Swimming action-based attitude analysis method and device

Abstract

The invention provides a posture analysis method and device based on swimming actions, and the method comprises the steps: obtaining the data of multiple parts of a sporter during swimming through a nine-axis sensor, and converting the data to a three-dimensional coordinate system in a unified manner; the method comprises the following steps: performing segmentation and feature extraction on data according to motion stages of entering water, paddling, exiting water, breathing and turning around, calculating a three-dimensional motion track of a key point through an attitude deviation model, and projecting the three-dimensional motion track to a two-dimensional plane to generate an actual projection track. And in combination with standard attitude data, generating a theoretical trajectory through a clustering algorithm and calculating a trajectory difference to form an attitude deviation index. The posture deviation indexes of all the action stages are combined and input into a comprehensive analysis model, the balance, synchronism and streamline of the swimming posture are evaluated, and correction suggestions are generated. Therefore, swimming action postures can be accurately recognized and analyzed, scientific correction suggestions are provided for athletes, and the training effect and the exercise performance are improved.

Description

Technical Field

[0001] The present invention relates to the field of sports, and in particular to a method and device for analyzing swimming movements and postures. Background Art

[0002] In swimming, it is crucial for athletes to maintain the correct posture. The correct swimming posture can not only improve the athlete's movement efficiency, reduce resistance in the water, increase speed and stability, but also prevent muscle strain and joint injuries, ensuring the long-term health of athletes. However, the adjustment and correction of swimming posture is often a complex and difficult process.

[0003] Traditionally, swimmers rely on coach guidance and video playback to correct their swimming posture. However, this method has the following defects. First, unlike other sports, athletes are in the water during training and cannot hear the coach's instructions in real time. They can only rely on muscle memory to complete the movements. Second, the coach's guidance and correction opinions rely on their personal experience and observation ability, which is prone to subjective bias and cannot guarantee that posture problems can be accurately and comprehensively found every time. In addition, traditional methods cannot provide real-time feedback on posture problems during swimming. Athletes can only reflect and correct through video playback after training, which greatly reduces the training efficiency, and the cost of using video playback is high. Furthermore, it is difficult to capture the subtle posture changes of athletes by naked eye observation and video analysis alone, especially when swimming at high speed, the details of the movement are difficult to accurately judge, resulting in limited correction effect. Finally, traditional methods lack quantitative analysis of swimming posture, cannot provide data-supported correction suggestions, and it is difficult for athletes to clearly determine the direction and effect of improvement. These problems jointly limit the accuracy and efficiency of swimming posture correction. Summary of the invention

[0004] In view of the above actual situation, the present invention proposes a swimming movement posture analysis method and device to solve the problems in the prior art that the coach's guidance is easily affected by subjectivity, posture problems cannot be fed back in real time, the cost of correcting posture is high, and there is a lack of accurate capture and quantitative analysis of subtle posture changes.

[0005] A swimming posture analysis method, characterized in that it comprises the following steps: S1, obtaining data to be processed, wherein the data to be processed is data of the head, shoulders, arms, waist and legs of a swimmer during exercise, and the data to be processed is obtained through a nine-axis sensor; S2, performing data conversion on the data to be processed, thereby obtaining conversion data, wherein the conversion data is coordinate data obtained by uniformly converting the data to be processed into a three-dimensional coordinate system; S3, combining the converted data with the data to be processed, and segmenting the data according to the action phases, thereby obtaining segmented data, wherein the action phases include entry, paddling, exit, breathing, and turning phases; S4, extracting features from the segmented data and performing action feature stratification to obtain feature stratified data; S5, calling the posture deviation model, inputting the feature layered data into the posture deviation model, thereby obtaining the three-dimensional motion trajectory data of the key point; based on the three-dimensional motion trajectory data of the key point, projecting onto three mutually perpendicular two-dimensional planes, obtaining the actual projection trajectory; generating a theoretical trajectory according to the actual projection trajectory of the key point on the two-dimensional plane and the pre-collected standard posture data, and calculating the difference between the actual trajectory and the theoretical trajectory, and forming a posture deviation index by combining the trajectory difference of the key point on the two-dimensional plane; S6, merging the posture deviation indicators of each action stage to form an overall coherent posture deviation indicator; S7, calling a comprehensive analysis model, inputting the overall coherent posture deviation index into the comprehensive analysis model, analyzing the balance, synchronization, and streamline, thereby evaluating the balance, synchronization, and streamline of the swimming posture and obtaining an evaluation result; Step S8, issuing a warning and generating swimming posture correction suggestions based on the balance, synchronization and streamline evaluation results.

[0006] Furthermore, in the step S2, the coordinate data is obtained by converting the data to be processed into a global coordinate system through a rotation matrix, and the rotation matrix is ,in are the rotation angles around the x, y, and z axes respectively. The nine-axis nine-axis sensor includes a gyroscope. The rotation angle is obtained by integrating the angular velocity of the gyroscope. The specific calculation formula is as follows: , , .

[0007] Furthermore, the S3 step includes the following sub-steps: S31, calculating motion state parameters of the data to be processed and the global coordinate data through a sliding window, wherein the motion state parameters include the mean and standard deviation of the acceleration, angular velocity, magnetometer and global coordinate data, which are used to describe the motion situation in each time window; S32, constructing a state vector based on the motion state parameters, wherein the state vector includes the mean and standard deviation of the acceleration, angular velocity, magnetometer and global coordinate data in each time window, which is used to comprehensively reflect the motion situation; S33, based on the state vector, classification is performed by a classification algorithm to obtain an action stage label in each sliding window, wherein the classification algorithm is a support vector machine; S34, based on the action stage label, the original data is segmented to generate segmented data of the corresponding stage, and the original data includes the data to be processed and the global coordinate data.

[0008] Furthermore, the S4 step is to extract and stratify the features of the segmented data through a deep learning network, wherein the deep learning network includes a convolutional layer and an LSTM layer, wherein the convolutional layer extracts local spatiotemporal features, and the LSTM layer is used to capture the temporal features of the data.

[0009] Furthermore, the step S5 includes the following sub-steps: S51, inputting the feature hierarchical data into a posture deviation model, wherein the posture deviation model is used to calculate the motion trajectory of the key point in the three-dimensional coordinate system, thereby obtaining the three-dimensional motion trajectory data of the key point; S52, based on the three-dimensional motion trajectory data of the key point, projecting it onto three mutually perpendicular two-dimensional planes, where the two-dimensional planes are an XY plane, a YZ plane, and an XZ plane, thereby obtaining a projection trajectory of the key point on the two-dimensional plane; S53, generating a theoretical trajectory through a clustering algorithm according to the actual projection trajectory of the key point on the two-dimensional plane and the pre-acquired standard posture data, wherein the theoretical trajectory is used to characterize the trajectory of the key point under the standard motion posture; S54, calculating the trajectory difference between the two-dimensional projection trajectory and the theoretical trajectory, wherein the trajectory difference includes a trajectory deviation value and a trajectory deviation direction, which are used to quantify the motion deviation of the key point; S55, integrating the trajectory differences of each key point on all two-dimensional planes to form a posture deviation index, wherein the posture deviation index is used to evaluate the accuracy and consistency of the overall motion posture.

[0010] Furthermore, the posture deviation model is a hybrid model that integrates convolutional neural network, graph convolutional network and attention mechanism.

[0011] Furthermore, it is characterized in that the comprehensive analysis model includes a balance analysis sub-model, a synchronization analysis sub-model and a streamline analysis sub-model.

[0012] Furthermore, the specific calculation formula of the balance analysis sub-model is as follows: , where T is the total time of the entire swimming process, D balancede The specific calculation formula is as follows: .

[0013] Furthermore, the specific calculation formula of the synchronization analysis sub-model is as follows: , where N is the total number of key points, T is the total time of the entire swimming process, The specific calculation formula is as follows: , the motion phase calculation formula is as follows: ,in is the initial phase of key point k, Tk is the motion period of each key point; the specific calculation formula of the streamline analysis sub-model is as follows: , where N is the total number of key points; the smoothness index S k The calculation formula is as follows: , where T is the total time of the entire swimming process; the specific calculation formula for smoothness is as follows: ; Acceleration A k (t) The specific calculation formula is as follows: , P k (t) is the overall coherent posture deviation index P overall (t) The extracted motion trajectory.

[0014] In addition, the present invention also discloses a swimming action posture analysis device, characterized in that the device comprises: An acquisition unit, used for acquiring data to be processed, wherein the data to be processed is data of the head, shoulders, arms, waist and legs of a swimmer in motion, and the data to be processed is obtained through a nine-axis sensor; A conversion unit, used for performing data conversion on the data to be processed, so as to obtain conversion data, wherein the conversion data is coordinate data obtained by uniformly converting the data to be processed into a three-dimensional coordinate system; a segmentation unit, used to combine the converted data with the data to be processed, and perform data segmentation according to the action phases, so as to obtain segmented data, wherein the action phases include entry, paddling, exit, breathing, and turning phases; A feature extraction unit, used to extract features from the segmented data and perform action feature stratification to obtain feature stratified data; The deviation calculation unit is used to call the posture deviation model, input the feature layered data into the posture deviation model, so as to obtain the three-dimensional motion trajectory data of the key point; based on the three-dimensional motion trajectory data of the key point, the actual projection trajectory is obtained by projecting onto three mutually perpendicular two-dimensional planes; according to the actual projection trajectory of the key point on the two-dimensional plane and the pre-collected standard posture data, a theoretical trajectory is generated by a clustering algorithm, and the difference between the actual trajectory and the theoretical trajectory is calculated, and the trajectory difference of the key point on the two-dimensional plane is integrated to form a posture deviation index; A merging unit, used to merge the posture deviation indicators of each action stage to form an overall coherent posture deviation indicator; An analysis unit is used to call a comprehensive analysis model, input the overall coherent posture deviation index into the comprehensive analysis model, analyze the balance, synchronization, and streamline, thereby evaluating the balance, synchronization, and streamline of the swimming posture and obtaining an evaluation result; The suggestion generation unit issues warnings and generates swimming posture correction suggestions based on the balance, synchronization, and streamline assessment results.

[0015] The present invention proposes a method and device for analyzing swimming postures, which obtains and merges segmented posture deviation indicators by segmenting swimming action stages to form an overall coherent posture deviation indicator, which significantly solves the problems of inaccurate posture problem identification, unreal-time training feedback, difficulty in capturing subtle posture changes, and lack of quantitative analysis caused by relying on subjective experience and observation ability in the prior art, thereby achieving a comprehensive, accurate and real-time analysis of swimming postures. By inputting a comprehensive analysis model, the balance, synchronization and streamline of swimming postures can be comprehensively evaluated, and data-supported correction suggestions can be generated, which greatly improves the efficiency, effect and cost of training, and helps athletes to clarify the direction of improvement and improve their swimming skills. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a flow chart of a swimming posture analysis method proposed by the present invention; Figure 2 A schematic diagram of a data segmentation process based on a swimming action posture analysis method proposed by the present invention; Figure 3 A schematic diagram of a mid-posture deviation analysis process based on a swimming action posture analysis method proposed by the present invention; Figure 4 A schematic diagram of the structure of a swimming motion posture analysis device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The following will be combined with the accompanying drawings of the present invention to clearly and completely describe the simulation technology route in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] As swimming becomes more popular, more and more people are paying attention to how to improve swimming skills and efficiency. However, during training, swimmers often find it difficult to accurately understand the deviations and problems in their own posture, which leads to poor training results and may even cause sports injuries. Existing swimming posture analysis methods lack comprehensiveness and accuracy, and are difficult to meet swimmers' needs for professional guidance and scientific correction. To this end, this application proposes a swimming posture analysis method, please refer to the attached Figure 1 As shown, the following steps are included: Step S1, obtaining data to be processed, wherein the data to be processed is data of the head, shoulders, arms, waist, and legs of a swimmer during exercise, and the data to be processed is obtained through a nine-axis sensor; Specifically, this step aims to obtain the head, shoulder, arm, waist and leg data of the swimmer during exercise, and the data is obtained by a nine-axis sensor embedded in a swimsuit. First, a flexible circuit board is used, which is suitable for integration with swimsuit materials due to its thin and soft characteristics. The flexible circuit board can withstand bending and stretching during exercise without affecting the comfort of the athlete. Nine-axis sensors, including accelerometers, gyroscopes and magnetometers, are arranged on the flexible circuit board. The sensors capture multi-dimensional motion data and provide a basis for subsequent analysis. Next, waterproof materials are used to encapsulate the flexible circuit board and the nine-axis sensor. In one or more embodiments, the waterproof material includes epoxy resin and silica gel. The waterproof material can effectively prevent moisture from penetrating into the circuit, ensure the stability and reliability of the sensor when working in water, and meet the IP68 waterproof requirements. During the packaging process, it should be noted that the sensing part of the sensor should be kept exposed or encapsulated with a material with good water permeability to ensure that the sensor can accurately sense the external environment. Then the packaged flexible circuit board and nine-axis sensor are embedded in the fabric of the swimsuit. In one or more embodiments, the sensor is firmly fixed in a specific position of the swimsuit by stitching and hot pressing to ensure that the sensor can be close to the body of the athlete and will not move due to swimming movements. The weaving process needs to ensure the accurate position of the sensor, especially the key parts of the head, shoulders, arms, waist and legs, to ensure that the collected data accurately reflects the posture and movement of the athlete. In addition, the swimsuit also includes a wireless charging function, which is convenient for users to charge, and an integrated linear patch vibration motor is used to remind athletes in real time when there is a problem with the posture in the water. A data acquisition module is integrated on the flexible circuit board of the swimsuit, which is responsible for collecting and storing or transmitting the data of the nine-axis sensor. The data acquisition module has a wireless transmission function so that the data can be transmitted to the data processing system on the shore in real time for analysis. The data acquisition module should also have sufficient storage capacity so that when wireless transmission is not available, the data can be stored locally, and the data can be transmitted and processed after the exercise. Before formal use, the nine-axis sensor system embedded in the swimsuit is calibrated to ensure that the data of each sensor is accurate. In one or more embodiments, the calibration method includes static calibration and dynamic calibration, which adjust the static offset and dynamic response of the sensor respectively. Finally, the swimsuit is field tested to check the performance of the sensor during actual swimming to ensure that the data acquisition system can work normally in various swimming postures. The nine-axis sensor includes an accelerometer, a gyroscope and a magnetometer. The accelerometer measures the linear acceleration in three-dimensional space and can capture the acceleration changes of the athlete for analyzing the direction and strength of movement; the gyroscope measures the angular velocity in three-dimensional space and can capture the rotation and rotation of the athlete for analyzing posture changes; the magnetometer measures the magnetic field strength in three-dimensional space and can provide a direction reference for correcting the data of the accelerometer and gyroscope to improve the accuracy of posture calculation.The data acquisition module is responsible for real-time acquisition of motion data, which is processed through the embedded data acquisition module, including filtering and denoising, to ensure data quality. The data is transmitted to the data processing system on shore through wireless transmission or local storage for further analysis and processing. Waterproof materials are used to encapsulate sensors and circuits to prevent moisture from entering the internal circuits and ensure the normal operation of the system in water. The waterproof materials need to have good sealing properties and do not affect the sensing ability of the sensors. Through the above steps, the head, shoulder, arm, waist, and leg data of swimmers in motion are successfully acquired, providing a reliable data basis for subsequent posture analysis and correction.

[0020] Step S2, performing data conversion on the data to be processed, thereby obtaining conversion data, wherein the conversion data is coordinate data obtained by uniformly converting the data to be processed into a three-dimensional coordinate system; Specifically, the raw data from the nine-axis sensor is transformed so that all data can be represented in the same three-dimensional coordinate system for subsequent processing and analysis. The raw data includes acceleration data, gyroscope data, and magnetometer data collected by the nine-axis sensor. For each collection point i, the data can be expressed as: ,in, is the data of the accelerometer on the x, y, and z axes, is the data of the gyroscope on the x, y, and z axes, is the data of the magnetometer on the x, y, and z axes. This step is to unify the data collected at different sensor locations into a global three-dimensional coordinate system. The global coordinate system is , the raw data provided by the nine-axis sensor is relative to the local coordinate system of the sensor itself, and needs to be converted to the global coordinate system through the rotation matrix. The rotation matrix R is used to describe the rotation relationship between the local coordinate system and the global coordinate system. For each acquisition point i, the rotation matrix is ​​calculated using the gyroscope data: ,in, They are the rotation angles around the x, y, and z axes, respectively, obtained by integrating the angular velocity of the gyroscope: , , , use the rotation matrix to transform the acceleration and magnetometer data in the local coordinate system to the global coordinate system: , ,in, , It should be noted that the accelerometer data contains gravity acceleration, and the gravity component needs to be eliminated first: ,in, , g is the standard value of gravity acceleration (9.81 m / s²). The magnetometer data is affected by the ambient magnetic field and needs to be corrected. In this embodiment, correction is performed using the soft iron and hard iron correction models: , where M is the magnetic field soft iron effect matrix and b is the magnetic field hard iron effect bias vector. The angular velocity also needs to be converted to the global coordinate system. Using the same rotation matrix R i Convert the angular velocity data in the local coordinate system to the data in the global coordinate system: ,in, After the above steps, the converted acceleration, angular velocity and magnetometer data are obtained, and these data are integrated into a unified three-dimensional coordinate data set: After data conversion, all sensor data are uniformly represented in the global three-dimensional coordinate system to form a conversion data set .

[0021] In summary, step S2 unifies sensor data from different locations into a common three-dimensional coordinate system to ensure data comparability and consistency. This lays the foundation for subsequent data segmentation, feature extraction, and posture analysis. By eliminating the influence of gravity and environmental magnetic fields, the accuracy and reliability of the data are guaranteed, thereby improving the accuracy of posture analysis and evaluation.

[0022] Step S3, combining the converted data with the data to be processed, segmenting the data according to the action phases, thereby obtaining segmented data, wherein the action phases include entry, paddling, exit, breathing, and turning phases; For details, please refer to the attached Figure 2 As shown, the S3 step includes the following sub-steps: S31, calculating motion state parameters of the data to be processed and the global coordinate data through a sliding window, wherein the motion state parameters include the mean and standard deviation of the acceleration, angular velocity, magnetometer and global coordinate data, which are used to describe the motion situation in each time window; Specifically, the data to be processed, including acceleration, angular velocity, magnetometer data and global coordinate data, are processed using a sliding window technique. The sliding window size w is set to between several hundred milliseconds and one second according to the specific application scenario. In this embodiment, it is set to one second. For each sliding window W j , calculate the mean and standard deviation of acceleration, angular velocity, magnetometer and global coordinate data, the acceleration mean and standard deviation are: , ; The angular velocity mean and standard deviation: , ; The mean and standard deviation of the magnetometer data: , ; The mean and standard deviation of the global coordinate data: , , where a i ,ω i 、m i and r i Respectively represent the acceleration, angular velocity, magnetometer data and global coordinate data at the i-th moment.

[0023] S32, constructing a state vector based on the motion state parameters, wherein the state vector includes the mean and standard deviation of the acceleration, angular velocity, magnetometer and global coordinate data in each time window, which is used to comprehensively reflect the motion situation; Specifically, the motion state parameters in each time window are integrated into a vector by constructing a state vector. This effectively describes the motion situation in each time window and provides basic data support for subsequent action stage recognition and data segmentation. The state vector S is constructed based on the mean and standard deviation of the acceleration, angular velocity, magnetometer and global coordinate data. j : , this state vector includes the mean and standard deviation of acceleration, angular velocity, magnetometer and global coordinate data, which comprehensively reflects the movement within the sliding window.

[0024] S33, based on the state vector, classification is performed by a classification algorithm to obtain an action stage label in each sliding window, wherein the classification algorithm is a support vector machine; Specifically, the state vector is classified by a classification algorithm to identify the action stage label in each sliding window, providing a basis for subsequent data segmentation and action analysis. j As the input of the classification algorithm, in one or more embodiments, a support vector machine is selected as the classification algorithm. Before classification, the SVM model needs to be trained with labeled training data. The training data includes the state vector of the known action stage label, where the input state vector S j , action stage label y j Including entry, paddling, exit, breathing, and turning. According to the characteristics of the data in each stage, specifically, the mean reflects the average movement within a time window. The means of different action stages will be different because the movement patterns of each action stage are different. For example, the mean acceleration of the paddling stage is higher, while the mean acceleration of the breathing stage is lower. The standard deviation reflects the degree of change within a time window. A large standard deviation indicates a large change, and a small standard deviation indicates a small change. For example, the standard deviation of acceleration in the turning stage is larger because the movement changes drastically, while the standard deviation in the paddling stage is relatively small because the movement is relatively stable. During the training process, SVM learns a classification decision function f(x) so that the input state vector S j can be correctly classified. For the training data (Sj ,y j ), SVM looks for an optimal hyperplane that maximizes the classification interval. The definition of the optimal hyperplane is as follows: , where w is the weight vector and b is the bias. The SVM optimization goal is: , where C is the regularization parameter that controls the balance between the classification interval and the misclassification penalty. After the training is completed, the trained SVM model is used to classify the new state vector. For each new state vector S j , prediction is made through the decision function f(x): , according to the output value of the decision function, determine the action stage label Through the above classification process, the action stage label in each sliding window is identified. Each state vector S j Each will be given a label , indicating the action phases within the corresponding time window, such as entry, paddling, exit, breathing, and turning.

[0025] S34, segmenting the original data based on the action stage label to generate segmented data of the corresponding stage, wherein the original data includes the data to be processed and the global coordinate data; Specifically, this step is to segment the original data according to the action stage labels obtained by classification to generate segmented data of the corresponding stage, wherein the input data includes the data to be processed and the global coordinate data as well as the above-mentioned action stage labels. Align the action stage labels with the original data. In this embodiment, each sliding window contains w data points, and there is a certain overlap between the sliding windows. First, the action stage label corresponding to each data point is calculated. Alignment is achieved by repeating the label w times. Specifically, the time range corresponding to the sliding window j is: , at each time point t i The tags are: , if, , based on the action stage label of each data point, the original data is segmented. The segmented data includes the data of the entry, stroke, exit, breathing, and turning stages. Specifically, for each action stage k such as entry, stroke, exit, breathing, and turning, a segmented data set D is constructed. k : And store the segmented data separately according to the action stages.

[0026] In general, by segmenting the original data based on the action stage labels, the data is divided into data for the entry, stroke, exit, breathing, and turning stages. This segmentation method can clearly distinguish the data of different action stages, thereby generating segmented data for the corresponding stages. This segmented data provides a basis for subsequent action feature extraction and posture evaluation.

[0027] Step S4, extracting features from the segmented data, and performing action feature stratification and integration, thereby obtaining feature stratification data; Specifically, in this embodiment, a feature extraction and stratification method based on deep learning is adopted, and the segmented data is processed using a convolutional neural network and a long short-term memory network to obtain high-dimensional, nonlinear, and time-series related features, and the stratification and integration of features are achieved through a network hierarchical structure. , where each segmented data set contains sensor data and global coordinate data of the corresponding action stage. In this implementation, segmented data feature extraction and feature stratification are performed through a deep learning network, and the deep learning network includes a convolution layer and an LSTM layer, wherein the convolution layer extracts local spatiotemporal features, and the convolution kernel performs local scanning on the input data to capture the local pattern of the data. Specifically, for the input data D k , the convolution operation is defined as: ,in is the convolution kernel, is the bias, represents the convolution operation, is the activation function. The LSTM layer is used to capture the temporal characteristics of the data, and can handle long-term dependencies, capturing temporal related information through memory units and gating mechanisms. Specifically, for the feature map output by the convolutional layer , the LSTM operation is defined as: ,in It is hidden. is a memory unit, is the input of the tth time step. In general, feature stratification is achieved through the network hierarchy, and each layer captures feature information at different levels. The first layer of local features, that is, the spatiotemporal features extracted by the convolutional layer mentioned above: , and the second time series feature layer, that is, the time series features extracted by the LSTM layer mentioned above: Next, the features at different levels are integrated to form a comprehensive feature representation. Specifically, the integrated feature vector , where concat represents the concatenation operation of feature vectors. Through feature stratification and integration, the final feature stratification data F is generated. final ,used for subsequent posture deviation analysis and comprehensive evaluation. .

[0028] In summary, by combining the deep learning method of convolutional neural network and long short-term memory network, high-dimensional and time-related motion features are automatically extracted from segmented data. Through the hierarchical structure of the deep learning network, the stratification and integration of features are realized, which can not only capture the local spatiotemporal characteristics of the data, but also effectively handle long-term temporal dependencies. This method is efficient, accurate and automated, providing high-quality feature data for subsequent posture deviation analysis and comprehensive evaluation, making the overall analysis more refined and comprehensive.

[0029] Step S5, calling the posture deviation model, inputting the feature hierarchical data into the posture deviation model, thereby obtaining the three-dimensional motion trajectory data of the key point; based on the three-dimensional motion trajectory data of the key point, projecting onto three mutually perpendicular two-dimensional planes, obtaining the actual projection trajectory; generating a theoretical trajectory according to the actual projection trajectory of the key point on the two-dimensional plane and the pre-collected standard posture data, and calculating the difference between the actual trajectory and the theoretical trajectory, and forming a posture deviation index by combining the trajectory difference of the key point on the two-dimensional plane; For details, please refer to the attached Figure 3 As shown, the step S5 includes the following sub-steps: S51, inputting the feature hierarchical data into a posture deviation model, wherein the posture deviation model is used to calculate the motion trajectory of the key point in the three-dimensional coordinate system, thereby obtaining the three-dimensional motion trajectory data of the key point; Specifically, in this embodiment, the posture deviation model is a hybrid model that integrates convolutional neural network, graph convolutional network and attention mechanism, and processes the feature hierarchical data obtained in the above steps to obtain the three-dimensional motion trajectory data of the key points. integrated Input model. Use CNN layer to extract spatial information in features: , where C is the output of the CNN layer, W cnn is the convolution kernel, b cnn is the bias, and then the relationship between key points is captured through the graph convolutional network: , where H is the output of the GCN layer and A is the adjacency matrix. Next, we introduce the attention mechanism to enhance the model's focus on key points: , where Z is the output of the attention mechanism. Then the output of the attention mechanism is input into the fully connected layer for processing: , where F is the output of the fully connected layer, W f is the weight matrix, b f is the offset. The final output is the three-dimensional motion trajectory of the key points: , where Y is the three-dimensional motion trajectory of the key point, W o is the weight matrix, b ois the offset. It should be noted that the key points include the head, shoulders, arms, waist and legs. The key points are identified by the pre-set sensor positions and the label data during the model training process. The three-dimensional motion trajectory Y contains the position changes of the key points in the three-dimensional coordinate system. Specifically, for each key point k and each time step t, the trajectory data can be expressed as: ,in , , They represent the three-dimensional coordinates of the key point k at time step t respectively.

[0030] S52, based on the three-dimensional motion trajectory data of the key point, projecting it onto three mutually perpendicular two-dimensional planes, where the two-dimensional planes are an XY plane, a YZ plane, and an XZ plane, thereby obtaining a projection trajectory of the key point on the two-dimensional plane; Specifically, in step S51, the motion trajectory data of the key points in the three-dimensional coordinate system are obtained through the posture deviation model. The three-dimensional motion trajectory data of the key points can comprehensively reflect the changes in the athlete's posture. However, in order to more intuitively analyze the motion laws and posture deviations of these key points, it is necessary to project the three-dimensional trajectory data onto a two-dimensional plane. In this way, the analysis of the key point movement can be simplified and further deviation detection can be facilitated. In order to project the three-dimensional trajectory data onto a two-dimensional plane, it is necessary to perform a mathematical transformation on the position of each key point in the three-dimensional coordinate system. The mathematical transformation is achieved through matrix operations. Specifically, the projection plane is first defined, where the XY plane only retains the X and Y coordinates; the YZ plane only retains the Y and Z coordinates; and the XZ plane only retains the X and Z coordinates. For each key point p k,t =(x k,t ,y k,t ,z k,t ) is the position in three-dimensional space of the key point k at time step t. k,t , projected by the following formula to obtain the projection trajectory of the key points on the two-dimensional plane: Projection on the XY plane: Projection on the YZ plane: Projection on the XZ plane: The actual coordinates of each key point on different two-dimensional planes are obtained through the above-mentioned projection transformation. The line connecting the actual coordinates of each key point is the two-dimensional projection trajectory, which not only simplifies the analysis of three-dimensional motion, but can also be used for posture deviation detection and analysis in subsequent steps. Specifically, by comparing and analyzing these two-dimensional projection trajectories, the motion law and posture deviation of the key points can be evaluated more intuitively.

[0031] S53, generating a theoretical trajectory through a clustering algorithm according to the actual projection trajectory of the key point on the two-dimensional plane and the pre-collected standard posture data, wherein the theoretical trajectory is used to characterize the trajectory of the key point under the standard motion posture; Specifically, the projection trajectory of the key points obtained in step S52 on the two-dimensional plane , and , and simultaneously obtain multiple pre-collected standard motion posture data sets, wherein the motion posture data sets contain the projection trajectories of key points under standard motion postures on a two-dimensional plane. Next, a clustering algorithm is used to perform cluster analysis on the standard posture data. In this embodiment, K-means clustering is used for cluster analysis, and N standard posture data sets are provided, each of which contains the two-dimensional projection trajectories of key points. The two-dimensional projection trajectories in these data sets are used as input, and clustering is performed using the K-means clustering algorithm. According to the diversity of the standard posture data, the number of cluster centers K is determined to be set, and K cluster centers are obtained, each of which represents a key point trajectory pattern under a standard motion posture. Next, the actual projection trajectory data is matched with the cluster center to determine which cluster the actual trajectory belongs to. Based on the matching results, the corresponding cluster center is selected as the theoretical trajectory of the key point. Specifically, for the actual projection trajectory of the key point on the XY plane , by calculating the distance between the actual projection trajectory and each cluster center, the nearest cluster center is found , and use it as a theoretical trajectory In this embodiment, the Euclidean distance between the actual projection trajectory and each cluster center is calculated, and the specific calculation is as follows: ,in, Represent the actual projection trajectory and the coordinate value of the cluster center at the jth time point. In this way, the nearest cluster center can be found. , specifically: The same is true for the other YZ and XZ planes, so I won't go into detail here. In general, the actual projection trajectory data reflects the athlete's real motion trajectory, while the standard posture data represents the standard motion trajectory under ideal conditions. By comparing the actual projection trajectory data with the standard posture data, the corresponding theoretical trajectory can be obtained.

[0032] S54, calculating the trajectory difference between the two-dimensional projection trajectory and the theoretical trajectory, wherein the trajectory difference includes a trajectory deviation value and a trajectory deviation direction, which are used to quantify the motion deviation of the key point; Specifically, the trajectory deviation value in this step is the distance between the calculated two-dimensional projection trajectory and the theoretical trajectory, which is used as the trajectory deviation value; the trajectory deviation direction is the deviation direction of the calculated two-dimensional projection trajectory relative to the theoretical trajectory, which is used to describe the deviation direction of the motion trajectory. Specifically, the trajectory deviation value: for the projection trajectory on the XY plane and theoretical trajectory , , for the trajectories on the YZ plane and the XZ plane, the deviation values ​​are calculated similarly: ; , the trajectory deviation value is used to quantify the degree of motion deviation of the key point. The larger the deviation value, the greater the degree to which the actual motion trajectory deviates from the standard motion trajectory. Wherein the trajectory deviation direction: for the trajectory deviation direction θ on the XY plane XY The calculation formula is: ,For the trajectory deviation directions on the YZ plane and XZ plane, the following calculations are performed similarly: , ,The trajectory deviation direction is used to describe the movement deviation direction of the key point. By analyzing the deviation direction, the specific direction in which the key point deviates from the standard trajectory during the movement can be understood, providing directional guidance for movement correction. In a specific embodiment, such as the analysis of hand key points in swimming, the trajectory of the hand key points is very important in swimming. Through the above steps, the theoretical trajectory of the hand key points under the standard swimming style can be obtained. By calculating the deviation value and deviation direction between the actual movement trajectory and the theoretical trajectory, it is possible to evaluate whether the swimmer's hand movements are standardized and provide specific correction suggestions.

[0033] S55, integrating the trajectory differences of each key point on all two-dimensional planes to form a posture deviation index, wherein the posture deviation index is used to evaluate the accuracy and consistency of the overall motion posture.

[0034] Specifically, for each key point k, the weighted average of the trajectory deviation values ​​is calculated in the XY plane, YZ plane, and XZ plane. k,t,XY ,d k,t,YZ ,d k,t,XZ Respectively represent the trajectory deviation values ​​of key point k on the XY, YZ, and XZ planes, and calculate the comprehensive trajectory deviation value D k : , where α, β, and γ are weight coefficients, satisfying α+β+γ=1.

[0035] And for each key point k, the weighted average of the trajectory deviation direction is calculated on the XY plane, YZ plane and XZ plane. k,t,XY ,θ k,t,YZ ,θ k,t,XZRespectively represent the trajectory deviation direction of the key point k on the XY, YZ, and XZ planes, and calculate the comprehensive trajectory deviation direction : .

[0036] Next, the comprehensive trajectory deviation value and deviation direction of all key points K are weighted averaged to form the overall posture deviation index P. Next, the posture deviation index P is calculated, where D k is the comprehensive trajectory deviation value of key point k, is the comprehensive trajectory deviation direction of key point k, which is calculated as follows: ,in is the weight coefficient of each key point, satisfying .

[0037] In general, this step first obtains the trajectory difference data from the previous step, including the trajectory deviation value and deviation direction of each key point on different two-dimensional planes. Then, the comprehensive trajectory deviation value and deviation direction of each key point are obtained by weighted averaging the trajectory deviation values ​​and deviation directions on different two-dimensional planes. This step takes into account the trajectory differences and deviation directions on different planes to ensure the comprehensiveness of the comprehensive evaluation. Finally, the overall posture deviation index is formed by weighted averaging the comprehensive trajectory deviation values ​​and deviation directions of all key points.

[0038] Step S6, merging the posture deviation indicators of each action stage to form an overall coherent posture deviation indicator; Specifically, in the previous steps, the action phases, such as entry, paddling, exit, breathing, turning, etc., have been identified and segmented, and the posture deviation index of each phase has been calculated separately. Now it is necessary to merge these phased posture deviation indexes to form an overall, coherent posture deviation index to evaluate the accuracy and consistency of the posture during the entire movement. Specifically, the posture deviation indexes of all phases are sorted in chronological order to ensure that the connection between each phase is continuous. Among them, the posture deviation index for each phase i is P i , the corresponding time period is T i The posture deviation indicators of all stages are spliced ​​in chronological order to form a complete time series P total Its expression is , where n is the number of action stages. If necessary, the posture deviation index is weighted averaged according to the importance of each stage. Let the weight of each stage be ,satisfy The formula for calculating the overall posture deviation index is expressed as: , if the weights of each stage are equal, then P overall is a simple average.

[0039] In summary, step S6 combines the posture deviation indicators of each action stage through data sorting, time series splicing and weighted averaging to form an overall coherent posture deviation indicator. The core of this step is to ensure the continuity and coherence of the data, and at the same time, through a reasonable weighted average method, to ensure that the importance of each stage in the overall evaluation is properly reflected. It should be noted that the main significance of the action stage segmentation and merging is the combination of refined analysis and overall coherence evaluation. This process makes the analysis more accurate by refining the posture deviation of each action stage, while ensuring the consistency and coherence of the overall action, and finally comprehensively obtains a comprehensive evaluation result.

[0040] Step S7, calling a comprehensive analysis model, inputting the overall coherent posture deviation index into the comprehensive analysis model, analyzing the balance, synchronization, and streamline, thereby evaluating the balance, synchronization, and streamline of the swimming posture and obtaining an evaluation result; Specifically, the comprehensive analysis model includes a balance analysis sub-model, a synchronization analysis sub-model and a streamline analysis sub-model.

[0041] The balance analysis sub-model is used for balance analysis, thereby reflecting the athlete's ability to maintain body stability and balance during swimming. Specifically, it measures the degree of deviation between the left and right sides of the body during movement. If the deviation is large, it means that the athlete is unbalanced during swimming.

[0042] In step S6, the overall coherent posture deviation index P is obtained. overall (t), covers the posture deviation data of all key points in the whole swimming process. According to the position of the key point, each key point k is defined as the left side or the right side. In this embodiment, the left key points include the left shoulder, left elbow, left wrist, and left knee; the right key points include the right shoulder, right elbow, right wrist, and right knee.

[0043] Suppose we have N key points, denoted as k 1 ,k 2 ,…,k N , the posture deviation of each key point at time t is P overall,k (t).

[0044] Construct the left and right keypoint sets:

[0045]

[0046] Then calculate the posture deviation index of the left and right key points. For each time point t, the posture deviation index P of the left key point left (t) is calculated as follows: ; Posture deviation index P of the key point on the right right (t) is calculated as follows: Next, we calculate the balance deviation. The balance is measured by calculating the difference in posture deviation between the left and right key points. For each time point t, we calculate the difference in posture deviation between the left and right key points: ; Finally, the overall balance is evaluated by calculating the average balance deviation during the entire swimming process. The specific calculation formula is as follows: , where T is the total time of the entire swimming process.

[0047] The synchronization analysis sub-model is used for synchronization analysis, thereby reflecting the coordination of the movements of the key points of the athlete during the swimming process. Specifically, it measures the degree to which the various parts of the body move synchronously during the movement. If the movements of different key points are not synchronized, it means that the athlete has a coordination problem during the swimming process.

[0048] In step S6, the overall coherent posture deviation index P is obtained. overall (t), covers the posture deviation data of all key points in the whole swimming process. The posture deviation of each key point k at time t is P overall,k (t).

[0049] Specifically, synchronization analysis is based on the evaluation of the time difference between key points. Ideally, each key point should move at the same rhythm. The motion phase of each key point k is defined as To characterize its motion rhythm. By analyzing the posture deviation curve P of each key point overall,k (t), extract its motion phase . Let the motion period of each key point be T k , then the motion phase can be expressed as: ,in, is the initial phase of key point k. Next, the phase difference of the key points is calculated. For each pair of key points k i and k j , calculate their phase difference at time t : Next, we calculate the synchronization deviation, which is measured by calculating the average phase difference between all key points. For each time point t, we calculate the average phase difference D of all key point pairs. synchrony (t): , where N is the total number of key points and T is the total time of the entire swimming process.

[0050] The streamline analysis submodel is used for streamline analysis, which reflects the smoothness and consistency of the athlete's motion trajectory during swimming. Specifically, it measures whether the body maintains a smooth, streamlined trajectory during movement. If the motion trajectory is not smooth, it means that the athlete has unnecessary movements or poor posture during swimming, which increases the resistance in the water.

[0051] In step S6, the overall coherent posture deviation index P is obtained. overall (t), covers the posture deviation data of all key points during the entire swimming process.

[0052] For each key point k, from P overall (t) to extract its motion trajectory P k (t). By k (t) Calculate the first and second order derivatives to obtain the velocity V of the key point k (t) and acceleration A k (t): ; ; Next, the trajectory smoothness is calculated. The streamline is measured by calculating the smoothness of the key point motion trajectory. Smoothness can be evaluated by analyzing the rate of change of acceleration, that is, the derivative of acceleration: In this embodiment, the smoothness index S k Calculated by the RMS value of the acceleration derivative: , where T is the total time of the entire swimming process. Finally, the smoothness index of each key point is combined, and the streamline is evaluated by combining the smoothness index of all key points. The specific calculation formula is as follows: , where N is the total number of key points. By calculating the smoothness index S of each key point k And its comprehensive smoothness index S overall , you can evaluate the streamline of the athlete throughout the swimming process. The higher the streamline index, the smoother the athlete's trajectory, the less resistance, and the higher the efficiency.

[0053] In summary, a comprehensive analysis model is used in this step, and the overall coherent posture deviation indicators obtained from the previous step are input into the model. By analyzing these indicators, the model conducts a detailed evaluation of the athlete's swimming posture in three aspects: balance, synchronization, and streamline. Balance analysis measures the stability of the body by comparing the posture deviations of the left and right key points; synchronization analysis evaluates the coordination of various parts of the body by calculating the posture deviations at different stages; and streamline analysis evaluates the smoothness of the motion trajectory through the speed and acceleration change rate of the key points. Finally, the model can comprehensively evaluate the accuracy and consistency of the athlete's posture during swimming.

[0054] Step S8, issuing a warning and generating swimming posture correction suggestions based on the balance, synchronization and streamline evaluation results.

[0055] Specifically, based on the evaluation results of balance, synchronization and streamline, a comprehensive analysis is conducted to determine the main problems that athletes have during swimming. avg_balance If the D is large, it means that the athlete has a significant imbalance during swimming. Balance problems will cause the movement trajectory to deviate during swimming, thereby increasing resistance and reducing swimming efficiency. Synchronicity problems, if D synchrony If the S is large, it means that the athlete's limbs are not coordinated during swimming. Synchronicity problems will lead to discontinuity of movements during swimming, thus affecting the effective transmission of propulsion. Streamline problems, if S overall If the value is lower, it means that the athlete has greater resistance during swimming. Streamline problems will lead to uneven posture during swimming, thereby increasing resistance in the water and reducing swimming efficiency. In order to solve these problems, the linear patch vibration motor in the system will send out a vibration prompt in real time when posture deviation is detected, reminding the athlete to adjust the posture immediately, thereby improving training efficiency and the accuracy of posture correction.

[0056] Based on the comprehensive analysis results, specific swimming posture correction suggestions are generated.

[0057] Balance correction suggestion: Perform balance training, such as balance exercises and core strength training to enhance the body's balance ability. In this embodiment, the recommended exercises include single-leg standing exercises, core stability training, and balance board exercises.

[0058] Synchronicity correction suggestion: Perform synchronization training, such as rhythm training and limb coordination training to improve the coordination of the limbs. In this embodiment, the recommended exercises include double-arm rowing exercises, rhythm training, and coordination exercises.

[0059] Streamline correction suggestion: Perform streamline training, such as reducing the resistance generated during swimming through posture adjustment and streamline exercises. In this embodiment, the recommended exercises include streamline gliding exercises, posture correction training, and resistance reduction exercises.

[0060] The resulting corrective recommendations are then output in the form of a report for athletes and coaches to review. The report should detail the balance, synchronization, and flow assessment results, as well as specific corrective recommendations for each problem.

[0061] Based on the description of the above-mentioned swimming action posture analysis method embodiment, the present application embodiment also discloses a swimming action posture analysis device, which can be a computer program (including program code) for running the above-mentioned swimming action posture analysis method. Figure 4 As shown, the following units can be run based on the swimming action posture analysis device: An acquisition unit 110 is used to acquire data to be processed, wherein the data to be processed is data of the head, shoulders, arms, waist and legs of a swimmer in motion, and the data to be processed is obtained through a nine-axis sensor; The conversion unit 120 is used to perform data conversion on the data to be processed, so as to obtain conversion data, wherein the conversion data is coordinate data obtained by uniformly converting the data to be processed into a three-dimensional coordinate system; A segmentation unit 130 is used to combine the converted data with the data to be processed, and perform data segmentation according to the action phases to obtain segmented data, wherein the action phases include entry, paddling, exit, breathing, and turning phases; A feature extraction unit 140 is used to extract features from the segmented data and perform action feature stratification to obtain feature stratified data; The deviation calculation unit 150 is used to call the posture deviation model, input the feature layered data into the posture deviation model, so as to obtain the three-dimensional motion trajectory data of the key point; based on the three-dimensional motion trajectory data of the key point, the actual projection trajectory is obtained by projecting onto three mutually perpendicular two-dimensional planes; a theoretical trajectory is generated according to the actual projection trajectory of the key point on the two-dimensional plane and the pre-collected standard posture data, and the difference between the actual trajectory and the theoretical trajectory is calculated, and the posture deviation index is formed by integrating the trajectory difference of the key point on the two-dimensional plane; A merging unit 160, used to merge the posture deviation indicators of each action stage to form an overall coherent posture deviation indicator; The analysis unit 170 is used to call the comprehensive analysis model, input the overall coherent posture deviation index into the comprehensive analysis model, analyze the balance, synchronization, and streamline, thereby evaluating the balance, synchronization, and streamline of the swimming posture and obtaining an evaluation result; The suggestion generating unit 180 is used to generate swimming posture correction suggestions based on the balance, synchronization and streamline evaluation results.

[0062] The above is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art shall not deviate from the spirit and scope of the present invention, and shall be within the scope of protection of the claims attached to the present invention.

Claims

1. A swimming posture analysis method, characterized in that: The following steps are involved: S1, obtaining data to be processed, wherein the data to be processed is data of the head, shoulders, arms, waist and legs of a swimmer during exercise, and the data to be processed is obtained through a nine-axis sensor; S2, performing data conversion on the data to be processed, thereby obtaining conversion data, wherein the conversion data is coordinate data obtained by uniformly converting the data to be processed into a three-dimensional coordinate system; S3, combining the converted data with the data to be processed, and segmenting the data according to the action phases, thereby obtaining segmented data, wherein the action phases include entry, paddling, exit, breathing, and turning phases; S4, extracting features from the segmented data and performing action feature stratification to obtain feature stratified data; S5, calling the posture deviation model, inputting the feature layered data into the posture deviation model, thereby obtaining the three-dimensional motion trajectory data of the key points; Based on the three-dimensional motion trajectory data of key points, the actual projection trajectory is obtained by projecting them onto three mutually perpendicular two-dimensional planes; Generate a theoretical trajectory based on the actual projection trajectory of the key points on the two-dimensional plane and the pre-collected standard posture data, calculate the difference between the actual trajectory and the theoretical trajectory, and form a posture deviation index based on the trajectory difference of the key points on the two-dimensional plane; S6, merging the posture deviation indicators of each action stage to form an overall coherent posture deviation indicator; S7, calling a comprehensive analysis model, inputting the overall coherent posture deviation index into the comprehensive analysis model, analyzing the balance, synchronization, and streamline, thereby evaluating the balance, synchronization, and streamline of the swimming posture and obtaining an evaluation result; Step S8, issuing a warning and generating swimming posture correction suggestions based on the balance, synchronization and streamline evaluation results.

2. A swimming posture analysis method according to claim 1, characterized in that: In the step S2, the coordinate data is obtained by converting the data to be processed into the global coordinate system through a rotation matrix, and the rotation matrix is ,in are the rotation angles around the x, y, and z axes respectively. The nine-axis nine-axis sensor includes a gyroscope. The rotation angle is obtained by integrating the angular velocity of the gyroscope. The specific calculation formula is as follows: , , .

3. A swimming posture analysis method according to claim 2, characterized in that: The S3 step includes the following sub-steps: S31, calculating motion state parameters of the data to be processed and the global coordinate data through a sliding window, wherein the motion state parameters include the mean and standard deviation of the acceleration, angular velocity, magnetometer and global coordinate data, which are used to describe the motion situation in each time window; S32, constructing a state vector based on the motion state parameters, wherein the state vector includes the mean and standard deviation of the acceleration, angular velocity, magnetometer and global coordinate data in each time window, which is used to comprehensively reflect the motion situation; S33, based on the state vector, classification is performed by a classification algorithm to obtain an action stage label in each sliding window, wherein the classification algorithm is a support vector machine; S34, based on the action stage label, the original data is segmented to generate segmented data of the corresponding stage, and the original data includes the data to be processed and the global coordinate data.

4. A swimming posture analysis method according to claim 2, characterized in that: The S4 step is to extract and stratify the features of the segmented data through a deep learning network, wherein the deep learning network includes a convolutional layer and an LSTM layer, wherein the convolutional layer extracts local spatiotemporal features, and the LSTM layer is used to capture the temporal features of the data.

5. A swimming posture analysis method according to claim 2, characterized in that: The S5 step includes the following sub-steps: S51, inputting the feature hierarchical data into a posture deviation model, wherein the posture deviation model is used to calculate the motion trajectory of the key point in the three-dimensional coordinate system, thereby obtaining the three-dimensional motion trajectory data of the key point; S52, based on the three-dimensional motion trajectory data of the key point, projecting it onto three mutually perpendicular two-dimensional planes, where the two-dimensional planes are an XY plane, a YZ plane, and an XZ plane, thereby obtaining a projection trajectory of the key point on the two-dimensional plane; S53, generating a theoretical trajectory through a clustering algorithm according to the actual projection trajectory of the key point on the two-dimensional plane and the pre-acquired standard posture data, wherein the theoretical trajectory is used to characterize the trajectory of the key point under the standard motion posture; S54, calculating the trajectory difference between the two-dimensional projection trajectory and the theoretical trajectory, wherein the trajectory difference includes a trajectory deviation value and a trajectory deviation direction, which are used to quantify the motion deviation of the key point; S55, integrating the trajectory differences of each key point on all two-dimensional planes to form a posture deviation index, wherein the posture deviation index is used to evaluate the accuracy and consistency of the overall motion posture.

6. A swimming posture analysis method according to claim 5, characterized in that: The posture deviation model is a hybrid model that integrates convolutional neural network, graph convolutional network and attention mechanism.

7. A swimming posture analysis method according to any one of claims 1 to 6, characterized in that: The comprehensive analysis model includes a balance analysis sub-model, a synchronization analysis sub-model and a streamline analysis sub-model.

8. A swimming posture analysis method according to claim 7, characterized in that: The specific calculation formula of the balance analysis sub-model is as follows: , where T is the total time of the entire swimming process, D balancede The specific calculation formula is as follows: .

9. A swimming posture analysis method according to claim 7, characterized in that: The specific calculation formula of the synchronization analysis sub-model is as follows: , where N is the total number of key points, T is the total time of the entire swimming process, The specific calculation formula is as follows: , the motion phase calculation formula is as follows: ,in is the initial phase of key point k, Tk is the motion period of each key point; the specific calculation formula of the streamline analysis sub-model is as follows: , where N is the total number of key points; the smoothness index S k The calculation formula is as follows: , where T is the total time of the entire swimming process; the specific calculation formula for smoothness is as follows: ; Acceleration A k (t) The specific calculation formula is as follows: , P k (t) is the overall coherent posture deviation index P overall (t) The extracted motion trajectory.

10. A swimming posture analysis device, characterized in that: The device comprises: An acquisition unit, used for acquiring data to be processed, wherein the data to be processed is data of the head, shoulders, arms, waist and legs of a swimmer in motion, and the data to be processed is obtained through a nine-axis sensor; A conversion unit, used for performing data conversion on the data to be processed, so as to obtain conversion data, wherein the conversion data is coordinate data obtained by uniformly converting the data to be processed into a three-dimensional coordinate system; a segmentation unit, used to combine the converted data with the data to be processed, and segment the data according to the action phases, so as to obtain segmented data, wherein the action phases include entry, paddling, exit, breathing, and turning phases; A feature extraction unit, used to extract features from the segmented data and perform action feature stratification to obtain feature stratified data; The deviation calculation unit is used to call the posture deviation model, input the feature layered data into the posture deviation model, so as to obtain the three-dimensional motion trajectory data of the key point; based on the three-dimensional motion trajectory data of the key point, the actual projection trajectory is obtained by projecting onto three mutually perpendicular two-dimensional planes; a theoretical trajectory is generated according to the actual projection trajectory of the key point on the two-dimensional plane and the pre-collected standard posture data, and the difference between the actual trajectory and the theoretical trajectory is calculated, and the posture deviation index is formed by integrating the trajectory difference of the key point on the two-dimensional plane; A merging unit, used to merge the posture deviation indicators of each action stage to form an overall coherent posture deviation indicator; An analysis unit is used to call a comprehensive analysis model, input the overall coherent posture deviation index into the comprehensive analysis model, analyze the balance, synchronization, and streamline, thereby evaluating the balance, synchronization, and streamline of the swimming posture and obtaining an evaluation result; The suggestion generation unit is used to issue warnings and generate swimming posture correction suggestions based on the balance, synchronization and streamline assessment results.

Images