A training action comparison method and system

Abstract

The present application provides a training action comparison method and system. The method captures and records pre-stored actions; captures the motion data of the action to be recognized; classifies the captured action data into single-plane actions and three-dimensional actions; for single-plane actions, records the starting position, ending position, and angle, and compares them with the pre-stored action data to determine the consistency of the actions; for three-dimensional actions, gradually decomposes them into the XY plane, XZ plane, and YZ plane, and compares the three different plane coordinate sequences with the pre-stored actions, thereby realizing the comparison of two sets of actions during the practice process, supervising and reminding the user to train according to the pre-stored actions during the training process, and ensuring the accuracy and precision of the training actions.

Description

Technical Field

[0001] The present invention relates to the technical field of training, and in particular, to a method for comparing training actions. Background Art

[0002] The training action comparison method is a technique for comparing two action sequences, which originated from the need to evaluate the action performance of athletes in the fields of sports and medicine. In sports, the evaluation of action performance is an important indicator for assessing the skill level and progress of athletes. In the medical field, the evaluation of action performance is used to monitor and evaluate the rehabilitation progress of patients, as well as to guide treatment and training plans.

[0003] In order to better achieve the effectiveness of training actions, humans have made many attempts. For example, Chinese Patent CN106984027A discloses "a method and device for action comparison analysis and a display, which detect the current limb actions of a target user; obtain the standard action data corresponding to the actions in the current screen of the target display; compare and analyze the detected current limb actions with the standard action data to obtain a comparison and analysis result; and output feedback information corresponding to the comparison and analysis result." However, this action comparison method only considers actions in a single plane and cannot identify the details of three-dimensional actions, so the error is relatively large. Especially for combined collaborative actions with displacements and angle changes in both multi-dimensional space and plane space, it is simply impossible to identify and analyze them. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention provides a method for comparing training actions, which improves the accuracy and precision of action recognition, adapts to more complex action forms, can more accurately identify actions, reduces the false judgment rate, improves the accuracy and efficiency of recognition, and at the same time can improve the speed of on-site action recognition.

[0005] A method for comparing training actions, characterized by comprising the following steps:

[0006] S1, capturing the motion data of the action to be recognized;

[0007] S2, classifying the captured action data into single-plane actions and three-dimensional actions;

[0008] S3, for single-plane actions, recording the starting position, ending position and angle, and comparing them with the pre-stored action data to determine the consistency of the actions;

[0009] S4, for three-dimensional actions, gradually decomposing them into the XY plane, XZ plane and YZ plane, and comparing the three different plane coordinate sequences with the pre-stored actions;

[0010] In the step S2, the action is classified as a single-plane action or a three-dimensional action according to whether the change amplitudes on the three axes are all greater than the threshold value;

[0011] The step of gradually decomposing the three-dimensional action into the XY plane, the XZ plane and the YZ plane and comparing the three different plane coordinate sequences includes the following steps:

[0012] i. Divide the action into different plane coordinate sequences of the three planes, namely the XY plane, the XZ plane and the YZ plane, respectively;

[0013] ii. Transform the reverse change of the angle of each axis in the spatial direction into different key point sets;

[0014] iii. Record the comparison between the key point data and the pre-stored action data to determine the consistency of the action.

[0015] Further, before the step S1, capture and record the pre-stored action, and record the change angle of each segment of the XYZ axis, so as to be able to compare with the action captured quickly and on-site.

[0016] Further, the comparison between the single-plane action and the pre-stored action data is carried out by a single-node simple angle decomposition algorithm.

[0017] Further, when executing S4, it is through the Euclidean distance D combined with the included angle θ 合 to evaluate the similarity of the two sequences, where the distance included angle θ 合 =(θ xy +θ xz +θ yz ) / 3, where

[0018] Further, before executing S2, according to the starting point where the on-site action is consistent with the pre-stored action, ignore the inconsistent part in the front of the on-site action.

[0019] Further, in the step S1, it is divided into action capture with time requirements and action capture ignoring time elements.

[0020] Further, the Euclidean distance D combined with the included angle θ 合 evaluates the similarity of the two sequences by respectively setting the weights of D and θ 合 and performing weighted sum comparison and evaluation.

[0021] In addition, the present application also discloses a training action comparison system, which is characterized in that the system includes:

[0022] An action data capture module for capturing motion data of an action to be recognized;

[0023] An action data classification module for classifying the captured action data into single-plane actions and three-dimensional actions;

[0024] A single-plane action comparison module for single-plane actions, recording the starting position, ending position, and angle, and comparing with pre-stored action data to determine the consistency of the action;

[0025] A three-dimensional action comparison module for three-dimensional actions, gradually decomposing it into the XY plane, XZ plane, and YZ plane, and comparing the coordinate sequences of the three different planes with pre-stored actions;

[0026] The action data classification classifies the action as a single-plane action or a three-dimensional action according to whether the change amplitudes on the three axes are all greater than a threshold value;

[0027] The step of gradually decomposing the three-dimensional action into the XY plane, XZ plane, and YZ plane and comparing the coordinate sequences of the three different planes includes the following steps:

[0028] i. Divide the action into different plane coordinate sequences of the three planes, namely the XY plane, XZ plane, and YZ plane, respectively;

[0029] ii. Transform the reverse change of the angle of each axis in the spatial direction into different key point sets;

[0030] iii. Record the key point data and compare it with the pre-stored action data to determine the consistency of the action.

[0031] Advantages of the present invention:

[0032] The training action comparison method of the present invention improves the accuracy and precision of action recognition, adapts to more complex action forms, can more accurately identify actions, reduces the misjudgment rate, improves the accuracy and efficiency of recognition, and at the same time can improve the speed of on-site action recognition, having high practical value. Description of the Drawings

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the structures shown in these drawings.

[0034] Figure 1 It is a schematic flowchart of the method of the present application;

[0035] Figure 2 Schematic diagram for comparing single-plane actions of this application;

[0036] Figure 3 Schematic diagram of the three-dimensional motion trajectory in the space coordinate system xyz of this application;

[0037] Figure 4 Schematic diagram of the key point set after decomposing the three-dimensional actions of this application;

[0038] Figure 5 Schematic diagram of simplifying the connection of key point sets after decomposing the three-dimensional actions of this application;

[0039] Figure 6 Schematic diagram of the key point sets of the XY plane, XZ plane, and YZ plane after decomposing the three-dimensional actions of this application;

[0040] Figure 7 Schematic diagram of data analysis for ignoring the inconsistent parts of the on-site action data in the action comparison of this application. Detailed implementation manners

[0041] The following will describe the detailed implementation manners of the present disclosure in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all embodiments. The components of the embodiments of the present disclosure described and illustrated herein can generally be arranged and designed in various different configurations.

[0042] Therefore, the detailed description of the embodiments of the present disclosure provided in the accompanying drawings is not intended to limit the scope of the present disclosure claimed, but merely represents the selected embodiments of the present disclosure, and is only used to illustrate and explain the present disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of the present disclosure without creative efforts belong to the scope of protection of the present disclosure.

[0043] As shown in this application and the claims, unless the context clearly indicates otherwise, words such as "a", "an", "one", and / or "the" are not specifically singular and may also include plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0044] As Figure 1 shown, the present invention discloses a training action comparison method, including the following steps:

[0045] Step 1, capturing and recording the pre-stored actions;

[0046] Step 2, capturing the motion data of the action to be recognized;

[0047] Step 3: Classify the captured motion data into single-plane motions and three-dimensional motions;

[0048] Step 4: For single-plane motions, record the starting position, ending position, and angle, and compare with the pre-stored motion data to determine the consistency of the motion;

[0049] Step 5: For three-dimensional motions, gradually decompose them into three different plane coordinate sequences and compare with the pre-stored motions.

[0050] For the step of capturing and recording pre-stored motions, we need to pre-store some standard professional motions in advance. This step can be performed by professional training coaches or diagnostic attending physicians. The step of pre-storing professional motions is carried out before training motions. It involves classifying the motions to be recognized into different categories in advance, and selecting multiple typical motions for each category to record and store. These motions need to be carefully designed and accurately executed by professionals during recording to ensure that they represent the typical motions in that category.

[0051] During the recording process, sensors such as gyroscopes and accelerometers need to be fixed on key parts of the human body to record motion data. These data are processed and saved as a set of feature vectors for subsequent use. At the same time, to improve the recognition accuracy, it is also necessary to record the start and end states of each motion, as well as information such as the position and angle of key points.

[0052] After the model is trained, these pre-stored professional motions will be used as a comparison and reference to identify real-time motion data. Therefore, the accuracy and richness of pre-stored professional motions are crucial for recognition accuracy. After the pre-stored motion information is completed, the next step is to capture and identify training information and compare it with the pre-stored motions.

[0053] In the step of capturing the motion data of the motion to be recognized, capturing the motion data of the motion to be recognized means using sensors such as gyroscopes and accelerometers to record the relevant postures and motion information of the motion executor during the motion, and storing and representing them in digital form. In this process, the gyroscope and accelerometer can obtain the direction and rotation angle information of the motion executor by measuring the motion acceleration and angular velocity, and convert this information into the form of Euler angles for storage.

[0054] Before motion capture, it is necessary to calibrate the gyroscope once to obtain the initial angle information. After that, sensors such as gyroscopes and accelerometers will continuously sample the posture and motion of the motion executor. The collected data is filtered and fused through techniques such as digital signal processing to remove noise and errors, obtain accurate angle information, and store it in the form of Euler angles.

[0055] In this technical solution, Euler angles are used for storage, that is, the posture and motion information of the motion executor are represented and stored in the form of a set of three Euler angles, namely: pitch angle, yaw angle, and roll angle. This method can not only accurately record the direction and rotation angle information of the motion executor, but also has a smaller storage space, which is convenient for storage and processing.

[0056] When the action to be recognized is captured, it is often difficult to determine the type of action when comparing the captured action with the pre-stored action: If single actions and three-dimensional actions are mixed together, it is difficult for the recognition system to accurately determine the type of action. This will lead to a decrease in the recognition rate of the system for actions, and at the same time cause difficulties in subsequent processing. Secondly, it is difficult to effectively process different types of actions: There are significant differences in the motion patterns and characteristics between single actions and three-dimensional actions. If no classification processing is carried out, the recognition system will have difficulty effectively processing different types of actions. Finally, it increases the computing burden of the system: Without distinguishing between single actions and three-dimensional actions, the system needs to perform complete processing and comparison on all action data, which will increase the computing burden of the system and affect the efficiency and response speed of the system.

[0057] Therefore, in this technical solution, classifying the captured action data into single-plane actions and three-dimensional actions can help us more accurately identify and analyze different types of actions. The classification of the captured action data into single-plane actions and three-dimensional actions can be further understood as a step in the action recognition method. The purpose is to classify the action data according to its motion state for subsequent processing and recognition. Specifically, a single-plane action is a motion performed on the same plane. A single-plane action means that the action occurs on a plane. The classification and judgment of a single-plane action can be based on any axis to adapt to different types of training actions. For example, when measuring the range of motion of the elbow joint of the human body, the arm needs to make a horizontal swing within the same plane. As shown in Figure 2, the hand is lifted horizontally upward by 60°. For example, the measurement of the range of motion of single joints such as the wrist joint, elbow joint, ankle joint, and knee joint, and multi-joint actions such as side balance and push-ups. A three-dimensional action involves motion on three axes. A three-dimensional action means that the action occurs in a three-dimensional space. For example, the action of swinging the arm while running, the complete action of swinging the arm while playing golf, the action of lifting the arm from picking up a water cup to drinking it, etc. The arm of the human body swings upward and also has a horizontal swing. For example, squats, pull-ups, skipping rope, running, jumping, etc. Such a classification processing and comparison method can improve the accuracy and flexibility of action comparison, and can more accurately determine whether the user's action is consistent with the pre-stored action. Through accurate comparison and analysis, the system can give corresponding feedback and evaluation results to help the user improve the accuracy and effect of action execution.

[0058] When classifying, it can be distinguished by judging the number of coordinate systems involved in the action. For single-plane actions, usually only the coordinate system within one plane is involved, while for three-dimensional actions, multiple coordinate systems are involved. Therefore, the motion state can be judged by processing the action data, and then classified.

[0059] Specifically, in this application, for the classification of plane actions in the captured action data, it is judged whether the change amplitude of a certain axis is less than a certain threshold to determine whether the action is a plane action. If the change amplitude of a certain axis is less than the threshold, it means that the movement in the direction of this axis is basically translational, and this action can be classified as a plane action. For the classification of three-dimensional actions, it is judged whether the change amplitudes on the three axes are all greater than a certain threshold to determine whether the action is a three-dimensional action. If the change amplitudes of the three axes are all greater than the threshold, it means that the action has large movement changes in three-dimensional space, and this action can be classified as a three-dimensional action.

[0060] It should be noted here that the setting of the threshold needs to be adjusted according to the actual situation and requirements. The method used in this application is based on experience and trial-and-error method. The specific operation steps are as follows:

[0061] S11. Collect sufficient action data samples, including single-plane actions and three-dimensional actions, covering different types and difficulties of actions as much as possible.

[0062] S21. Preprocess the data and extract key feature parameters, such as starting position, ending position, angle, speed.

[0063] S31. Try different threshold settings, test each threshold, and record the test results, including the number of correctly recognized actions, the number of misrecognized actions, the number of missed recognized actions, etc.

[0064] S41. Analyze the test results, evaluate the performance and advantages and disadvantages of different thresholds, and select the optimal threshold as the set value.

[0065] S51. Fine-tune and optimize the threshold according to the actual situation and requirements.

[0066] The classified data can play a key role in the subsequent action recognition process, helping to improve the accuracy and reliability of action recognition.

[0067] After classification, the next step is to respectively judge the action data of the two major categories. For single-plane actions, comparing plane actions first can improve the algorithm efficiency and reduce the computational complexity. The principle will be described in detail below:

[0068] In the step of recording the starting position, ending position and angle of a single planar motion and comparing them with pre-stored motion data to determine the consistency of the motion, specifically, the recognition of a single planar motion requires recording the starting position, ending position and angle, and comparing them with pre-stored motion data to determine the consistency of the motion.

[0069] First, for a single planar motion, it is necessary to determine the plane in which the motion is located, such as a motion on a two-dimensional plane. Then, according to the captured motion data, record the starting position and ending position of the motion, and at the same time record the angle information. Secondly, for the pre-stored motion data, it is also necessary to record its starting position, ending position and angle information. Finally, compare the captured motion data with the pre-stored motion data to determine their consistency. This comparison can be carried out by various methods. Here, the single-node simple angle decomposition algorithm is used. The single-node simple angle decomposition algorithm is to convert the captured motion data into a series of coordinate points, record the position coordinates and motion direction angles of each point on the plane. Determine the starting and ending positions, and calculate the straight-line distance between the two points. Calculate the motion direction angles of the starting and ending points, and calculate the difference between them. Compare the difference with the corresponding pre-stored motion data to determine the similarity between the two motions. If the similarity is higher than the preset threshold, it is considered that the two motions are the same; otherwise, it is considered that they are different. It should be noted that the threshold mentioned here is also determined as described above, and will not be elaborated here. Generally speaking, the purpose of this step is to match the single planar motion with the pre-stored motion to determine whether the motion to be recognized is the same as the pre-stored motion.

[0070] Next, an example will be given. As shown in Figure 2, for a simple motion comparison, the pre-compared motion is only a single motion, such as raising the hand horizontally by 60°. In this case, the consistency of the two motions can be compared by calculating the angle from the same starting point to the ending point through the gyroscope of the wearable device.

[0071] More often, the motions to be compared are a series of complex spatial three-dimensional XYZ-axis time-sequence motions. For example, the pre-action to be compared is a complex three-dimensional motion of the arm. Finally, it is shown through the time-sequence coordinate axes of XYZ as Figure 3 shown.

[0072] At this time, it is no longer possible to compare the consistency of the pre-action and the motion captured on-site through a simple motion comparison method. For three-dimensional motions, the present invention compares them by gradually decomposing them into different planar coordinate sequences of three planes, namely the XY plane, the XZ plane and the YZ plane. Specifically, it is divided into three steps:

[0073] i. Divide the movement into different planes in three planes: the XY plane, the XZ plane, and the YZ plane

[0074] Coordinate sequence;

[0075] ii. Transform it into different sets of key points through the reverse change of the angle in the spatial direction of each axis;

[0076] iii. Record the comparison between the key point data and the pre-stored movement data to determine the consistency of the movement;

[0077] Specifically, i. Divide the movement into different plane coordinate sequences in three planes: For a movement in a three-dimensional space, it is necessary to first determine its projection on the three planes. Then, for each plane, record its projection coordinate sequence as a sequence and perform subsequent comparisons. For example, the projection on the XY plane can be obtained by keeping the Z-axis coordinate constant and using the remaining two coordinates as the projection on the XY plane.

[0078] ii. As Figure 6 shown, transform it into different Key point set Combined : For each plane coordinate sequence, it is necessary to transform it into a key point sequence. This can be achieved by recording the key points of the coordinate changes when the angle changes in the reverse direction of each axis. As Figure 4 shown, for example, for the coordinate sequence on the XY plane, the extreme points in the X-axis and Y-axis directions, as well as the mean points in the X-axis and Y-axis directions, can be recorded, and then these points are combined into a key point sequence.

[0079] iii. Record the comparison between the key point data and the pre-stored movement data to determine the consistency of the movement: Compare the key point sequence of each plane with the pre-stored movement data to determine the consistency of the movement. Specifically, here the movement comparison is achieved through the multi-node random movement decomposition algorithm. When comparing the key point sequences of the plane coordinate system, the Euclidean distance D and the included angle θ 合 are comprehensively used to evaluate the similarity of the two sequences. Among them, the Euclidean distance can be used to evaluate the distance difference between the two sequences, and the cosine of the included angle can be used to evaluate the direction difference between the two sequences.

[0080] Specifically, when using the Euclidean distance for comparison, it is necessary to calculate the distance between each key point in the two sequences, and then sum up all the distances after weighting to obtain the total distance. This total distance can be used to measure the similarity between the two sequences. The smaller the distance, the higher the similarity. For example, assume the captured action sequence is A and the pre-stored action sequence is B. For the key point sequences of each plane (XY, XZ, YZ), we can represent them as A {plane} and B {plane} . Taking the XY plane as an example, for the key point p i , whose coordinate is represented as (x i , y i ), then A xy = {(x1, y1), (x2, y2}),...,(x n , y n )}, and B xy Similarly.

[0081] Next, we need to calculate the distance d between A {plane} and B {plane} to evaluate the consistency of the two action sequences on this plane. One method for calculating the distance is the Euclidean distance, and its formula is:

[0082]

[0083] where n is the number of key points on this plane, are the coordinates of the i-th key point in A {plane} and B {plane} respectively.

[0084] We have three planes (XY, XZ, YZ), and we can use the following formula to calculate the total distance D:

[0085]

[0086] where d i is the distance on each plane, that is, the Euclidean distance between A {plane} and B {plane} .

[0087] Finally, we can use D as an index to evaluate the consistency of the action sequences A and B. If D is smaller, it means the consistency of the two action sequences on the three planes is higher.

[0088] For the comparison of angles, we can calculate through the following formula:

[0089]

[0090] In three planes, we can regard the sequence of key points in each plane as a vector, and use the above formula to calculate the angle between the corresponding vectors in each plane. Then we can take the average of the three angles to obtain the final angle difference value.

[0091] Specifically, let θ xy , θ xz and θ yz be the angles between vectors in the three planes respectively, then the final angle difference value θ can be expressed as:

[0092] θ 合 =(θ xy + θ xz + θ yz ) / 3

[0093] If the value of θ 合 is smaller, it indicates that the difference between the two actions is smaller, and their consistency is higher.

[0094] After calculating the Euclidean distance and the angle, they can be used as two indicators. Considering their magnitudes and relative importance simultaneously to comprehensively determine the consistency of actions. Set weights for the two indicators and perform weighted summation. In this application, the weight of the Euclidean distance is set to 0.7, and the weight of the angle is set to 0.3. After performing weighted summation, finally compare this result with a threshold. If this comprehensive indicator is less than a preset threshold, it is considered that this three-dimensional action is consistent with the pre-stored action, otherwise it is considered inconsistent. It should be noted here that during the process of capturing action comparison, we need to compare the captured action data with the pre-stored action data to determine their similarity. Since there may be certain noise and errors in the action data, a threshold needs to be set during the comparison, and the result of the similarity measurement is compared with this threshold. The indicators such as the Euclidean distance and the angle are used during the comparison because they can objectively measure the difference or similarity degree between two vectors, and have good mathematical properties and interpretability. By calculating these indicators, we can obtain a result of similarity measurement, but this result may fluctuate due to factors such as data noise and errors. Setting a reasonable threshold can filter out these fluctuations to a certain extent and ensure the accuracy of the captured action.

[0095] Therefore, during the comparison process, we generally compare the result of the similarity measurement between the captured action data and the pre-stored action data with the preset threshold. If the result of the similarity measurement is greater than or equal to the threshold, it is considered that the two actions are similar, otherwise it is considered that they are not similar.

[0096] Of course, the specific threshold can be adjusted according to the specific application scenario and experimental data. The adjustment method is as described above, and will not be elaborated here.

[0097] To achieve fast motion comparison, the present application also pre-records the change angle of each segment of the XYZ axis of each motion in the pre-stored motion data, rather than recapturing and calculating it every time a comparison is made. This can greatly reduce the comparison time and improve the response speed of the system. Therefore, after capturing and recording the pre-stored motion, the change angle of each segment of the XYZ axis is recorded in advance to enable quick comparison with the motion captured on-site.

[0098] Specifically, the pre-stored motion data is captured and recorded in advance. During the recording of each motion, the change angle of each segment of the XYZ axis captured by the gyroscope used is recorded. These recorded angle change data will become the key features of each motion for subsequent motion comparison.

[0099] Therefore, after capturing a new motion on-site, the system processes the motion, decomposes it into different plane coordinate sequences of the XYZ axis. Then, each plane coordinate sequence is converted into a set of key point sets through reverse transformation. Next, the key point data is compared with the key point data in the pre-stored motion data to determine which pre-stored motion the motion is most similar to.

[0100] The present application improves the accuracy and precision of motion recognition by classifying and comparing the captured motions. The main inventive point of the invention is to classify motions into single-plane motions and three-dimensional motions, and adopt different comparison methods for these two types of motions. For single-plane motions, the method of recording the starting position, ending position, and angle is used for comparison, while for three-dimensional motions, the method of decomposing the motion into three plane coordinate systems and recording key point data for comparison is adopted. In addition, the invention also provides a method for pre-storing motion data, that is, recording the change angle of each segment of the XYZ axis in advance for quick comparison.

[0101] This classification and comparison method can effectively improve the accuracy and stability of motion recognition, and can also adapt to more complex motion forms. Compared with traditional motion recognition methods, the invention can more accurately identify motions, reduce the misjudgment rate, and improve the accuracy and efficiency of recognition. In addition, the invention can also adapt to more complex motion forms and can improve the speed of on-site motion recognition, with high practical value.

[0102] Traditional action recognition methods usually only consider actions in a single plane and cannot recognize the details of three-dimensional actions, so the error is relatively large. However, this invention decomposes three-dimensional actions into different plane coordinate sequences in three planes for comparison, which can recognize actions more accurately. Secondly, it reduces the dependence on data length and direction: traditional action recognition methods usually use quantities with length and direction such as vectors for comparison, which makes the action recognition result vulnerable to the influence of data length and direction. This invention decomposes actions into coordinate sequences in three planes for comparison, avoiding this problem. Finally, it improves the real-time performance and efficiency of action recognition: after capturing and recording pre-stored actions, this invention records the change angles of each segment of the XYZ axes in advance so as to quickly compare with the actions captured on-site, thus improving the real-time performance and efficiency of action recognition.

[0103] In actual operation, people may perform actions in the wrong direction for various reasons, and these incorrect actions will affect the execution of subsequent actions, resulting in a lower completion rate of the final action. Therefore, this invention processes on-site actions, ignores the inconsistent parts in the front, and only focuses on the parts consistent with the pre-stored actions. This can avoid the influence of actions in the wrong direction on subsequent actions, improve the accuracy of action comparison, reduce the error rate, and thus improve the completion rate of the overall action.

[0104] Specifically, for example, when determining three-dimensional actions in this invention, the on-site actions are gradually decomposed into different plane coordinate sequences in three planes, namely the XY plane, the XZ plane, and the YZ plane, for comparison. On each plane, the actions are respectively divided into different plane coordinate sequences, and through the reverse change of the angle of each axis in the spatial direction, they are transformed into different key sets. Then, the key point data is recorded and compared with the pre-stored action data to determine the consistency of the actions. After capturing and recording the pre-stored actions, this invention records the change angles of each segment of the XYZ axes in advance so as to quickly compare with the actions captured on-site. As shown in Figure 7, before executing S4, according to the starting point where the on-site action is consistent with the pre-stored action, the inconsistent parts in the front of the on-site action are ignored. In this process, this invention comprehensively determines the consistency of the actions through the formulas for calculating the Euclidean distance and the included angle, as Figure 5 shown, only when a certain segment of the on-site action is in the same direction as segment A of the XYZ of the pre-stored action, does the segmented comparison start. This can exclude actions in the wrong direction, improve the accuracy of action comparison, and further improve the completion rate of the overall action. Of course, the same principle applies when comparing single-plane actions.

[0105] During the capture process, the time factor also needs to be considered in specific situations. For example, in some specific situations, the speed of the action is also a very important factor. For example, in the training of muscle tone, it is required that not only the actions are consistent, but also a certain speed is needed. In other specific situations, the time factor does not need to be considered. For example, for the situation where the patient's training actions are slow due to their own reasons, the doctor only needs to require the patient to do the actions slowly and correctly. Therefore, during the action capture process, we divide it into two modes: action capture with time requirements and action capture that ignores the time factor. For action capture with time requirements, the time of each segment of the XYZ axis with the pre-action and the on-site action is compared. Specifically, the length of time of each segment can be used as an additional dimension and incorporated into the comprehensive determination together with the Euclidean distance and the angle. In this way, the speed and accuracy of the action can be comprehensively evaluated, improving the training effect. For action capture that ignores the time factor, no time is carried on any segment on the XYZ axis for comparison, and only requires that the patient's action is consistent with the pre-action. In this way, the training can be made more flexible to meet the needs of different patients.

[0106] This application accurately compares the user's actions with the pre-stored actions and evaluates their consistency. By classifying the actions and combining the comparison of the starting position, ending position, angle, and the sequence of three-plane coordinates, the characteristics and changes of the actions can be analyzed more comprehensively, thereby improving the accuracy and reliability of action comparison. This method provides accurate action evaluation and feedback, helps users improve action execution skills and effects, and improves the training effect and efficiency.

[0107] The present invention also discloses a training action comparison system, including a pre-stored action module, an action data capture module, an action data classification module, a single-plane action comparison module, and a three-dimensional action comparison module.

[0108] In this embodiment, the functions of each module / unit are as follows:

[0109] The pre-stored action module is used to capture and record the pre-stored actions;

[0110] The action data capture module is used to capture the motion data of the action to be recognized;

[0111] The action data classification module is used to classify the captured action data into single-plane actions and three-dimensional actions;

[0112] The single-plane action comparison module is used to record the starting position, ending position, and angle of the single-plane action and compare it with the pre-stored action data to determine the consistency of the action;

[0113] A three-dimensional motion comparison module is configured to gradually decompose a three-dimensional motion into three different planar coordinate sequences and compare them with pre-stored motions.

[0114] The embodiments described above merely represent the implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the appended claims.

Claims

1. A training action comparison method, characterized in that, It includes the following steps: S1. Capture the motion data of the action to be recognized; S2. Classify the captured action data into single-plane actions and three-dimensional actions; S3. For single-plane actions, record the starting position, ending position, and angle, and compare them with the pre-stored action data to determine the consistency of the actions; S4. For three-dimensional actions, gradually decompose them into the XY plane, XZ plane, and YZ plane, and compare the three different plane coordinate sequences with the pre-stored actions; In step S2, according to whether the change amplitudes on the three axes are all greater than the threshold, classify the action as a single-plane action or a three-dimensional action; The step of gradually decomposing the three-dimensional action into the XY plane, XZ plane, and YZ plane and comparing the three different plane coordinate sequences includes the following steps: i. Divide the action into different plane coordinate sequences of the three planes, namely the XY plane, XZ plane, and YZ plane, respectively; ii. Transform it into different key-point sets through the reverse change of the angle of each axis in the spatial direction; iii. Record the key-point data and compare it with the pre-stored action data to determine the consistency of the actions.

2. The training action comparison method according to claim 1, characterized in that: Before step S1, capture and record the pre-stored actions, and record the change angle of each segment of the XYZ axis so as to quickly compare with the actions captured on-site.

3. The training action comparison method according to claim 2, wherein: The comparison between the single-plane action and the pre-stored action data is carried out through the single-node simple angle decomposition algorithm.

4. A training action comparison method according to claim 2, characterized in that: When executing S4, the Euclidean distance D and the included angle θ are used 合 to evaluate the similarity of two sequences, where the distance included angle θ 合 =(θ xy +θ xz +θ yz ) / 3, where 5. A training action comparison method according to any one of claims 1-4, characterized in that: Before executing S2, according to the starting point where the on-site action is consistent with the pre-stored action, ignore the inconsistent part in the front of the on-site action.

6. A training action comparison method according to any one of claims 1-4, characterized in that: In step S1, it is divided into action capture with time requirements and action capture that ignores time elements.

7. A training action comparison method according to claim 4, characterized in that: The Euclidean distance D combined with the included angle θ 合 The similarity of two sequences is evaluated by separately setting the weights of D and θ 合 and performing weighted summation for comparison and evaluation.

8. A training action comparison system, characterized in that, The system includes: An action data capture module for capturing the motion data of the action to be recognized; An action data classification module for classifying the captured action data into single-plane actions and three-dimensional actions; A single-plane action comparison module for recording the starting position, ending position, and angle of the single-plane action and comparing them with the pre-stored action data to determine the consistency of the actions; A three-dimensional action comparison module for gradually decomposing the three-dimensional action into the XY plane, XZ plane, and YZ plane and comparing the three different plane coordinate sequences with the pre-stored actions; The action data classification is to classify the action as a single-plane action or a three-dimensional action according to whether the change amplitudes on the three axes are all greater than the threshold; The step of gradually decomposing the three-dimensional action into the XY plane, XZ plane, and YZ plane and comparing the three different plane coordinate sequences includes the following steps: i. Divide the action into different plane coordinate sequences of the three planes, namely the XY plane, XZ plane, and YZ plane, respectively; ii. Transform it into different key-point sets through the reverse change of the angle of each axis in the spatial direction; iii. Record the key-point data and compare it with the pre-stored action data to determine the consistency of the actions.

Images