Method and system for identifying pull-up action posture

Abstract

The invention discloses a method for identifying a pull-up action posture and an identification system thereof, action posture data in a pull-up action is extracted based on a human body posture tracking algorithm, and a pull-up detection model is established based on a KNN algorithm, so that the counting efficiency of pull-up can be improved, and the real-time identification capability of the pull-up can be improved. The method comprises the steps that motion posture data in pull-up motion are used as training data, model training is conducted on a pull-up detection model, and the motion posture data are position data and motion track data of all joint points of a human body extracted through a human body posture tracking algorithm in the pull-up process; the trained pull-up detection model is used for recognizing and counting pull-up action postures, first recognition is completed, and the pull-up detection model is obtained by modeling distance changes between human body joint points on the basis of a KNN algorithm and combining a pre-designed maximum distance screening strategy and an average distance screening strategy.

Description

Technical Field

[0001] The present application relates to the field of image data processing technology, and in particular to a method for identifying pull-up postures and a recognition system thereof, and even a corresponding edge computing device. Background Art

[0002] Pull-ups are a form of exercise that is usually performed in a gym or sports venue. This exercise requires people to use their own body weight to practice on a horizontal bar at a fixed point. It is a training exercise that has high requirements for the core muscles, arms and shoulders. Through this exercise, you can strengthen your muscles, improve body control and flexibility.

[0003] In a competitive environment, pull-ups are also a competition event. Athletes need to complete the pull-up action as many times as possible within a specified time. In such competitions, it is usually necessary to identify and count the pull-ups to determine the athlete's competition results.

[0004] Usually in a competitive environment, the recognition and counting of pull-ups is realized by extracting human features based on RGB features. However, this method is easily affected by the environment and computing equipment, and has poor real-time performance, which greatly affects the efficiency and real-time processing capability of the pull-up recognition and counting system. Therefore, it is necessary to provide a method for pull-up recognition and counting with high counting efficiency and strong real-time processing capability. Summary of the invention

[0005] In order to solve the defects in the above-mentioned pull-up recognition and counting process, the present application provides a method for identifying the pull-up action posture and its recognition system, which extracts the action posture data in the pull-up action based on the human posture tracking algorithm and establishes a pull-up detection model based on the KNN algorithm, which can improve the counting efficiency of pull-ups and enhance its real-time recognition ability.

[0006] In a first aspect, the present application provides a method for identifying a pull-up action posture, which is implemented based on key point measurement between a person and an object, and includes:

[0007] The pull-up detection model is trained using the action posture data in the pull-up action as training data, wherein the action posture data is the position data and motion trajectory data of each joint point of the human body extracted by using the human posture tracking algorithm during the pull-up process;

[0008] The trained pull-up detection model is used to identify and count the pull-up action postures to complete the first level of recognition. The pull-up detection model is based on the KNN algorithm by modeling the distance changes between human joints, combined with the pre-designed maximum distance screening strategy and average distance screening strategy.

[0009] Optionally, in an implementation of the first aspect, using a trained pull-up detection model to perform real-time recognition and counting of pull-up action postures includes:

[0010] Based on the pairwise distances between joint points in the predefined posture joint list, the pull-up detection model is used to normalize the direction and size of the pull-up action posture;

[0011] After the normalization process is completed, the maximum distance screening strategy and the average distance screening strategy are used to perform real-time recognition and counting.

[0012] Optionally, in an implementation of the first aspect, the method further includes: using a real-time human posture estimation library to train a pull-up detection model to model the distance change between human joints and a specific object to identify the start and end points and standardization of the pull-up action posture to complete a second recognition, and the specific object can be used as a reference object for determining the standardization of the pull-up action posture.

[0013] Optionally, in an implementation of the first aspect, when the specific object is a horizontal bar, identifying the start and end of a pull-up action posture includes:

[0014] When the distances between the joints of both hands of the human body and the horizontal bar are less than a predetermined threshold, the action posture is determined to be the starting posture, otherwise, it is determined to be the ending posture;

[0015] When the specific object is a horizontal bar, identify the standard of pull-up posture, including:

[0016] When the chin of the human body passes through the horizontal bar and the distance between the joint points of the chin and the horizontal bar is zero, the movement posture is determined to be standard.

[0017] Optionally, in an implementation of the first aspect, the position data is specifically the three-dimensional pixel coordinates of the human body joints in each frame of data; the motion trajectory data is an array of three-dimensional image coordinates of all target joints to characterize the human body motion characteristics of the action performer in each frame.

[0018] It should be understood that in addition to the position data, this method will also collect the position coordinates of the four key points of the horizontal bar, namely the upper left, upper right, lower left, and lower right four key points of the horizontal bar. These four points are an array of two-dimensional image coordinates containing the key points of the horizontal bar, which are used to construct the two-dimensional image coordinates of the horizontal bar. Specifically, the RTMpose key point detection method can be used to detect and obtain the position data of the above four key points.

[0019] Optionally, in an implementation of the first aspect, the motion posture data is extracted from a picture or a real-time video stream, and the real-time video stream is captured in real time by a camera or input in real time from a video file.

[0020] In a second aspect, the present application also provides another method for identifying a pull-up action posture, which is based on key point measurement between a person and an object, and includes:

[0021] The pull-up detection model is trained using the action posture data in the pull-up action as training data, wherein the action posture data is the position data and motion trajectory data of each joint point of the human body extracted by using the human posture tracking algorithm during the pull-up process;

[0022] The pull-up detection model is trained using a real-time human posture estimation library to model the distance changes between human joints and specific objects to identify the start and end points and standardization of the pull-up posture to complete the second level of recognition. The specific object can be used as a reference to determine the standardization of the pull-up posture.

[0023] In a third aspect, the present application provides a system for identifying a pull-up posture, comprising:

[0024] Model training module, recognition and counting module;

[0025] The model training module is used to: use the action posture data in the pull-up action as training data to train the pull-up detection model, wherein the action posture data is the position data and motion trajectory data of each joint point of the human body extracted by using the human posture tracking algorithm during the pull-up process;

[0026] The recognition and counting module is used to: use the trained pull-up detection model to recognize and count the pull-up action postures to complete the first recognition, where the pull-up detection model is based on the KNN algorithm by modeling the distance changes between human joints, combined with the pre-designed maximum distance screening strategy and average distance screening strategy.

[0027] Optionally, in an implementation of the third aspect, the recognition system further includes: an action start and stop and standardization judgment module;

[0028] The action start and end and standardization judgment module is used to: use the real-time human posture estimation library to train the pull-up detection model to model the distance change between the human joints and specific objects to identify the start and end and standardization of the pull-up action posture to complete the second recognition. The specific object can be used as a reference for judging the standardization of the pull-up action posture.

[0029] In a fourth aspect, the present application provides an edge computing device on which is deployed the recognition system in the third aspect or its implementation method, and controls the recognition system to execute the method for recognizing pull-up movement posture as in the first aspect and any one of its implementation methods.

[0030] The technical solution provided by this application has the following technical effects:

[0031] The human posture tracking algorithm is used to extract the position data and motion trajectory data of each joint of the human body during the pull-up process as training data to train the pull-up detection model. The detection model is based on the KNN algorithm by modeling the distance changes between the human joints, combined with the pre-designed maximum distance screening strategy and average distance screening strategy; the human posture features extracted by the human posture tracking algorithm help the system to extract the most essential features of the pull-up action, that is, to conduct in-depth mining and analysis of the action posture data. The pull-up detection model established using the KNN algorithm can effectively detect the pull-up action accurately and efficiently, and realizes efficient and accurate recognition and counting of the human pull-up action, which can effectively improve the counting efficiency of pull-ups and its real-time recognition capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings exemplarily illustrate the embodiments and constitute a part of the specification, and together with the text description of the specification, are used to explain the exemplary implementation of the embodiments. The embodiments shown are for illustrative purposes only and do not limit the scope of the claims. In all drawings, the same reference numerals refer to similar but not necessarily identical elements.

[0033] Figure 1 A schematic diagram of a flow chart of a method for identifying a pull-up posture in an embodiment of the present application;

[0034] Figure 2 Another flowchart of the method for identifying the pull-up posture in the embodiment of the present application;

[0035] Figure 3 It is another flowchart diagram of the method for identifying the pull-up movement posture in the embodiment of the present application;

[0036] Figure 4 A structural schematic diagram of a recognition system for recognizing pull-up postures in an embodiment of the present application;

[0037] Figure 5 A schematic diagram of the structure of an edge computing device in an embodiment of the present application;

[0038] Figure 6 A schematic diagram of the positions of various joints of the human body in the embodiment of the present application;

[0039] Figure 7 This is a schematic diagram of the start of a pull-up movement in an embodiment of the present application;

[0040] Figure 8 A schematic diagram of counting pull-up movements in an embodiment of the present application;

[0041] Fig. 9 This is a schematic diagram of the end of the pull-up action in the embodiment of the present application. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0043] It should be noted that the descriptions involving "first", "second", etc. in the embodiments of the present application are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0044] In the description of the present application, it should be understood that the numerical labels before the steps do not indicate the order in which the steps are executed, but are only used to facilitate the description of the present application and to distinguish each step, and therefore should not be understood as a limitation on the present application.

[0045] In order to facilitate understanding of the technical solutions in the embodiments of the present application, the technical terms involved in the embodiments of the present application are briefly described below, as follows:

[0046] Med iapipe algorithm: is a real-time human posture tracking algorithm that can achieve high-precision detection and tracking of human posture, and can accurately detect the movement trajectory and posture changes of the human body, so it is widely used in many application scenarios, such as intelligent robots, virtual reality, human-computer interaction, etc.

[0047] KNN algorithm: K-Nearest Neighbors (KNN) is a supervised learning algorithm and a type of nonlinear discriminant analysis method. The basic idea of ​​the algorithm is: for a classification or regression problem, if the k nearest neighbors (i.e. similar samples) in the training set are known, the category of a new sample can be predicted based on their categories. If the new sample is similar to any of the nearest neighbors in the training set, the category to which the nearest neighbor belongs is used as the possible category of the new sample; otherwise, the least similar category to which the nearest neighbor belongs is used as the possible category of the new sample. The KNN algorithm is widely used in text classification, image recognition, speech recognition and other fields.

[0048] RTMpose (Real-Time Motion Prediction) is a real-time motion prediction technology. It can realize real-time recognition and tracking of human behavior by analyzing dynamic human postures in video or image sequences. RTMPose technology is widely used in human-computer interaction, intelligent monitoring, virtual reality and other fields, and can help users identify human actions and behaviors more accurately. The core of RTMPose technology is to detect the key points of the human body in multi-frame images or videos, and predict future actions based on these key points. These key points can be the motion trajectories of large joints such as ankles, knees, and hip joints. By modeling and optimizing the key points, human behavior can be predicted more accurately. In order to improve the prediction accuracy, RTMPose technology can also be combined with technical means such as deep learning and computer vision. It should be noted that in the embodiment of the present application, RTMpose mainly realizes the detection of the key points of the horizontal bar to construct a model of the two-dimensional image coordinates of the horizontal bar, and provides a model and data support for the recognition of the pull-up action posture.

[0049] The following is a further detailed description of the method for identifying pull-up postures and the technical solution of the identification system in the present application in conjunction with the specific embodiments.

[0050] Embodiment 1

[0051] The method for identifying pull-up movement postures in the present application can use a trained pull-up detection model to identify and count pull-up movement postures, and complete the first level of recognition alone.

[0052] Figure 1 A flow chart of a method for identifying pull-up movement posture in an embodiment of the present application.

[0053] like Figure 1 As shown, the method for identifying the pull-up posture in the embodiment of the present application includes:

[0054] S101. Use the motion posture data in the pull-up action as training data to perform model training on the pull-up detection model.

[0055] In step S101, the action posture data is the position data and motion trajectory data of each joint of the human body extracted by the human posture tracking algorithm during the pull-up process. These data contain various action posture information of the human body during the pull-up process, which can effectively reflect the movement state of the human body and are ideal training data for action recognition and counting. The human posture tracking algorithm can specifically be the Mediapipe algorithm. The human posture features extracted based on the Mediapipe algorithm help the system extract the most essential features of the pull-up action and can improve the accuracy of the system counting. And the Mediapipe algorithm performs well in low-latency and high-performance application scenarios.

[0056] Optionally, in one implementation of the present application, the position data is specifically the three-dimensional pixel coordinates of the human body joints in each frame of data; the motion trajectory data is an array of three-dimensional image coordinates of all target joints to characterize the human body motion characteristics of the action performer in each frame.

[0057] By using the Mediapipe algorithm to extract the human posture features in the pull-up action pictures at different angles in the training set, the extracted data is converted into an object containing joint point coordinates and pull-up posture samples. These posture features include the position and motion trajectory of each joint point of the human body, so as to fully capture the details of the pull-up action. Then, these feature data are used as a sample set to train the KNN algorithm model. In this way, the KNN algorithm can learn and recognize the standard pull-up action posture, providing a standard for action recognition and counting in the subsequent step S102.

[0058] The target joint points can be 33 joint points of the human body, which are distributed as follows: Figure 6 The joints shown in the figure are labeled as 0-32. The Mediapipe posture tracking algorithm can detect and track human posture data in real time. Mediapipe can track the changes of 33 joints of the human body in each frame, and finally use the three-dimensional pixel coordinates of each joint point to represent the node position, i.e. (X, Y, Z). Therefore, the present invention uses an array containing 33 tuples (X, Y, Z) to represent the human body features of the action taker in each frame.

[0059] Optionally, the action posture data is extracted from a picture or a real-time video stream, which is captured by a camera in real time or input from a video file in real time. The Mediapipe algorithm can process the video stream captured by a camera in real time or input from a video file in real time, and efficiently track the changes in the human joints in each frame. This is extremely important for building an accurate and efficient pull-up counting system.

[0060] S102: Use the trained pull-up detection model to identify and count the pull-up movement postures.

[0061] First, by inputting the action posture data to be recognized, the trained model is used to extract features and match patterns to obtain the probability distribution of each action posture; then, the occurrence frequency of each action posture is calculated based on the probability distribution, thereby realizing the recognition and counting of pull-up action postures.

[0062] In step S102, the pull-up detection model is obtained by modeling the distance changes between human joints based on the KNN algorithm, combined with the pre-designed maximum distance screening strategy and average distance screening strategy. The KNN algorithm is a method for finding data classification and clustering based on neighborhood similarity, which has good locality and globality, and has good classification and clustering effects for data sets containing multiple features. By combining the pre-designed maximum distance screening strategy and average distance screening strategy, the trained model can be more adaptable to actual application scenarios and has higher accuracy and stability.

[0063] Optionally, in one implementation of the embodiment of the present application, using a trained pull-up detection model to perform real-time recognition and counting of pull-up action postures includes:

[0064] Based on the pairwise distances between joint points in the predefined posture joint list, the pull-up detection model is used to normalize the direction and size of the pull-up action posture;

[0065] After the normalization process is completed, the maximum distance screening strategy and the average distance screening strategy are used to perform real-time recognition and counting.

[0066] Specifically, similarity judgment based on the pre-designed maximum distance screening strategy and average distance screening strategy: In order to use the KNN algorithm to realize the recognition of the pull-up action, it is necessary to process the posture data in each sample to be identified into a feature vector and compare it with the standard in step S201, so as to realize the recognition of the action. Therefore, the present invention uses the paired distances between the predefined posture joint lists, such as the distance between the wrist and the trunk, the ankle and the waist, and the distance between the two wrists. And normalize the direction and size. Then, the similarity judgment of the action is performed through the maximum distance screening and the average distance screening. The maximum distance screening is intended to filter out samples that are almost the same as the target sample but have only a few different values ​​in the feature vector.

[0067] Optionally, the maximum distance screening strategy screening may be specifically performed according to Formula 1, where Formula 1 is:

[0068] d s =min(max(|se|·w),max(|s-e'|·w));

[0069] In formula 1, for each sample s, the maximum distance between it and the given posture standard e and the flipped posture standard e′ is calculated, where |·| represents the absolute value, and then multiplied by the preset axis weight w. Then the maximum value of these distances is taken. After the maximum distance values ​​of all samples are calculated, these distances are sorted and the first N samples with the smallest distances are selected. s Go to the next step of screening.

[0070] Optionally, in an implementation of the embodiment of the present application, the maximum distance screening strategy may be used to screen by performing average distance screening according to Formula 2, where Formula 2 is:

[0071] μ s =min(mean(|s'-e|·w),(mean|s'-e'|·w));

[0072] In formula 2, for each remaining sample s', calculate the average distance between it and the given posture e and its flipped posture e'. mean represents taking the average value, multiplying it by the axis weight w, and then taking the average of these distances. After the average distance values ​​of all samples are calculated, these distances are sorted and the first M samples with the smallest distances are selected. s ,Finally, these samples are used for pull-up action recognition and counting.

[0073] In an embodiment of the present application, a human posture tracking algorithm is used to extract position data and motion trajectory data of various joints of the human body during the pull-up process as training data to train a pull-up detection model. The detection model is based on the KNN algorithm by modeling the distance changes between human joints, combined with a pre-designed maximum distance screening strategy and an average distance screening strategy. The human posture features extracted by the human posture tracking algorithm help the system to extract the most essential features of the pull-up action, that is, to conduct in-depth mining and analysis of the action posture data. The pull-up detection model established using the KNN algorithm can effectively detect the pull-up action accurately and efficiently, and realizes efficient and accurate recognition and counting of human pull-up actions, which can effectively improve the counting efficiency of pull-ups and its real-time recognition capability.

[0074] Embodiment 2

[0075] The method for identifying pull-up action postures in the present application can not only use the trained pull-up detection model to identify and count the pull-up action postures to complete the first level of recognition; it can also use the real-time human posture estimation library to train the pull-up detection model to model the distance changes between human joints and specific objects, and then identify the start and end points and standardization of the pull-up action postures to complete the second level of recognition.

[0076] Figure 2 Another flowchart of the method for identifying pull-up posture in an embodiment of the present application.

[0077] like Figure 2 As shown, the method for identifying the pull-up posture in the embodiment of the present application includes:

[0078] S201. Use the motion posture data in the pull-up action as training data to perform model training on the pull-up detection model.

[0079] S202: Use the trained pull-up detection model to identify and count the pull-up movement postures.

[0080] Steps S201 and S202 are respectively the same as steps S101 and S102 in the above-mentioned embodiment 1. For detailed descriptions thereof, reference may be made to the descriptions of the relevant parts in the above-mentioned embodiment 1, which will not be described in detail herein.

[0081] S203, using the real-time human posture estimation library to train the pull-up detection model to model the distance change between the human body joints and specific objects to identify the start and end points and standardization of the pull-up action posture to complete the second recognition.

[0082] In physical exercise and pull-up counting, the start, end and standard of the action are also extremely important judgment indicators. The start, end and standard judgment of the action based on the human-object distance information can be made by analyzing the distance between the key points of the human body and a specific object (such as a lever). Specifically, the pull-up motion detection model is trained using a real-time human posture estimation library, and the pull-up action posture is recognized by establishing a model of the distance change between the human joint points and a specific object, including the start and end of the action and the completion of the standard.

[0083] Optionally, in one implementation of the present application, when the specific object is a horizontal bar, identifying the start and end of the pull-up action posture includes: when the distance between each joint point of the human body's hands and the horizontal bar is less than a predetermined threshold, determining the action posture as the starting posture, otherwise, determining it as the ending posture. Figure 6 The joint points shown as 15, 16, 17, 18, 19, 20, 21, and 22.

[0084] Specifically, the distance between each joint point of the human hands and the horizontal bar can be calculated using Formula 3, and Formula 3 is as follows:

[0085]

[0086] In formula 3, p i =(x i ,y i ) is the coordinate of the i-th joint point, i∈{9, 10, 15, 16, 17, 18, 19, 20, 21, 22}, l is the equation of the straight line of the horizontal bar (connecting the upper left key point of the horizontal bar (x 1 ,y 1 ) and the upper right key point (x 2 ,y 2 ) of the straight line), distance(p i ,l) represents point p i The distance to the line l.

[0087] Optionally, in one implementation of the present application, when the specific object is a horizontal bar, identifying the standard of the pull-up posture includes: when the chin of the human body passes through the horizontal bar and the distance between the joint point of the chin of the human body and the horizontal bar is zero, determining that the posture is standard. The joint point of the chin can be Figure 6 The joint points shown in 9 and 10. In other words, when the chin joint passes through the single bar, the distance between the chin joint point and the single bar is zero, which can be used to determine whether the movement is standard during the exercise.

[0088] Figure 7Schematic diagram of the start of the pull-up action in the embodiment of the present application. Figure 7 As shown, when the athlete's hand holds the horizontal bar and the distance between the hand joint point and the horizontal bar is less than a threshold, it is determined that the action has started.

[0089] Figure 8 This is a schematic diagram of counting the pull-up action in the embodiment of the present application. The athlete's hands hold the horizontal bar, the chin joint passes through the bar, the action is standard, and KNN determines that the action is a pull-up, and the count is 1 or +1.

[0090] Fig. 9 This is a schematic diagram of the end of the pull-up action in the embodiment of the present application. When the athlete's hand leaves the bar and the distance between the hand joint and the horizontal bar is greater than the threshold, it is determined that the action is ended and the counting is terminated.

[0091] Figure 7 , Figure 8 and Fig. 9 The target frame of the horizontal bar part shown in the figure is a horizontal bar square frame detected by RTMpose key point detection. There are four key points at the four vertices of the target frame, that is, each frame has four key points: upper left, lower left, upper right and lower right. It is easy to understand that the array of the two-dimensional image coordinates of the horizontal bar can be identified through these four key points.

[0092] In the embodiment of the present application, the change in distance between the human joint point and the specific object is modeled in the pull-up detection model to identify the start and end and standard of the pull-up action posture, and complete the second level of action start and end and standard judgment; combined with the first level of identification completed by using the pull-up detection model to identify and count the pull-up action posture in Example 1, the double identification of pull-ups is finally achieved. Through this method, it can be ensured that the pull-up detection model can not only accurately identify the start and end of the pull-up, but also effectively evaluate the standard of the action, thereby improving the accuracy and reliability of the pull-up counting.

[0093] When the camera position is fixed, the position of the horizontal bar will not change. At this time, only the key point data of the horizontal bar in the first frame can be extracted, thereby improving the inference speed of the model.

[0094] Embodiment 3

[0095] In the method of identifying the pull-up action posture in the present application, a real-time human posture estimation library can be used to train a pull-up detection model to model the distance changes between human joints and specific objects, thereby identifying the start and end points and standardization of the pull-up action posture, and completing the second level of recognition independently.

[0096] Figure 3 This is another flowchart of the method for identifying the pull-up movement posture in the embodiment of the present application.

[0097] like Figure 3 As shown, the method for identifying the pull-up posture in the embodiment of the present application includes:

[0098] S301. Use the motion posture data in the pull-up action as training data to perform model training on the pull-up detection model.

[0099] Step S301 is the same as step S101 in the above-mentioned embodiment 1. The detailed description thereof can be found in the description of the relevant part in the above-mentioned embodiment 1, and will not be described in detail here.

[0100] S302. Use the real-time human posture estimation library to train a pull-up detection model to model the distance change between the human joints and specific objects, so as to recognize the start and end points and standardity of the pull-up action posture to complete the second recognition.

[0101] Step S302 is the same as step S203 in the above-mentioned embodiment 1. The detailed description thereof can be found in the description of the relevant part in the above-mentioned embodiment 1, and will not be described in detail here.

[0102] In the embodiment of the present application, it can be ensured that the pull-up detection model can not only accurately identify the start and end of the pull-up, but also effectively evaluate the standardization of the movement.

[0103] Embodiment 4

[0104] In summary, the technical solutions in the above method embodiments are also designed in the embodiments of the present application to identify the pull-up action posture recognition system, which has the following advantages:

[0105] 1. The system has high accuracy and strong robustness: The system uses KNN's maximum distance screening and average distance screening, combined with the distance changes between the human body's joints and the horizontal bar during exercise, to design a dual recognition and counting system, providing high-precision judgment of the start and end of the action. Specifically, in the pull-up action, the system accurately judges the start and end of each action by analyzing the distance between the key points of the human body and the horizontal bar, ensuring the accuracy of the counting. The maximum distance screening can filter out samples that are almost the same as the target sample but have only a few different values ​​in the feature vector, and the average distance screening ensures that the overall posture is consistent with the standard sample. Therefore, the present invention can effectively reduce the misjudgment caused by prediction jitter and ensure the stability of the counting. By combining the human body posture characteristics and the key point characteristics of the horizontal bar, the system has a stronger anti-interference ability against interference caused by factors such as light changes and complex backgrounds, which significantly improves the robustness of the system. This design enables the system to maintain high accuracy and stable performance in different environments;

[0106] Second, this system does not require users to wear additional equipment or use other sensors. It only needs the camera to capture the joints of the human body and the key points of the horizontal bar to judge and count movements. This design not only improves the naturalness and convenience of the user experience, but also makes the system suitable for more usage scenarios. Whether at home, in the gym or outdoors, it can easily count and evaluate pull-ups. In addition, the system evaluates the standardization of each pull-up through human-object distance calculation, and can provide real-time feedback to help users improve the quality of their movements and achieve better exercise results.

[0107] Figure 4 A structural schematic diagram of a recognition system for recognizing pull-up movement posture in an embodiment of the present application.

[0108] like Figure 4 As shown, the recognition system 40 for recognizing the pull-up posture in the embodiment of the present application includes:

[0109] Model training module 401, recognition and counting module 402;

[0110] The model training module 401 is used to: use the action posture data in the pull-up action as training data to perform model training on the pull-up detection model, wherein the action posture data is the position data and motion trajectory data of each joint point of the human body extracted by using the human posture tracking algorithm during the pull-up process;

[0111] The recognition and counting module 402 is used to: use the trained pull-up detection model to recognize and count the pull-up action postures to complete the first recognition, wherein the pull-up detection model is based on the KNN algorithm by modeling the distance changes between human joints, combined with the pre-designed maximum distance screening strategy and average distance screening strategy.

[0112] Optional, such as Figure 4 As shown in the dotted line part, the recognition system 40 in the embodiment of the present application not only includes: a model training module 401, a recognition and counting module 402, but also includes: an action start and stop and standardization judgment module 403;

[0113] The action start and end and standardization judgment module 403 is used to: train the pull-up detection model using the real-time human posture estimation library, and to model the distance change between the human joints and the specific object to identify the action start and end and standardization of the pull-up action posture to complete the second level of recognition, and the specific object can be used as a reference for judging the standardization of the pull-up action posture. This system can simultaneously complete the first level and the second level of dual recognition.

[0114] In addition, in other embodiments of the present application, the recognition system 40 may also include a model training module 401, an action start and end and standardization judgment module 403, but does not include a recognition and counting module 402. This system can complete the second level of recognition alone.

[0115] Embodiment 5

[0116] Figure 5 A structural diagram of an edge computing device in an embodiment of the present application.

[0117] like Figure 5 As shown, the edge computing device 50 in the embodiment of the present application includes: a memory 501 and a processor 502, wherein the recognition system is deployed in the memory 501, and the processor 502 executes the method for recognizing the pull-up posture in the above embodiment by calling the recognition system deployed in the memory 501. Among them, the edge computing device 50 can be a smart camera with an AI chip, etc.

[0118] In the edge computing device 50, the memory 501 includes at least one type of computer-readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (such as SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 501 can be an internal storage module of the edge computing device 50, such as a hard disk or memory of the edge computing device 50. In other embodiments, the memory 501 can also be an external storage device of the edge computing device 50, such as a plug-in hard disk equipped on the edge computing device 50, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Of course, the memory 501 can also include both the internal storage module of the edge computing device 50 and its external storage device. In this embodiment, the memory 501 is generally used to store the operating system and various application software installed on the edge computing device 50, such as the program code of the bullet screen display method, etc. In addition, the memory 501 can also be used to temporarily store various types of data that have been output or are to be output.

[0119] The processor 502 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other chips in some embodiments. The processor 502 is generally used to control the overall operation of the edge computing device 50, such as performing control and processing related to data interaction or communication with the edge computing device 50. In this embodiment, the processor 502 is used to run the program code stored in the memory 501 or process data.

[0120] In addition, in some special edge computing devices, a network interface is also provided on the edge computing device, and the network interface may include a wireless network interface or a wired network interface, which is generally used to establish a communication link between the edge computing device 50 and other computer devices. For example, the network interface is used to connect the edge computing device 50 to an external terminal through a network, and to establish a data transmission channel and a communication link between the edge computing device 50 and the external terminal. The network may be an intranet, the Internet, the Global System of Mobile communication (GSM), Wideband Code Division Multiple Access (WCDMA), 4G network, 5G network, Bluetooth, Wi-Fi and other wireless or wired networks.

[0121] Obviously, those skilled in the art should understand that each module or each step of the above-mentioned embodiment of the present application can be implemented with a general-purpose computer device, which can be concentrated on a single computer device, or distributed on a network composed of multiple computer devices. Optionally, they can be implemented with a program code executable by a computer device, so that they can be stored in a storage device and executed by a computer device, and in some cases, the steps shown or described can be performed in an order different from that herein, or they can be made into each integrated circuit module respectively, or multiple modules or steps therein can be made into a single integrated circuit module to implement. In this way, the embodiment of the present application is not limited to any specific hardware and software combination.

[0122] It should be noted that the above are only preferred embodiments of the present application, and the patent protection scope of the present application is not limited thereto. Any equivalent structure or equivalent process transformation made using the contents of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for identifying pull-up postures, the method being based on key point measurement between a person and an object, characterized in that: The method comprises: Using the action posture data in the pull-up action as training data, the pull-up detection model is trained, wherein the action posture data is the position data and motion trajectory data of each joint point of the human body extracted by using the human posture tracking algorithm during the pull-up process; The first level of recognition is completed by using a trained pull-up detection model to identify and count the pull-up action postures, wherein the pull-up detection model is based on the KNN algorithm by modeling the distance changes between human joints, combined with a pre-designed maximum distance screening strategy and an average distance screening strategy.

2. The method according to claim 1, characterized in that The method of using the trained pull-up detection model to perform real-time recognition and counting of pull-up action postures includes: Based on the paired distances between the joint points in the predefined posture joint list, the pull-up detection model is used to normalize the direction and size of the pull-up action posture; After the normalization process is completed, the maximum distance screening strategy and the average distance screening strategy are used in turn to perform real-time recognition and counting.

3. The method according to claim 1 or 2, characterized in that: The method also includes: using a real-time human posture estimation library to train the pull-up detection model to model the distance change between the human body joints and a specific object to identify the start and end of the pull-up action posture and the standardization to complete the second recognition, wherein the specific object can be used as a reference object for determining the standardization of the pull-up action posture.

4. The method according to claim 3, characterized in that When the specific object is a horizontal bar, the step of identifying the start and end of the pull-up posture includes: When the distances between the joints of both hands of the human body and the horizontal bar are less than a predetermined threshold, the action posture is determined to be the starting posture, otherwise, it is determined to be the ending posture; When the specific object is a horizontal bar, the standard of identifying the pull-up posture includes: When the chin of the human body passes through the horizontal bar and the distance between the joint points of the chin and the horizontal bar is zero, the movement posture is determined to be standard.

5. The method according to claim 1 or 2, characterized in that: The position data is specifically the three-dimensional pixel coordinates of the human joint points in each frame of data; the motion trajectory data is an array of three-dimensional image coordinates of all target joint points to characterize the human motion characteristics of the person performing the action in each frame.

6. The method according to claim 5, characterized in that The motion posture data is extracted from a picture or a real-time video stream, and the real-time video stream is collected in real time by a camera or input in real time by a video file.

7. A method for identifying pull-up postures, the method being based on key point measurement between a person and an object, characterized in that: The method comprises: Using the action posture data in the pull-up action as training data, the pull-up detection model is trained, wherein the action posture data is the position data and motion trajectory data of each joint point of the human body extracted by using the human posture tracking algorithm during the pull-up process; The pull-up detection model is trained using a real-time human posture estimation library to model the distance change between human joints and a specific object to identify the start and end points and standardization of the pull-up action posture to complete the second level of recognition. The specific object can be used as a reference object for determining the standardization of the pull-up action posture.

8. A system for identifying pull-up postures, characterized in that: include: Model training module, recognition and counting module; The model training module is used to: use the action posture data in the pull-up action as training data to perform model training on the pull-up detection model, wherein the action posture data is the position data and motion trajectory data of each joint point of the human body extracted by using the human posture tracking algorithm during the pull-up process; The recognition and counting module is used to: use the trained pull-up detection model to recognize and count the pull-up action postures to complete the first recognition, wherein the pull-up detection model is based on the KNN algorithm by modeling the distance changes between human joints, combined with a pre-designed maximum distance screening strategy and an average distance screening strategy.

9. The identification system according to claim 8, characterized in that: The recognition system also includes: an action start and stop and standardization judgment module; The action start and end and standardization judgment module is used to: use the real-time human posture estimation library to train the pull-up detection model to model the distance change between the human body joints and a specific object to identify the start and end and standardization of the pull-up action posture to complete the second recognition. The specific object can be used as a reference object for judging the standardization of the pull-up action posture.

10. An edge computing device, on which is deployed a recognition system as described in claim 8 or 9 above, and the recognition system executes the method for recognizing pull-up movement posture as described in any one of claims 1-7 above.

Images