Human body posture recognition model construction method and motion posture evaluation method and system

Abstract

The invention firstly discloses a human body posture recognition model construction method, which comprises the following steps: acquiring a disclosed human body posture data set, and performing preprocessing; on the basis of a YOLOv8-pose network framework, an improved SPPF and a deformable convolution DCNv4 are introduced, detection of a human body in an image and estimation of a posture are realized by combining a target frame and key point prediction, and an improved human body posture recognition network framework is obtained; and repeatedly training the improved human body posture recognition network framework according to the human body posture data set until the recognition rate reaches 90% or above, and obtaining a final human body posture recognition model. The improved human body posture recognition model is constructed through the construction method, and the problems that information of a pooling layer is lost, the number of parameters of the model is increased by using large kernel convolution, the calculation complexity is increased, and overfitting and performance reduction are caused are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of target detection and recognition, and in particular to a method for constructing a human posture recognition model, and a method and system for evaluating motion posture. Background Art

[0002] With the popularization of national fitness, exercise has become an indispensable part of life, especially in the field of sports. In high-intensity training, it is necessary to maintain a good posture to avoid unnecessary sports injuries. At present, the recognition and evaluation of sports postures mostly rely on manual recognition and judgment by experts. There are few relatively accurate sports posture recognition and evaluation systems on the market, which not only consumes manpower costs, but also inevitably cannot guarantee the accuracy and fairness of the judgment due to individual differences in the process of manual recognition and judgment.

[0003] In recent years, with the rapid development of artificial intelligence technology, the field of human posture estimation has also received widespread attention, especially the popularization and application of the YOLO series of target detection algorithms, and the YOLO-POSE series has also come into being.

[0004] YOLOv8-pose is a new method for human pose estimation model, which is a two-dimensional multi-person pose estimation based on the popular YOLO target detection framework. YOLOv8-pose adopts a more efficient and accurate feature pyramid construction method - SPPF, which improves the inference speed of the model by reducing computational redundancy while maintaining the network's ability to effectively identify targets of different sizes.

[0005] However, the pooling layer in this model has the problem of information loss, especially in the process of dimensionality reduction, some detailed information may be lost, resulting in a decrease in the ability to accurately locate the target. At the same time, the characteristic of traditional large-kernel convolution is to expand the perception field of view by increasing the size of the convolution kernel in order to better capture the contextual information and global features of the target. Although this helps to improve the model's accuracy in identifying and locating targets, the use of large-kernel convolution will also increase the number of parameters and computational complexity of the model, which may lead to overfitting and performance degradation. Therefore, although the traditional pooling layer and large-kernel convolution have obvious advantages in the YOLO framework, their shortcomings cannot be ignored. Summary of the invention

[0006] The present invention aims to solve one of the technical problems existing in the related art at least to a certain extent.

[0007] The primary purpose of the present invention is to provide a method for constructing a human posture recognition model, to construct an improved human posture recognition model, to solve the problem of information loss in the pooling layer, and to use large kernel convolution to increase the number of model parameters and computational complexity, leading to overfitting and performance degradation.

[0008] Another object of the present invention is to provide a human body motion posture evaluation method based on the constructed human body posture recognition model, so as to solve the problem of lack of professional and accurate motion posture evaluation methods in the field of sports, reduce the cost of sports evaluation, and ensure the fairness and accuracy of the evaluation.

[0009] Another object of the present invention is to provide a motion posture evaluation system to apply the above-mentioned motion posture evaluation method.

[0010] Based on the above-mentioned motion posture evaluation method, the present invention also provides a carrier thereof, including a non-transitory computer-readable storage medium, an electronic device and a computer program product.

[0011] In order to achieve the above-mentioned object, the present invention provides a method for constructing a human posture recognition model, comprising:

[0012] Obtain public human posture datasets and perform preprocessing;

[0013] Based on the YOLOv8-pose network framework, the improved spatial pyramid pooling module SPPF and deformable convolution DCNv4 are introduced, and the target box and key point prediction are combined to realize the detection and posture estimation of the human body in the image, and obtain an improved human posture recognition network framework;

[0014] The improved human posture recognition network framework is repeatedly trained according to the human posture dataset until the recognition rate reaches more than 90%, and the final human posture recognition model is obtained.

[0015] A further preferred technical solution of the present invention is that the human body posture data set is divided into a training set, a validation set and a test set;

[0016] Train the human posture recognition network framework with the training set and adjust the network weights;

[0017] Use the validation set to adjust the hyperparameters of the human posture recognition model;

[0018] The recognition rate of the human posture recognition model is evaluated using the test set.

[0019] Preferably, the improved human posture recognition network framework introduces the SP_DCNv4 module combining SPPF and DCNv4 in the Backbone layer as a convolutional space pyramid pooling layer. After receiving the feature map from the previous layer as input, this layer performs maximum pooling on the feature map with three pooling kernels of the same size, and then DCNv4 splices the outputs of different pooling layers to form a feature vector of a fixed length as the input of subsequent fully connected layers or other network layers.

[0020] Another aspect of the present invention provides a method for evaluating a motion posture, comprising:

[0021] Collect key movements that need to be evaluated when standard personnel use standard postures to perform sports; and construct a standard sports posture information matrix based on the basic information of the standard personnel and the key information extracted from the key movements of the standard personnel;

[0022] Collect the key movements that need to be evaluated when the subjects are exercising; and construct the subject's movement posture information matrix based on the subject's basic information and the key information extracted from the subject's key movements;

[0023] The final human posture recognition model constructed by the human posture recognition model construction method performs posture recognition on the key movements of the standard person and the key movement images of the tested person, respectively, to obtain key joint annotation information; the obtained annotation information is combined with the standard motion posture information matrix and the tested person's motion posture information matrix to obtain the standard feature matrix and the tested feature matrix;

[0024] According to the standard feature matrix and the tested feature matrix, the key joint angle similarity comparison method is used to perform motion posture matching and complete the motion posture evaluation.

[0025] Preferably, the key movements of the standard personnel that need to be evaluated are collected when they use standard postures to perform sports; and a standard sports posture information matrix is ​​constructed based on the basic information of the standard personnel and the key information extracted from the key movements of the standard personnel; including:

[0026] Obtain the height data of standard personnel and record it as Hs;

[0027] When a standard person performs a sport in a standard posture, a camera is used to capture images of the key movements that need to be evaluated, and a real number set is obtained. Extract the number of key actions that need to be evaluated in the movement, denoted as P, the number of joints, denoted as G, and the joint information, denoted as S;

[0028] Constructing standard motion posture information matrix

[0029] The method collects key movements of the tested person that need to be evaluated when the tested person is exercising, and constructs a movement posture information matrix of the tested person based on the basic information of the tested person and the key information extracted from the key movements of the tested person; including:

[0030] Obtain the height data of the person being tested and record it as Hc;

[0031] When the subject is exercising, use a camera to capture images of the key movements that need to be evaluated, and obtain another real number set Extract the number of joints that need to be evaluated in the movement, denoted as G, and the joint information, denoted as S;

[0032] Construct the motion posture information matrix of the tested person

[0033] Preferably, the final human posture recognition model constructed by the above-mentioned human posture recognition model construction method performs posture recognition on the key movements of the standard person and the key movement images of the tested person respectively to obtain key joint annotation information; the obtained annotation information is combined with the standard motion posture information matrix and the tested person motion posture information matrix to obtain the standard feature matrix and the tested feature matrix; including:

[0034] Use the final human posture recognition model to estimate and recognize the human posture of the key action images of the standard person and the person under test, and obtain the joint point annotation information of each image;

[0035] Using the obtained joint point annotation information, calculate the standard motion posture information matrix Each key joint angle θ in; According to the key joint angle θ and the standard motion posture information matrix Construct the standard feature matrix P, expressed as:

[0036] P=[p 1 ,p 2 ,p 3 ,…,p i ,…,p n ]

[0037] Where n is the number of key action images, p i is the key angle feature set of the i-th frame of the sequence, expressed as:

[0038] p i =[θ 1 ,θ 2 ,θ 3 ,…θ j ]

[0039] θ j is the angle of the jth key joint;

[0040] The obtained joint point annotation information is used to calculate the motion posture information matrix of the tested person Each key joint angle θ in the test; according to the key joint angle θ and the motion posture information matrix of the tested person Construct the tested feature matrix T, expressed as:

[0041] T=[t 1 ,t 2 ,t 3 ,…,ti ,…,t n ]

[0042] Where n is the number of key action images, t i is the key angle feature set of the i-th frame of the sequence, expressed as:

[0043] t i =[θ 1 ,θ 2 ,θ 3 ,…θ j ]

[0044] θ j is the angle of the jth key joint.

[0045] Preferably, the method of matching the motion postures and completing the motion posture evaluation is performed by using a key joint angle similarity comparison method based on the standard feature matrix and the tested feature matrix; including:

[0046] According to the standard feature matrix P and the tested feature matrix T, posture matching is performed on each action of the tested person. The calculation formula of posture matching is expressed as:

[0047]

[0048] Among them, ||·|| means finding the Euclidean norm of a vector;

[0049] Analyze each key action, assuming that the i-th joint angle of the tested person is b i , the corresponding standard action is obtained through posture matching as p pose , the corresponding angle of the i-th joint is p pose,i , get the deviation value calculation formula S i , expressed as:

[0050]

[0051] Calculate the corresponding joint score g i , expressed as:

[0052]

[0053] Finally, the average score is calculated based on the scores of each joint as the total score G:

[0054]

[0055] Where n is the total number of key joints;

[0056] Complete the assessment of movement posture.

[0057] In another aspect, the present invention provides a sports posture assessment system, characterized in that it includes:

[0058] The data collection and processing module is used to collect the key movements that need to be evaluated when the standard personnel use the standard posture to exercise; and to construct the standard movement posture information matrix based on the basic information of the standard personnel and the key information extracted from the key movements of the standard personnel;

[0059] At the same time, the key movements that need to be evaluated when the subjects are exercising are collected; and the movement posture information matrix of the subjects is constructed based on the basic information of the subjects and the key information extracted from the key movements of the subjects;

[0060] The model building and posture recognition module is used to build the final human posture recognition model using the human posture recognition model building method mentioned above; and perform posture recognition on the key movements of the standard person and the key movement images of the tested person respectively to obtain key joint annotation information; combine the obtained annotation information with the standard motion posture information matrix and the tested person motion posture information matrix to obtain the standard feature matrix and the tested feature matrix;

[0061] The posture matching and output module is used to match the motion posture according to the standard feature matrix and the tested feature matrix, adopt the key joint angle similarity comparison method, complete the motion posture evaluation, and output the evaluation results.

[0062] In yet another aspect of the present invention, a non-transitory computer-readable storage medium is provided, on which computer instructions are stored, and the computer instructions enable a computer to execute the above-mentioned motion posture evaluation method.

[0063] In another aspect, the present invention provides an electronic device, comprising: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus, and the processor calls the logic instructions in the memory to execute the above-mentioned motion posture evaluation method.

[0064] In yet another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer executes the above-mentioned motion posture evaluation method.

[0065] Beneficial effects: The present invention optimizes the YOLOv8POSE network and introduces the SF_DCNv4 module. SP_DCNv4 has strong feature extraction and representation learning capabilities. Through deep convolution operations, a deeper understanding of the image can be achieved, thereby improving the model's ability to abstract image features, as well as its ability to extract image features and its perception range, while also improving the model's processing speed. Traditional pooling operations are usually performed at a fixed scale, which may result in some information loss on input images of different scales. SP_DCNV4 can pool input images at different scales, thereby maintaining local invariance and being able to adapt to inputs of different scales. At the same time, it can also extract certain information from the background area, greatly improving the accuracy of the model, and improving the flexibility and generalization of the model.

[0066] The present invention proposes a motion posture evaluation method and system based on an improved YOLOv8-pose model, which can evaluate a variety of motion postures, has universality, can perform professional evaluation on all motions, and has strong adaptability and adjustment capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A flowchart of a method for constructing a human posture recognition model according to the present invention;

[0068] Figure 2 It is a structural diagram of the SP_DCNv4 module introduced in the present invention;

[0069] Figure 3 It is a structural diagram of the improved human posture recognition network framework of the present invention;

[0070] Figure 4 is a flow chart of the motion posture evaluation method of the present invention;

[0071] Figure 5 is a point map of 17 key points of characters in an embodiment of the present invention;

[0072] Figure 6 8 key joint angle diagrams in an embodiment of the present invention. DETAILED DESCRIPTION

[0073] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0074] Combine the following Figure 1-Figure 6 The present invention describes a method for constructing a human posture recognition model, a motion posture evaluation method and system, a medium, an electronic device and a computer program.

[0075] Embodiment 1: This embodiment provides a method for constructing a human posture recognition model. Figure 1 As shown, the specific steps are:

[0076] S100: Obtain a public human posture dataset and perform preprocessing.

[0077] The public COCO2017 dataset is used in this embodiment. The dataset contains 160,000 images marked with key points for pose estimation tasks. The dataset supports 17 key points of people, which facilitates detailed pose estimation. The training set, test set, and validation set in the dataset are 118287, 40670, and 5000 respectively. The data is re-divided into training set, test set, and validation set according to 8:1:1, and the training weight is increased.

[0078] The 17 key points of the characters are: nose, left eye, right eye, left ear, right ear, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left hip, right hip, left knee, right knee, left ankle and right ankle.

[0079] S200, a network framework based on YOLOv8-pose, introduces the improved spatial pyramid pooling module SPPF and deformable convolution DCNv4.

[0080] This step is mainly used to improve the network framework, realize feature extraction and analysis, and combine the target box and key point prediction to realize the detection and posture estimation of the human body in the image.

[0081] The improved YOLOv8-pose network structure introduces the SP_DCNv4 module. The SP_DCNv4 module structure diagram is as follows Figure 2As shown in the figure, it combines SPPF and deformable convolution DCNv4. After receiving the feature map from the previous layer as input, SP_DCNv4 uses three pooling kernels of the same size (5×5, 5×5, 5×5) to perform maximum pooling on the feature map. Then, the deformable convolution network DCNv4 adaptively adjusts the outputs of different pooling layers according to different target scales, so as to more accurately capture features on objects of different scales for splicing, forming a fixed-length feature vector as the input of subsequent fully connected layers or other network layers.

[0082] DCNv4 uses adaptive aggregation windows and dynamic aggregation weights with unbounded value ranges to process multiple channels in the same group in one thread, which share sampling offsets and aggregation weights. This reduces workloads such as memory reads and bilinear interpolation coefficient calculations, and can merge multiple memory access instructions.

[0083] The improved YOLOv8-pose network structure is as follows Figure 3 As shown in the figure, it consists of the following three parts: Backbone layer, Neck layer and Head layer. The Backbone layer contains 5 convolution blocks, 4 C2f modules and SP-DCNv4 modules, which are mainly used for feature extraction. Residual connections and bottleneck structures are introduced to reduce network complexity and improve performance. The Neck layer adopts multi-scale feature fusion technology to fuse feature maps at different stages to enhance feature representation capabilities. The Head layer includes a detection head and a classification head. The detection head consists of a series of convolutional layers and deconvolutional layers to generate detection results, and the classification head uses global average pooling to classify each feature map to achieve the final target detection and classification tasks.

[0084] S300, repeatedly training the improved human posture recognition network framework according to the human posture data set until the recognition rate reaches more than 90%, thereby obtaining a final human posture recognition model.

[0085] This step is used for model training. The improved YOLOv8-pose model is trained and tested based on the divided data set to obtain the network weights. According to the changes in the model parameters during the training process, the learning rate is adjusted to control the convergence and stability of the training process. The performance of the improved YOLOv8-pose model on the validation set is evaluated, and the weights are evaluated, including the calculation of accuracy, recall rate, and mAP (mean average precision); until the recognition rate reaches more than 90%, the weight at this time is the final weight, and the final human posture recognition model is obtained for subsequent deployment and application.

[0086] Embodiment 2: This embodiment provides a method for evaluating a motion posture. The specific steps of the method are as follows: Figure 4 As shown, including:

[0087] S400, collecting key movements that need to be evaluated when standard personnel use standard postures to exercise; and constructing a standard movement posture information matrix based on the basic information of the standard personnel and the key information extracted from the key movements of the standard personnel. Specifically:

[0088] First, enter the name, age, height, weight, gender, etc. of the standard personnel, among which height is recorded as Hs in the important index column;

[0089] When a standard person performs a sport in a standard posture, a camera is used to capture images of the key movements that need to be evaluated, and a real number set is obtained. To extract useful information from the captured video, we first need to determine how many key actions are needed to make a complete motion process, and the number is recorded as P; then determine the number of joints in the motion, such as Figure 5 There are 17 points shown, numbered from 0 to 16, namely nose, left eye, right eye, left ear, right ear, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left hip, right hip, left knee, right knee, left ankle, right ankle, denoted as G = 17; finally, the parameters required to store the key point information are determined, which are the joint number, horizontal coordinate and vertical coordinate, denoted as S = 3, so the standard motion posture information matrix is ​​composed of the joint point information

[0090] S500, collecting key movements of the tested person that need to be evaluated when the tested person is exercising; and constructing a movement posture information matrix of the tested person based on the basic information of the tested person and the key information extracted from the key movements of the tested person. Specifically:

[0091] The same steps as for standard athletes are used to input the name, age, height, weight, gender, etc. of the person being tested, with height recorded as Hc in the important index column; the person being tested uses the same movement posture to exercise, and records the data of the key movements that need to be evaluated. Extract key information from the recorded exercise videos of the person being tested, and construct the information matrix of the person's movement posture

[0092] S600: The final human posture recognition model constructed in Example 1 performs posture recognition on the key movements of the standard person and the key movement images of the tested person, respectively, to obtain key joint annotation information; the obtained annotation information is combined with the standard motion posture information matrix and the tested person's motion posture information matrix to obtain the standard feature matrix and the tested feature matrix. Specifically:

[0093] S610, using the final human posture recognition model to perform human posture estimation and recognition on the key action images of the standard person and the tested person, and obtain joint point annotation information of each image;

[0094] S620, using the acquired joint point annotation information, calculate the standard motion posture information matrix Each key joint angle θ in

[0095] In this embodiment, there are 8 key joint angles in total, and their positions are as follows: Figure 6 As shown, ∠a is right shoulder, left shoulder and left elbow, ∠b is right shoulder, left shoulder and left elbow, ∠c is left shoulder, left elbow and left wrist, ∠d is right shoulder, left elbow and right wrist, ∠e is left shoulder, left hip and left knee, ∠f is right shoulder, right hip and right knee, ∠g is left hip, left knee and left ankle, ∠h is right hip, right knee and right ankle.

[0096] For example, to calculate ∠a, let θa be the angle between the joints of bones 5, 6, and 7, whose coordinates are (x1, y1), (x2, y2), and (x3, y3), first calculate the distance between the joints of bones 5 and 6 (normalized):

[0097]

[0098] Similarly, we get l K,S and l J,S , and then calculate the joint angle θa:

[0099]

[0100] S630, according to the key joint angle θ and the standard motion posture information matrix Construct the standard feature matrix P, expressed as:

[0101] P=[p 1 ,p 2 ,p 3 ,…,p i ,…,p n ]

[0102] Where n is the number of key action images, p i is the key angle feature set of the i-th frame of the sequence, expressed as:

[0103] p i =[θ X ,θ Y ,θ Z ,θ [ ,θ \ ,θ ] ,θ ^ ]

[0104] S640, using the same method as steps S630 to S640, the motion posture information matrix of the person under test is transformed, and the measured feature matrix T is expressed as:

[0105] T=[t 1 ,t 2 ,t 3 ,…,t i ,…,t n ]

[0106] S700: According to the standard feature matrix and the tested feature matrix, a key joint angle similarity comparison method is used to perform motion posture matching and complete motion posture evaluation. Specifically:

[0107] According to the standard feature matrix P and the tested feature matrix T, posture matching is performed on each action of the tested person. The calculation formula of posture matching is expressed as:

[0108]

[0109] Among them, ||·|| means finding the Euclidean norm of a vector;

[0110] Analyze each key action, assuming that the i-th joint angle of the tested person is b i , the corresponding standard action is obtained through posture matching as p pose , the corresponding angle of the i-th joint is p pose,i , get the deviation value calculation formula S i , expressed as:

[0111]

[0112] Calculate the corresponding joint score g i , expressed as:

[0113]

[0114] Finally, the average score is calculated based on the scores of each joint as the total score G:

[0115]

[0116] Complete the assessment of movement posture.

[0117] Based on the above method, it is possible to quantify and display the current action of the person being tested, thereby completing the evaluation of an action based on the score of each joint angle and the overall score, accurately reflecting the standardization of an action. This not only helps to make standard judgments on the posture of the action, but also can correct the action response based on the results obtained, thereby largely avoiding sports injuries caused by incorrect sports knowledge.

[0118] Embodiment 3: A motion posture assessment system, comprising:

[0119] The data collection and processing module is used to collect the key movements that need to be evaluated when the standard personnel use the standard posture to exercise; and to construct the standard movement posture information matrix based on the basic information of the standard personnel and the key information extracted from the key movements of the standard personnel;

[0120] At the same time, the key movements that need to be evaluated when the subjects are exercising are collected; and the movement posture information matrix of the subjects is constructed based on the basic information of the subjects and the key information extracted from the key movements of the subjects;

[0121] A model building and posture recognition module is used to build a final human posture recognition model using the method of claim 1; and to perform posture recognition on the key movements of the standard person and the key movement images of the tested person, respectively, to obtain key joint annotation information; and to combine the obtained annotation information with the standard motion posture information matrix and the tested person's motion posture information matrix to obtain a standard feature matrix and a tested feature matrix;

[0122] The posture matching and output module is used to match the motion posture according to the standard feature matrix and the tested feature matrix, adopt the key joint angle similarity comparison method, complete the motion posture evaluation, and output the evaluation results.

[0123] The method of using the system of this embodiment to perform motion posture recognition is not described in detail, and is consistent with that of Embodiment 2.

[0124] Embodiment 4: This embodiment provides a non-transitory computer-readable storage medium on which computer instructions are stored. The computer instructions enable a computer to execute the motion posture recognition method described in Embodiment 2. The specific steps of the method are not repeated here.

[0125] Embodiment 5: This embodiment provides an electronic device, which may include: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus. The processor may call the logic instructions in the memory to execute the motion gesture recognition method described in Embodiment 2, and the specific steps of the method are not repeated here.

[0126] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0127] Embodiment 6: This embodiment provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the motion posture recognition method described in Embodiment 2. The specific steps of the method are not repeated here.

[0128] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0129] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a human posture recognition model, characterized in that: include: Obtain public human posture datasets and perform preprocessing; Based on the YOLOv8-pose network framework, the improved spatial pyramid pooling module SPPF and deformable convolution DCNv4 are introduced, and the target box and key point prediction are combined to realize the detection and posture estimation of the human body in the image, and obtain an improved human posture recognition network framework; The improved human posture recognition network framework is repeatedly trained according to the human posture dataset until the recognition rate reaches more than 90%, and the final human posture recognition model is obtained.

2. The method for constructing a human posture recognition model according to claim 1, characterized in that: The human body posture dataset is divided into a training set, a validation set and a test set; Train the human posture recognition network framework with the training set and adjust the network weights; Use the validation set to adjust the hyperparameters of the human posture recognition model; The recognition rate of the human posture recognition model is evaluated using the test set.

3. The method for constructing a human posture recognition model according to claim 1, characterized in that: The improved human posture recognition network framework introduces the SP_DCNv4 module combining SPPF and DCNv4 in the Backbone layer as a convolutional spatial pyramid pooling layer. After receiving the feature map from the previous layer as input, this layer performs maximum pooling on the feature map with three pooling kernels of the same size, and then DCNv4 concatenates the outputs of different pooling layers to form a feature vector of a fixed length as the input of subsequent fully connected layers or other network layers.

4. A method for evaluating a motion posture, characterized in that: include: Collect the key movements that need to be evaluated when standard personnel use standard postures to perform sports; And based on the basic information of standard personnel and the key information extracted from the key actions of standard personnel, a standard motion posture information matrix is ​​constructed; Collect the key movements that need to be evaluated when the subjects are exercising; and construct the subject's movement posture information matrix based on the subject's basic information and the key information extracted from the subject's key movements; The final human posture recognition model constructed by the method of claim 1 performs posture recognition on the key movements of the standard person and the key movement images of the tested person respectively to obtain key joint annotation information; The obtained annotation information is combined with the standard motion posture information matrix and the motion posture information matrix of the tested person to obtain the standard feature matrix and the tested feature matrix; According to the standard feature matrix and the tested feature matrix, the key joint angle similarity comparison method is used to perform motion posture matching and complete the motion posture evaluation.

5. The motion posture evaluation method according to claim 4, characterized in that: The method collects key movements that need to be evaluated when standard personnel use standard postures to perform sports; and constructs a standard sports posture information matrix based on basic information of the standard personnel and key information extracted from the key movements of the standard personnel; including: Obtain the height data of standard personnel and record it as Hs; When a standard person performs a sport in a standard posture, a camera is used to capture images of the key movements that need to be evaluated, and a real number set is obtained. Extract the number of key actions that need to be evaluated in the movement, denoted as P, the number of joints, denoted as G, and the joint information, denoted as S; Constructing standard motion posture information matrix The method collects key movements of the tested person that need to be evaluated when the tested person is exercising, and constructs a movement posture information matrix of the tested person based on the basic information of the tested person and the key information extracted from the key movements of the tested person; including: Obtain the height data of the person being tested and record it as Hc; When the subject is exercising, use a camera to capture images of the key movements that need to be evaluated, and obtain another real number set Extract the number of joints that need to be evaluated in the movement, denoted as G, and the joint information, denoted as S; Construct the motion posture information matrix of the tested person 6. The method for evaluating the motion posture according to claim 5, characterized in that: The final human posture recognition model constructed by the method of claim 1 performs posture recognition on the key actions of the standard person and the key action images of the tested person to obtain key joint annotation information; The obtained annotation information is combined with the standard motion posture information matrix and the motion posture information matrix of the tested person to obtain the standard feature matrix and the tested feature matrix; including: Use the final human posture recognition model to estimate and recognize the human posture of the key action images of the standard person and the person under test, and obtain the joint point annotation information of each image; Using the obtained joint point annotation information, calculate the standard motion posture information matrix Each key joint angle θ in; According to the key joint angle θ and the standard motion posture information matrix Construct the standard feature matrix P, expressed as: P=[p1,p2,p3,…,p i ,…,p n ] Where n is the number of key action images, p i is the key angle feature set of the i-th frame of the sequence, expressed as: p i =[θ1,θ2,θ3,…θ j ] θ j is the angle of the jth key joint; The obtained joint point annotation information is used to calculate the motion posture information matrix of the tested person Each key joint angle θ in the test; according to the key joint angle θ and the motion posture information matrix of the tested person Construct the tested feature matrix T, expressed as: T=[t1,t2,t3,…,t i ,…,t n ] Where n is the number of key action images, t i is the key angle feature set of the i-th frame of the sequence, expressed as: t i =[θ1,θ2,θ3,…θ j ] θ j is the angle of the jth key joint.

7. The method for evaluating the motion posture according to claim 6, characterized in that: The method of performing motion posture matching and completing motion posture evaluation based on the standard feature matrix and the tested feature matrix by using a key joint angle similarity comparison method comprises: According to the standard feature matrix P and the tested feature matrix T, posture matching is performed on each action of the tested person. The calculation formula of posture matching is expressed as: Among them, ||·|| means finding the Euclidean norm of a vector; Analyze each key action, assuming that the i-th joint angle of the tested person is b i , the corresponding standard action is obtained through posture matching as p pose , the corresponding angle of the i-th joint is p pose,i , get the deviation value calculation formula s i , expressed as: Calculate the corresponding joint score g i , expressed as: Finally, the average score is calculated based on the scores of each joint as the total score G: Where n is the total number of key joints; Complete the assessment of movement posture.

8. A sports posture assessment system, characterized in that: include: The data collection and processing module is used to collect key movements that need to be evaluated when standard personnel use standard postures to perform movements; And based on the basic information of standard personnel and the key information extracted from the key actions of standard personnel, a standard motion posture information matrix is ​​constructed; At the same time, the key movements that need to be evaluated when the subjects are exercising are collected; and the movement posture information matrix of the subjects is constructed based on the basic information of the subjects and the key information extracted from the key movements of the subjects; A model building and posture recognition module, used to build a final human posture recognition model using the method of claim 1; and to perform posture recognition on the key movements of the standard person and the key movement images of the tested person, respectively, to obtain key joint annotation information; The obtained annotation information is combined with the standard motion posture information matrix and the motion posture information matrix of the tested person to obtain the standard feature matrix and the tested feature matrix; The posture matching and output module is used to match the motion posture according to the standard feature matrix and the tested feature matrix, adopt the key joint angle similarity comparison method, complete the motion posture evaluation, and output the evaluation results.

9. A non-transitory computer-readable storage medium having computer instructions stored thereon, wherein the computer instructions enable a computer to execute the motion posture assessment method described in any one of claims 4 to 7.

10. An electronic device comprising: A processor, a communication interface, a memory and a communication bus, wherein the processor, Communication interface, the memory communicates with each other through the communication bus, and the processor calls the logic instructions in the memory. To execute the motion posture assessment method described in any one of claims 4-7.

Images