be analyzed, and comprehensively analyze it in combination with the reference trajectory distance stored in the standard database (with the same acquisition logic as the trajectory distance). The specific formula for the joint trajectory consistency index of the action to be analyzed is as follows: Among them, GyZ is the joint trajectory consistency index of the action to be analyzed, CkL is the reference trajectory distance value stored in the standard database, and GjL is the trajectory distance value of the action to be analyzed.

[0063] The specific steps for the action angle deviation index are as follows: comprehensively analyze the three-dimensional coordinates of each corresponding key point in each frame of the corresponding action image for each corresponding action cycle stored in the standard database and the action to be analyzed, obtain several corresponding pose angles of the action to be analyzed and each frame of the corresponding action image for each corresponding action cycle stored in the standard database (including but not limited to the angles between the left shoulder, left elbow, and left wrist, the angles between the right shoulder, right elbow, and right wrist, the angles between the head, left shoulder, and left hip, the angles between the head, right shoulder, and right hip, etc.), and comprehensively analyze it in combination with each reference pose angle in each frame of the reference action image for each reference action cycle stored in the standard database (with the same acquisition logic as the pose angle). The specific formula for the action angle deviation index of the action to be analyzed is as follows: Among them, DzP is the action angle deviation index of the action to be analyzed, CtJ abr is the r-th reference pose angle of the b-th frame of the reference action image of the a-th reference action cycle stored in the standard database, and ZjT imn is the n-th pose angle of the m-th frame of the corresponding action image of the i-th corresponding action cycle of the action to be analyzed, where a = 1, 2, 3,..., a0, a0 is the number of reference action cycles, b = 1, 2, 3,..., b0, b0 is the number of frames of the reference action image, r

[0064] = 1, 2, 3,..., r0, r0 is the number of reference pose angles, i = 1, 2, 3,..., i0, i0 is the number of corresponding action cycles, m = 1, 2, 3,..., m0, m0 is the number of frames of the corresponding action image, n = 1, 2, 3,..., n0, n0 is the number of pose angles, and a0 = i0, corresponding one by one, b0 = m0, corresponding one by one, r0 = n0, corresponding one by one.

[0065] For example, if the angle between the left shoulder, left elbow, and left wrist is the first pose angle, the vector dot product formula and the three-dimensional coordinates of the three key points of the left shoulder, left elbow, and left wrist can be used for calculation.

[0066] The specific formula for calculating the initial action fitness index of the action to be analyzed is as follows: Among them, CsQ is the initial action fitness index of the action to be analyzed, GyZ is the joint trajectory consistency index of the action to be analyzed, μ1 is the joint trajectory adjustment coefficient stored in the database, DzP is the action angle deviation index of the action to be analyzed, μ2 is the action angle adjustment coefficient stored in the database, and μ3 is the sensitivity adjustment coefficient stored in the database.

[0067] It should be explained that the expression form of the Tanh function is and its value range is (-1, 1), and the domain is (-∞, +∞).

[0068] The term 1 - Tanh(μ3 * GyZ * DzP) in the formula is used to adjust the superposition effect of the joint trajectory consistency index and the action angle deviation index, and avoid the initial action fitness index being too high or too low.

[0069] μ1, μ2, and μ3 can be obtained through the following steps: Using historical data, combined with the joint trajectory consistency index and the action angle deviation index, perform statistical regression analysis to quantify the specific impact of each factor on the initial action fitness index, so as to fit the initial weight value. Secondly, use the sensitivity analysis method to adjust the value range of each coefficient and observe its impact on the initial action fitness evaluation result to ensure the stability and rationality of the model. Based on the actual situation, correct and optimize the initially fitted coefficients.

[0070] In this implementation, by converting two-dimensional coordinates into three-dimensional coordinates and comparing them with the reference data in the standard database, the accuracy of motion analysis can be greatly improved. Moreover, three-dimensional analysis can more accurately restore the spatial performance of human motions, thereby avoiding errors caused by factors such as angles and perspectives. Secondly, by calculating the joint trajectory consistency index and the motion angle deviation index between the motion to be analyzed and the standard motion, the system can accurately evaluate the differences between the motion to be analyzed and the reference motion, which helps to improve the technique of the motion and ensure the correctness and efficiency of the motion. At the same time, since the number of frames and timestamps of different motion images may vary, the coordinates of key points of each frame of the motion to be analyzed and the standard database are time-aligned through interpolation, thus ensuring the synchronization between data and avoiding data mismatch problems caused by inconsistent frame rates. Furthermore, it ensures that the matching and comparison of motion cycles are based on accurate time series. Finally, the comprehensive analysis of joint trajectory consistency and motion angle deviation can provide a comprehensive motion assessment, thus more accurately describing the overall performance of the motion, especially in complex or delicate motion analysis. By using the reference motions and pose angles in the standard database, the analysis process is highly standardized and automated, which reduces the subjective bias of human judgment and ensures the consistency and repeatability in large-scale data analysis.

[0071] Specifically, the motion correction data includes the arm length value (the length when the two arms are extended), the leg length value, the environmental factor, the muscle tension distribution index, and the training load index. The specific steps to obtain the motion correction index of the motion to be analyzed are as follows: Obtain the reference arm length value, reference leg length value, reference environmental factor, reference muscle tension distribution index, and reference training load index stored in the standard database; comprehensively analyze the arm length value, leg length value, environmental factor, muscle tension distribution index, and training load index of the motion to be analyzed and the reference arm length value, reference leg length value, reference environmental factor, reference muscle tension distribution index, and reference training load index stored in the standard database to obtain the motion correction index of the motion to be analyzed.

[0072] Among them, the environmental factor is the comprehensive influence of the external environment on the athlete during the athlete's training process. It can be obtained by acquiring the temperature value (acquired by a temperature sensor), humidity value (acquired by a humidity sensor), wind speed value (acquired by an anemometer), and light intensity value (acquired by a light sensor) during the athlete's training process, and performing standardization processing. Based on the results of the standardization processing, weighted processing is performed, and the obtained result is this parameter.

[0073] The muscle tension distribution index is the tension distribution between different muscle groups (such as the anterior and posterior thigh muscles, shoulder and chest muscles, etc.) during exercise. It can measure the tension of each muscle group through electromyography and perform standard deviation processing, and the obtained result is this parameter.

[0074] The training load index measures the physiological load of athletes during training. It can be obtained by acquiring the speed value of athletes during training (obtained through a speed sensor), muscle fatigue (measuring the change in electrical activity of muscles during exercise through electromyography. Usually, fatigued muscles show a decrease in electrical activity frequency and amplitude), heart rate value (obtained through photoplethysmography in a smart bracelet), and then performing standardization processing. Based on the results of the standardization processing, weighted processing is carried out, and the resulting value is this parameter.

[0075] The reference arm length value is the arm length value of the athlete deduced from the video stream of the standard database.

[0076] The reference leg length value is the leg length value of the athlete deduced from the video stream of the standard database.

[0077] The reference environmental factor is the general impact of the environment where the athlete is located deduced from the video stream of the standard database, and its acquisition logic is consistent with that of the environmental factor.

[0078] The reference muscle tension distribution index is the tension distribution among different muscle groups of the athlete deduced from the video stream of the standard database, and its acquisition logic is consistent with that of the muscle tension distribution index.

[0079] The reference training load index is the physiological load of the athlete during training deduced from the video stream of the standard database, and its acquisition logic is consistent with that of the training load index.

[0080] The specific formula for calculating the action correction index of the action to be analyzed is as follows: Among them, DzX is the action correction index of the action to be analyzed, is the interaction coefficient stored in the database, CzH is the arm length value of the action to be analyzed, HcZ is the reference arm length value stored in the standard database, is the arm length adjustment coefficient stored in the database, TcZ is the leg length value of the action to be analyzed, ZtC is the reference leg length value stored in the standard database, is the leg length adjustment coefficient stored in the database, HjY is the environmental factor of the action to be analyzed, JhY is the reference environmental factor stored in the standard database, is the environmental adjustment coefficient stored in the database, JfZ is the muscle tension distribution index of the action to be analyzed, CjF is the reference muscle tension distribution index stored in the standard database, is the muscle tension adjustment coefficient stored in the database, XhL is the training load index of the action to be analyzed, ChX is the reference training load index stored in the standard database, is the training load adjustment coefficient stored in the database.

[0081] It should be explained that It can be obtained through the following steps: Based on historical data, determine the initial influence weights of each variable (arm length value, leg length value, environmental factor, muscle tension distribution index, training load index) on the motion correction index through statistical regression analysis. Then, use the sensitivity analysis method to adjust the value range of the coefficients to evaluate the stability and applicability of these parameters to the formula output. Next, further fit the weights through model optimization (such as machine learning algorithms or multi-objective optimization) to ensure that the formula can accurately reflect the correction of the motion.

[0082] A specific implementation example of calculating the motion correction index of the motion to be analyzed is as follows. The following data is available: including arm length value (m), leg length value (m), environmental factor, muscle tension distribution index (μV), training load index. The specific data is shown in Tables 1 and 2:

[0083] Table 1 Example of motion correction data for the motion to be analyzed

[0084]

[0085] Table 2 Example of motion correction data stored in the standard database

[0086]

[0087]

[0088] Interaction coefficient stored in the database Approximately: 0.247;

[0089] Arm length adjustment coefficient stored in the database Approximately: 0.452;

[0090] Leg length adjustment coefficient stored in the database Approximately: 0.367;

[0091] Environmental adjustment coefficient stored in the database Approximately: 0.481;

[0092] Muscle tension adjustment coefficient stored in the database Approximately: 0.643;

[0093] Training load adjustment coefficient stored in the database Is approximately: 0.275;

[0094] Substitute Table 1, Table 2 and the above coefficients into the specific formula for calculating the motion correction index of the motion to be analyzed, and obtain:

[0095] Motion correction index of the motion to be analyzed ≈ 0.716.

[0096] In this implementation scheme, by integrating individual characteristics such as arm length, leg length, environmental factors, muscle tension distribution, and training load, it is possible to correct movements according to the unique physiological and environmental conditions of each athlete, making movement analysis and evaluation more personalized, thus accurately reflecting the true state of the athlete, helping the athlete optimize movements according to their own situation, and then avoiding errors caused by external conditions or physiological differences. Secondly, environmental factors (such as temperature, humidity, wind speed, light) and physiological loads (such as muscle tension distribution, training load, etc.) all have a direct impact on the performance of athletes. Therefore, by comprehensively considering these factors, the movement performance of athletes can be corrected more comprehensively, and the deviation caused by ignoring these external and physiological factors can be avoided. At the same time, by refining the weights and adjustment coefficients of each factor (such as arm length adjustment coefficient, leg length adjustment coefficient, etc.), the movement to be analyzed can be accurately corrected, and sensitivity analysis and model optimization help evaluate the impact of each parameter on the movement correction index, so that the formula can be optimized according to historical data and actual situations, further improving the correction accuracy. Finally, using the reference data in the standard database for correction can avoid errors caused by individual differences between different athletes, ensuring the universality and consistency of movement correction.

[0097] Specifically, the specific steps for training management based on the comprehensive movement fit index of the movement to be analyzed are as follows: Compare and analyze the comprehensive movement fit index of the movement to be analyzed with the preset comprehensive movement fit index threshold; if the comprehensive movement fit index of the movement to be analyzed is lower than the preset comprehensive movement fit index threshold, take the first training management measure (that is, provide feedback to the athlete or coach, point out the deficiencies in the current movement, such as inconsistent movement trajectories, large angle deviations, and adjust the training plan and give suggestions for guided correction); if the comprehensive movement fit index of the movement to be analyzed is not lower than the preset comprehensive movement fit index threshold, take the second training management measure (that is, provide positive feedback to the athlete or coach, encourage them to maintain the current training progress, reasonably increase the exercise intensity and endurance training, and give suggestions for setting new training goals).

[0098] In this implementation, by providing feedback and making adjustments based on the comprehensive movement fit index of athletes, training management can be made more personalized. And personalized training management can help athletes focus on their weaknesses and areas for improvement, thereby enhancing the training effect. For example, for athletes below the threshold, their movement accuracy can be improved, while for those who have reached the standard, their physical fitness and endurance can be further enhanced. Secondly, by analyzing the movement fit of athletes in real time and providing feedback, training management can flexibly respond to the performance of athletes. And through timely feedback, coaches can adjust the training plan at any time, without having to wait for the results of long-term training to make improvements, thus effectively avoiding the lag in progress or technical bottlenecks caused by inappropriate training methods. Finally, the comprehensive movement fit index can clearly and quantitatively display the progress in training, enabling both athletes and coaches to intuitively see the results of their efforts, thereby providing a clear direction for training and also helping to motivate athletes, boost their confidence, and stimulate their training motivation.

[0099] Please refer to Figure 3 , an embodiment of the present invention provides a technical solution: an action data analysis system based on intelligent sports, including: a data acquisition module for acquiring video stream data of an action to be analyzed, where the video stream data includes a number of frame action image data, that is, the pixel value, two-dimensional coordinates, and corresponding depth information value of each pixel point in the action image; a prediction and division module for inputting the video stream data of the action to be analyzed into a pre-trained key point recognition model for prediction analysis to obtain the key point set of each frame of the action image of the action to be analyzed, and performing division processing to obtain the cycle key point set of each action cycle of the action to be analyzed; a data analysis module for performing data analysis on the cycle key point set of each action cycle of the action to be analyzed to obtain the initial action fit index of the action to be analyzed; a comprehensive correction module for obtaining the action correction data of the action to be analyzed, analyzing to obtain the action correction index of the action to be analyzed, and performing comprehensive analysis in combination with the initial action fit index to obtain the comprehensive action fit index of the action to be analyzed; a training management module for performing training management based on the comprehensive action fit index of the action to be analyzed.

[0100] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0101] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A method for analyzing motion data based on intelligent sports, characterized in that, Including the following steps: Obtain the video stream data of the action to be analyzed, where the video stream data includes several frames of action image data, that is, the pixel value, two-dimensional coordinates, and corresponding depth information value of each pixel point in the action image; Input the video stream data of the action to be analyzed into a pre-trained key point recognition model for prediction and analysis, obtain the key point set of each frame of the action image of the action to be analyzed, and perform division processing to obtain the periodic key point set of each action cycle of the action to be analyzed; And perform data analysis on the periodic key point set of each action cycle of the action to be analyzed to obtain the initial action matching degree index of the action to be analyzed; Obtain the action correction data of the action to be analyzed, analyze to obtain the action correction index of the action to be analyzed, and perform comprehensive analysis in combination with the initial action matching degree index to obtain the comprehensive action matching degree index of the action to be analyzed. The specific formula is as follows: Where ZhQ, CsQ, and DzX are the comprehensive action matching degree index, initial action matching degree index, and action correction index of the action to be analyzed in sequence, and α1, β1, α2, β2, and β3 are the initial coefficient, initial adjustment coefficient, correction coefficient, correction adjustment coefficient, and superposition coefficient stored in the database in sequence, and α1 + α2 = 1; And perform training management based on the comprehensive action matching degree index of the action to be analyzed.

2. The method for analyzing motion data based on intelligent sports according to claim 1, wherein The key point recognition model is specifically a pose estimation network, and the pose estimation network includes an input layer for receiving video stream data, an initial convolution layer, a multi-resolution flow layer, a feature fusion layer, a bottleneck layer, a key point detection layer, and an output layer.

3. The method for analyzing motion data based on intelligent sports according to claim 2, wherein, The specific steps to obtain the key point set of each frame of the action image of the action to be analyzed are as follows: In the input layer of the pose estimation network, receive several frames of action image data of the action to be analyzed and perform preprocessing; In the initial convolution layer of the pose estimation network, perform feature extraction processing on each frame of the preprocessed action image data of the action to be analyzed to obtain the feature map of each frame of the action image of the action to be analyzed; In the multi-resolution flow layer of the pose estimation network, perform parallel generation processing on the feature map of each frame of the action image of the action to be analyzed to obtain the multi-scale feature map of each frame of the action image of the action to be analyzed; In the feature fusion layer of the pose estimation network, perform fusion processing on the multi-scale feature map of each frame of the action image of the action to be analyzed to obtain the multi-scale fusion feature map of each frame of the action image of the action to be analyzed; In the bottleneck layer of the pose estimation network, perform optimized extraction processing on the multi-scale fusion feature map of each frame of the action image of the action to be analyzed to obtain the optimized deep feature map of each frame of the action image of the action to be analyzed; In the key point detection layer of the pose estimation network, perform detection processing on the optimized deep feature map of each frame of the action image of the action to be analyzed to obtain the key point heat map set of each frame of the action image of the action to be analyzed; In the output layer of the pose estimation network, perform prediction processing on the key point heat map set of each frame of the action image to obtain the key point set of each frame of the action image of the action to be analyzed, that is, the two-dimensional coordinates of several key points.

4. The method for analyzing motion data based on intelligent sports according to claim 1, wherein The specific cycle key point set is the two-dimensional coordinates of each key point in each frame of the action image. The specific steps to obtain the cycle key point set of each action cycle of the action to be analyzed are as follows: Read the two-dimensional coordinates of each key point in the key point set of each frame of the action image to be analyzed, and conduct comprehensive analysis to obtain the matching similarity indices of several groups of adjacent frame action images of the action to be analyzed; And perform judgment and matching processing based on the matching similarity indices of each group of adjacent frame action images of the action to be analyzed to obtain the two-dimensional coordinates of each key point of several frames of action images of each action cycle of the action to be analyzed, that is, the cycle key point set.

5. The method for analyzing motion data based on intelligent sports according to claim 4, wherein The specific steps to obtain the initial action fitness index of the action to be analyzed are as follows: Read the two-dimensional coordinates of each key point in each frame of the action image of each action cycle of the action to be analyzed, and perform matching coordinate conversion processing in combination with the two-dimensional coordinates of each pixel point and the corresponding depth information value to obtain the three-dimensional coordinates of each key point in each frame of the action image of each action cycle of the action to be analyzed; Conduct comprehensive analysis on the three-dimensional coordinates of each key point in each frame of the action image of each action cycle of the action to be analyzed and the reference three-dimensional coordinates of each reference key point in each frame of the reference action image of each reference action cycle stored in the standard database to obtain the joint trajectory consistency index of the action to be analyzed; And perform interpolation and corresponding processing on the three-dimensional coordinates of each key point in each frame of the action image of each action cycle of the action to be analyzed and the reference three-dimensional coordinates of each reference key point in each frame of the reference action image of each reference action cycle stored in the standard database to obtain the three-dimensional coordinates of each corresponding key point in each frame of the corresponding action image of each corresponding action cycle stored in the standard database for the action to be analyzed; And conduct comprehensive analysis on the three-dimensional coordinates of each corresponding key point in each frame of the corresponding action image of each corresponding action cycle stored in the standard database for the action to be analyzed to obtain the action angle deviation index of the action to be analyzed, and conduct comprehensive analysis in combination with the joint trajectory consistency index to obtain the initial action fitness index of the action to be analyzed.

6. The method for analyzing motion data based on intelligent sports according to claim 5, wherein, The specific formula for calculating the initial action fitness index of the action to be analyzed is as follows: Among them, CsQ, GyZ, and DzP are the initial action fitness index, joint trajectory consistency index, and action angle deviation index of the action to be analyzed respectively, and μ1, μ2, and μ3 are the joint trajectory adjustment coefficient, action angle adjustment coefficient, and sensitivity adjustment coefficient stored in the database respectively.

7. The method for analyzing motion data based on intelligent sports according to claim 1, wherein The action correction data includes arm length value, leg length value, environmental factor, muscle tension distribution index, and training load index. The specific steps to obtain the action correction index of the action to be analyzed are as follows: Obtain the reference arm length value, reference leg length value, reference environmental factor, reference muscle tension distribution index, and reference training load index stored in the standard database; Conduct comprehensive analysis on the arm length value, leg length value, environmental factor, muscle tension distribution index, and training load index of the action to be analyzed and the reference arm length value, reference leg length value, reference environmental factor, reference muscle tension distribution index, and reference training load index stored in the standard database to obtain the action correction index of the action to be analyzed.

8. The method for analyzing motion data based on intelligent sports according to claim 7, wherein The specific formula for calculating the action correction index of the action to be analyzed is as follows: Among them, DzX is the action correction index of the action to be analyzed, and CzH, TcZ, HjY, JfZ, and XhL are the arm length value, leg length value, environmental factor, muscle tension distribution index, and training load index of the action to be analyzed in sequence. HcZ, TcZ, JhY, CjF, and ChX are the reference arm length value, reference leg length value, reference environmental factor, reference muscle tension distribution index, and reference training load index stored in the standard database, which are the interaction coefficient, arm length adjustment coefficient, leg length adjustment coefficient, environmental adjustment coefficient, muscle tension adjustment coefficient, and training load adjustment coefficient stored in the database in sequence.

9. The method for analyzing motion data based on intelligent sports according to claim 1, wherein The specific steps for training management based on the comprehensive action fit index of the action to be analyzed are as follows: Compare and analyze the comprehensive action fit index of the action with the preset comprehensive action fit index threshold; If the comprehensive action fit index of the action to be analyzed is lower than the preset comprehensive action fit index threshold, take the first training management measure; If the comprehensive action fit index of the action to be analyzed is not lower than the preset comprehensive action fit index threshold, take the second training management measure.

10. An action data analysis system based on intelligent sports, applying the method for analyzing action data based on intelligent sports according to any one of claims 1-9, characterized in that, Including: A data acquisition module, which is used to acquire the video stream data of the action to be analyzed. The video stream data includes a number of frame action image data, that is, the pixel value, two-dimensional coordinate and corresponding depth information value of each pixel point in the action image; A prediction and division module, which is used to input the video stream data of the action to be analyzed into a pre-trained key point recognition model for prediction and analysis, obtain the key point set of each frame of action image of the action to be analyzed, and perform division processing to obtain the cycle key point set of each action cycle of the action to be analyzed; A data analysis module, which is used to perform data analysis on the cycle key point set of each action cycle of the action to be analyzed to obtain the initial action fit index of the action to be analyzed; A comprehensive correction module, which is used to obtain the action correction data of the action to be analyzed, analyze and obtain the action correction index of the action to be analyzed, and perform comprehensive analysis in combination with the initial action fit index to obtain the comprehensive action fit index of the action to be analyzed; A training management module, which is used to perform training management based on the comprehensive action fit index of the action to be analyzed.

Images