Automatic posture positioning and quantitative analysis method and system, computer device and medium

Abstract

The application discloses a kind of posture automatic positioning and quantitative analysis method, system, computer equipment and medium, the method includes: obtaining single model performance posture data set;Depth neural network framework including feature extraction network, automatic positioning subnetwork and quantitative analysis subnetwork is constructed, the feature extraction network is connected with automatic positioning subnetwork, quantitative analysis subnetwork respectively;Depth neural network framework is trained using the data set;Obtain the single model image to be analyzed;The single model image to be analyzed is input into trained depth neural network framework, realizes the automatic positioning and quantitative analysis of model posture.The application constructs a depth neural network framework including feature extraction network, automatic positioning subnetwork and quantitative analysis subnetwork, and performer can learn and study the posture law in excellent model work according to the quantitative data output by depth neural network framework, to restore high-level posture with aesthetic sense.

Description

Technical Field

[0001] This invention relates to a method, system, computer equipment, and medium for automatic posture positioning and quantitative analysis, belonging to the field of model performance posture estimation technology. Background Technology

[0002] For outstanding models in fashion shows or advertising performances, the works of high-level outstanding models are important materials for performers to learn and study trends and aesthetics. Most of these works are pictures of models in poses, and these pose pictures are usually analyzed by taking pose estimations so that performers can learn from them.

[0003] Pose estimation aims to extract human pose from given sensor output data and has been widely applied in pose tracking, human behavior recognition, human-computer interaction, and human image generation. In recent years, with the excellent performance of deep learning in image classification, object detection, and semantic segmentation, deep learning-based human pose estimation has become a research hotspot in computer vision, and many deep learning-based pose estimation methods have been proposed and achieved good results. However, ordinary pose estimation methods can only determine the relative position of a certain part of the body in the pose; positional information alone cannot adequately interpret the poses of models performing.

[0004] Currently, most methods in this field use deep learning or image processing to estimate the relative positions of skeletal points in a single image of a human figure. The drawbacks of these methods are poor robustness to the complex and subtle changes in poses during model performances; secondly, skeletal points alone cannot capture the model's form and muscle lines during a performance; and thirdly, these methods generally only determine the relative position of a specific part of the pose, and positional information alone cannot adequately interpret the model's posture. Summary of the Invention

[0005] In view of this, the present invention provides a method, system, computer device and storage medium for automatic posture localization and quantitative analysis. It constructs a deep neural network framework including a feature extraction network, an automatic localization subnetwork and a quantitative analysis subnetwork. Performers can learn and study the posture patterns in excellent model works based on the quantitative data output by the deep neural network framework, thereby reproducing a high-level posture with aesthetic appeal.

[0006] The first objective of this invention is to provide an automatic attitude positioning and quantitative analysis method.

[0007] The second objective of this invention is to provide an automatic attitude positioning and quantitative analysis system.

[0008] A third objective of this invention is to provide a computer device.

[0009] A fourth objective of this invention is to provide a storage medium.

[0010] The first objective of this invention can be achieved by adopting the following technical solution:

[0011] An automatic attitude localization and quantitative analysis method, the method comprising:

[0012] Obtain a dataset of poses for a single model performance.

[0013] A deep neural network framework is constructed, comprising a feature extraction network, an automatic localization subnetwork, and a quantitative analysis subnetwork, wherein the feature extraction network is connected to the automatic localization subnetwork and the quantitative analysis subnetwork, respectively.

[0014] A deep neural network framework was trained using the dataset.

[0015] Obtain the image of the single model to be analyzed;

[0016] The image of the single model to be analyzed is input into a trained deep neural network framework to achieve automatic localization and quantitative analysis of the model's posture.

[0017] Furthermore, the feature extraction network includes a pre-convolutional module and multiple combination modules;

[0018] In this module, the output of the previous combination module is connected to the input of the next combination module, the input of the first combination module is connected to the output of the pre-convolution module, and the last combination module is connected to the automatic localization subnetwork and the quantitative analysis subnetwork, respectively.

[0019] Furthermore, the combined module includes a residual module, a V module, and a 1×1 convolutional layer connected in sequence.

[0020] Furthermore, the V module is used to perform multiple downsampling and multiple upsampling on the feature matrix of the input V module;

[0021] The number of downsampling operations is the same as the number of upsampling operations.

[0022] During each downsampling operation, the corresponding feature matrix is ​​downsampled according to the max pooling operation, reducing its resolution by half.

[0023] During each upsampling, the corresponding feature matrix is ​​upsampled according to the deconvolution method, increasing its resolution by half.

[0024] Furthermore, before each downsampling and after the last downsampling, the corresponding feature matrix is ​​copied through the first residual processing;

[0025] The copied feature matrix is ​​subjected to a second residual processing to obtain the corresponding residual;

[0026] The residual is added to the feature matrix of the same resolution in subsequent upsampling processes.

[0027] Furthermore, the automatic positioning sub-network is specifically used for:

[0028] The feature matrix of the input autolocalization subnetwork is copied into a first feature matrix, a second feature matrix, and a third feature matrix;

[0029] The first and second feature matrices are convolved to make their resolution consistent with the number of channels;

[0030] Based on the channel attention mechanism and the spatial attention mechanism, the first feature matrix and the second feature matrix after convolution are processed to generate attention vector and attention map respectively;

[0031] The attention vector, attention map, and third feature matrix are added together and then restored using the third residual processing.

[0032] Furthermore, the single-person model performance posture dataset includes multiple labeled single-person model static posture images;

[0033] Each of the labeled static pose images of a single model is labeled with at least twenty dynamic points and their weights. The at least twenty dynamic points are located at the chin, right neck, left neck, right shoulder, left shoulder, right elbow, left elbow, right palm, left palm, right waist, left waist, abdomen, right hip, left hip, right knee, left knee, right ankle, left ankle, right toe, and left toe.

[0034] The second objective of this invention can be achieved by adopting the following technical solution:

[0035] An automatic attitude positioning and quantitative analysis system, the system comprising:

[0036] The first acquisition unit is used to acquire a dataset of individual model performance poses.

[0037] A construction unit is used to construct a deep neural network framework including a feature extraction network, an automatic localization subnetwork, and a quantitative analysis subnetwork, wherein the feature extraction network is connected to the automatic localization subnetwork and the quantitative analysis subnetwork, respectively.

[0038] Training unit, used to train a deep neural network framework using the dataset;

[0039] The second acquisition unit is used to acquire images of the single model to be analyzed.

[0040] The automatic positioning and quantitative analysis unit is used to input the single model image to be analyzed into a trained deep neural network framework to achieve automatic positioning and quantitative analysis of the model's posture.

[0041] The third objective of this invention can be achieved by adopting the following technical solution:

[0042] A computer device includes a processor and a memory for storing processor-executable programs, wherein when the processor executes the programs stored in the memory, it implements the above-described automatic attitude positioning and quantitative analysis method.

[0043] The fourth objective of this invention can be achieved by adopting the following technical solution:

[0044] A storage medium storing a program, which, when executed by a processor, implements the above-described automatic attitude positioning and quantitative analysis method.

[0045] The present invention has the following advantages over the prior art:

[0046] 1. This invention uses computer-based digital methods to analyze the performance works of outstanding models, automatically and intuitively demonstrating the principles presented by the models' performance postures, providing quantitative data support for performers, thereby helping them with daily training;

[0047] 2. This invention constructs a feasible and highly usable dataset of single-person model performance postures for dynamic points. The dataset includes more than 12,000 images of static performance works by single models, along with the positions and weights of 20 dynamic points, which can meet the training requirements of deep neural network frameworks.

[0048] 3. This invention uses an automatic positioning subnetwork to automatically locate the dynamic points on the model's posture, and uses a quantitative analysis subnetwork to assign weights to the dynamic points on the model's posture, thereby achieving automatic quantitative analysis of the model's posture while maintaining the accurate position of the dynamic points. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0050] Figure 1 This is a flowchart of the attitude automatic positioning and quantitative analysis method of Embodiment 1 of the present invention.

[0051] Figure 2This is an interface diagram of the annotation software according to Embodiment 1 of the present invention.

[0052] Figure 3 This is a structural diagram of the deep neural network framework of Embodiment 1 of the present invention.

[0053] Figure 4 This is a structural diagram of the V module in Embodiment 1 of the present invention.

[0054] Figure 5 This is a structural diagram of the automatic positioning subnetwork of Embodiment 1 of the present invention.

[0055] Figure 6 This is a structural diagram of the residual module in Embodiment 1 of the present invention.

[0056] Figure 7 This is a visualization of the automatic positioning and quantitative analysis results of the model's posture in Embodiment 1 of the present invention.

[0057] Figure 8 This is a structural block diagram of the attitude automatic positioning and quantitative analysis system of Embodiment 2 of the present invention.

[0058] Figure 9 This is a structural block diagram of the computer device according to Embodiment 3 of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0060] Example 1:

[0061] like Figure 1 As shown in the figure, this embodiment provides an automatic attitude localization and quantitative analysis method, which includes the following steps:

[0062] S101. Obtain a dataset of individual model performance poses.

[0063] Before step S101, the specific process of obtaining the single model performance pose dataset is as follows:

[0064] S11. In the solo model performance posture, obtain twenty dynamic points; where: the twenty dynamic points are located at the chin, right neck, left neck, right shoulder, left shoulder, right elbow, left elbow, right palm, left palm, right waist, left waist, abdomen, right hip, left hip, right knee, left knee, right ankle, left ankle, right toe, and left toe.

[0065] Unlike the human skeleton key points used in ordinary pose estimation methods, this embodiment deconstructs eleven key parts that can interpret the pose from the poses of solo models in daily training, namely chin, neck, shoulder, elbow, palm, waist, abdomen, hip, knee, ankle and toes; among them, except for the chin and abdomen which have one corresponding point, the other parts have two corresponding points on the left and right, for a total of twenty corresponding points; the corresponding points are called dynamic points.

[0066] S12. Assign a relative weight to each dynamic point.

[0067] It is worth noting that while performers can quickly determine how to pose a solo model based on these dynamic points, knowing the relative positions of these dynamic points on the image is insufficient to analyze the principles behind posing. Therefore, this embodiment studies the detailed conditions of each key part of a solo model's pose, i.e., the detailed conditions of each body part, and assigns a relative weight to each dynamic point. The weight ranges from 1 to 10, representing the performer's level of attention to the dynamic point and the amount of force applied to it, or the pressure that the key part needs to withstand. 1 is the minimum weight value, indicating the lowest level of attention and force applied to the dynamic point, while 10 is the maximum weight value, indicating the highest level of attention and force applied to the dynamic point.

[0068] S13. Based on the twenty dynamic points and weight assignment method, the static pose image set of a single model is labeled and converted into heat map form, thus obtaining the single model performance pose dataset.

[0069] Step S13 specifically includes:

[0070] (1) Multiple single model pose images were obtained from fashion magazine websites, fashion advertising websites and model photo websites by means of manual collection and machine crawling, thus obtaining a set of single model pose images.

[0071] (2) The single model pose image set is preprocessed to obtain the single model static pose image set; wherein: the preprocessing includes: software removal of duplicate images, deletion of non-static pose images and removal of non-professional model images; the single model static pose image set contains 12284 single model static pose images, covering various static poses posed by professional models such as standing, sitting, holding on and leaning.

[0072] (3) Based on the twenty dynamic points and weight assignment method, each image in the single-person model static pose image set is labeled and converted into a heat map, so as to obtain the labeled single-person model static pose image set and heat map set respectively.

[0073] (4) The labeled set of static pose images and heat maps of individual models are used as a dataset of performance poses of individual models.

[0074] It is worth noting that, due to the large amount of annotation work in step S13, the annotation process requires collaboration among multiple people. Therefore, this embodiment utilizes a web-based or server-based annotation system to facilitate remote collaborative annotation by relevant personnel. Figure 2 As shown.

[0075] After executing steps S11 to S13 in this embodiment, a dataset of individual model performance postures is obtained.

[0076] S102. Construct a deep neural network framework that includes a feature extraction network, an automatic localization subnetwork, and a quantitative analysis subnetwork.

[0077] like Figure 3 As shown, this embodiment constructs a deep neural network framework including a feature extraction network, an automatic localization sub-network combined with an attention mechanism module, and a quantitative analysis sub-network (Sub Net); wherein: the feature extraction network is connected to the automatic localization sub-network and the quantitative analysis sub-network, respectively.

[0078] Furthermore, the feature extraction network includes a pre-convolutional module and four combination modules; wherein, the output of the previous combination module is connected to the input of the next combination module, the input of the first combination module is connected to the output of the pre-convolutional module, and the last combination module is connected to the automatic localization subnetwork and the quantitative analysis subnetwork, respectively.

[0079] In this embodiment, the pre-convolution module is a front-end module of the deep neural network framework, used to reduce the resolution of the image input to the pre-convolution module to the resolution of the heatmap that needs to be predicted.

[0080] Furthermore, the combined module includes a residual module, a V module, and a 1×1 convolutional layer connected in sequence.

[0081] Furthermore, such as Figure 4 As shown, the V module is used to perform four downsampling and four upsampling operations on the feature matrix of the input V module. During each downsampling, the corresponding feature matrix is ​​downsampled according to the max pooling operation to reduce its resolution by half. During each upsampling, the corresponding feature matrix is ​​upsampled according to the deconvolution method to increase its resolution by half.

[0082] Furthermore, before each downsampling and after the last downsampling, the corresponding feature matrix is ​​copied through the first residual processing; the copied feature matrix is ​​then subjected to the second residual processing to obtain the corresponding residual; and the residual is added to the feature matrix of the same resolution in subsequent upsampling processes.

[0083] It is worth noting that in this embodiment, a simple deconvolution upsampling module is designed to improve the upsampling effect. After four upsampling operations, the resolution of the feature matrix returns to the resolution of the heatmap that needs to be predicted. In the V module of the fourth combined module, the feature matrix extracted after four downsampling operations is input into the quantitative analysis sub-network for prediction to predict the dynamic value, i.e. the weight of the dynamic point. The automatic positioning sub-network is connected to the 1×1 convolutional layer of the fourth combined module.

[0084] Specifically, such as Figure 5 As shown, the automatic localization subnetwork that incorporates the attention mechanism module includes a channel attention mechanism and a spatial attention mechanism. The module following the channel attention mechanism represents the attention vector generated by the channel attention mechanism, and the module following the spatial attention mechanism represents the attention map generated by the spatial attention mechanism.

[0085] Furthermore, the automatic location of sub-networks is specifically used for:

[0086] S21. Copy the feature matrix of the input autolocalization subnetwork into a first feature matrix, a second feature matrix, and a third feature matrix.

[0087] S22. Perform convolution on the first and second feature matrices to make their resolution consistent with the number of channels.

[0088] In step S22, the convolution process is performed using a 3×3 convolutional layer. This convolution process is used to increase the number of channels in the feature matrix.

[0089] S23. Based on the channel attention mechanism and the spatial attention mechanism, process the first feature matrix and the second feature matrix after convolution to generate attention vector and attention map respectively.

[0090] S24. Add the attention vector, attention map and third feature matrix together, and restore them by processing the third residual.

[0091] After executing steps S21 to S24, the feature matrix restored by the automatic positioning subnetwork is added to the feature matrix output by the quantitative analysis subnetwork, which is composed of three residual modules and an upsampling module, and then the predicted heatmap is output.

[0092] In this embodiment, the first residual processing, the second residual processing, and the third residual processing are all completed through the corresponding residual modules.

[0093] Specifically, the structure of all residual modules is as follows: Figure 6 As shown.

[0094] S103. Train a deep neural network framework using the dataset.

[0095] During the training of the deep neural network framework using the aforementioned dataset, the overall loss function of the deep neural network framework is as follows:

[0096]

[0097] Among them, L p L represents the loss function of the dynamic point. w The loss function representing the dynamic value, The parameters represent the balance between the two loss functions.

[0098] The L p As shown in the following formula:

[0099]

[0100] Where n represents the number of V modules, α represents the weight value, and L k This represents the heatmap loss output by each V module.

[0101] It is worth noting that, as shown in Equation (a), in the quantitative analysis subnetwork, in order to balance the loss value of the dynamic value as much as possible, the dynamic value is also heatmapped; as shown in Equation (b), after the end of each V module, a dynamic point heatmap is generated by a convolutional layer with a kernel size of 1×1, and the weights and participation in supervision are adjusted according to the depth of the module in the entire network, and the mean square error is calculated as the loss function. Equation (a) also calculates the mean square error as the loss function.

[0102] S104. Obtain the image of the single model to be analyzed.

[0103] S105. Input the single model image to be analyzed into the trained deep neural network framework to achieve automatic localization and quantitative analysis of the model's posture.

[0104] like Figure 7 As shown, this paper presents the automatic localization and quantitative analysis results of the model's pose obtained after inputting a trained deep neural network framework with an image of a single model to be analyzed.

[0105] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware, and the corresponding program can be stored in a computer-readable storage medium.

[0106] It should be noted that although the method operations of the above embodiments are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the order of execution of the described steps may be changed. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0107] Example 2:

[0108] like Figure 8 As shown, this embodiment provides an automatic attitude localization and quantitative analysis system. The system includes a first acquisition unit 801, a construction unit 802, a training unit 803, a second acquisition unit 804, and an automatic localization and quantitative analysis unit 805. The specific functions of each unit are as follows:

[0109] The first acquisition unit 801 is used to acquire a dataset of individual model performance poses.

[0110] The construction unit 802 is used to construct a deep neural network framework including a feature extraction network, an automatic localization subnetwork, and a quantitative analysis subnetwork, wherein the feature extraction network is connected to the automatic localization subnetwork and the quantitative analysis subnetwork, respectively.

[0111] Training unit 803 is used to train a deep neural network framework using the dataset;

[0112] The second acquisition unit 804 is used to acquire images of the single model to be analyzed.

[0113] The automatic positioning and quantitative analysis unit 805 is used to input the single model image to be analyzed into a trained deep neural network framework to achieve automatic positioning and quantitative analysis of the model's posture.

[0114] Example 3:

[0115] like Figure 9 As shown, this embodiment provides a computer device, which includes a processor 902, a memory, an input device 903, a display device 904, and a network interface 905 connected via a system bus 901. The processor 902 provides computing and control capabilities. The memory includes a non-volatile storage medium 906 and internal memory 907. The non-volatile storage medium 906 stores an operating system, computer programs, and a database. The internal memory 907 provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium 906. When the computer program is executed by the processor 902, it implements the automatic attitude positioning and quantitative analysis method of Embodiment 1 described above, as follows:

[0116] Obtain a dataset of poses for a single model performance.

[0117] A deep neural network framework is constructed, comprising a feature extraction network, an automatic localization subnetwork, and a quantitative analysis subnetwork, wherein the feature extraction network is connected to the automatic localization subnetwork and the quantitative analysis subnetwork, respectively.

[0118] A deep neural network framework was trained using the dataset.

[0119] Obtain the image of the single model to be analyzed;

[0120] The image of the single model to be analyzed is input into a trained deep neural network framework to achieve automatic localization and quantitative analysis of the model's posture.

[0121] Example 4:

[0122] This embodiment provides a storage medium, which is a computer-readable storage medium, storing a computer program. When the computer program is executed by a processor, it implements the automatic attitude localization and quantitative analysis method of Embodiment 1 above, as follows:

[0123] Obtain a dataset of poses for a single model performance.

[0124] A deep neural network framework is constructed, comprising a feature extraction network, an automatic localization subnetwork, and a quantitative analysis subnetwork, wherein the feature extraction network is connected to the automatic localization subnetwork and the quantitative analysis subnetwork, respectively.

[0125] A deep neural network framework was trained using the dataset.

[0126] Obtain the image of the single model to be analyzed;

[0127] The image of the single model to be analyzed is input into a trained deep neural network framework to achieve automatic localization and quantitative analysis of the model's posture.

[0128] It should be noted that the computer-readable storage medium in this embodiment can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0129] In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this embodiment, the computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable program. This propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable storage medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable storage medium can be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0130] The computer-readable storage medium described above can be used to write computer programs for executing this embodiment in one or more programming languages ​​or combinations thereof. These programming languages ​​include object-oriented programming languages—such as Java, Python, and C++—and conventional procedural programming languages—such as C or similar programming languages. The program can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0131] In summary, this invention constructs a deep neural network framework that includes a feature extraction network, an automatic localization subnetwork, and a quantitative analysis subnetwork. Performers can learn and study the posture patterns in excellent model works based on the quantitative data output by the deep neural network framework, thereby reproducing aesthetically pleasing and high-level postures.

[0132] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A method for automatic attitude positioning and quantitative analysis, characterized in that, The method includes: Obtain a dataset of poses for a single model performance, specifically including: Twenty dynamic points were obtained from the single model performance posture dataset; Among them, the twenty dynamic points are located at the chin, right neck, left neck, right shoulder, left shoulder, right elbow, left elbow, right palm, left palm, right waist, left waist, abdomen, right hip, left hip, right knee, left knee, right ankle, left ankle, right toe, and left toe; Assign a relative weight to each dynamic point; Based on twenty dynamic points and weight assignment methods, the static pose image set of a single model is labeled and converted into heat map form, thus obtaining the performance pose dataset of a single model. Multiple individual model pose images are acquired to obtain an individual model pose image set; Preprocess the single-model pose image set to obtain a single-model static pose image set; Based on the twenty dynamic points and the weight assignment method, each image in the single-person model static pose image set is labeled and converted into a heat map, thus obtaining the labeled single-person model static pose image set and heat map set respectively. The labeled set of static pose images and heatmaps of individual models are used as a dataset of performance poses of individual models. A deep neural network framework is constructed, comprising a feature extraction network, an automatic localization subnetwork, and a quantitative analysis subnetwork, wherein the feature extraction network is connected to the automatic localization subnetwork and the quantitative analysis subnetwork, respectively. The feature extraction network includes a pre-convolutional module and multiple combination modules; The pre-convolution module is a front-end module of the deep neural network framework, used to reduce the resolution of the image input to the pre-convolution module to the resolution of the heatmap to be predicted; In this module, the output of the previous combination module is connected to the input of the next combination module, the input of the first combination module is connected to the output of the pre-convolution module, and the last combination module is connected to the automatic localization subnetwork and the quantitative analysis subnetwork, respectively. The combined module includes a residual module, a V module, and a 1×1 convolutional layer connected in sequence. The V module is used to perform multiple downsampling and multiple upsampling on the feature matrix of the input V module; The automatic positioning subnetwork is specifically used for: The feature matrix of the input autolocalization subnetwork is copied into a first feature matrix, a second feature matrix, and a third feature matrix; The first and second feature matrices are convolved to make their resolution consistent with the number of channels; Based on the channel attention mechanism and the spatial attention mechanism, the first feature matrix and the second feature matrix after convolution are processed to generate attention vector and attention map respectively; The attention vector, attention map, and third feature matrix are added together and then restored using the third residual processing. The feature matrix restored by the automatic localization subnetwork is added to the feature matrix output by the quantitative analysis subnetwork, which is composed of three residual modules and an upsampling module, and then the predicted heatmap is output. A deep neural network framework was trained using the dataset. Obtain the image of the single model to be analyzed; The image of the single model to be analyzed is input into a trained deep neural network framework to achieve automatic localization and quantitative analysis of the model's posture.

2. The automatic attitude positioning and quantitative analysis method according to claim 1, characterized in that, The V module is used to perform multiple downsampling and multiple upsampling on the feature matrix of the input V module; The number of downsampling operations is the same as the number of upsampling operations. During each downsampling operation, the corresponding feature matrix is ​​downsampled according to the max pooling operation, reducing its resolution by half. During each upsampling, the corresponding feature matrix is ​​upsampled according to the deconvolution method, increasing its resolution by half.

3. The automatic attitude positioning and quantitative analysis method according to claim 2, characterized in that, Before each downsampling and after the last downsampling, the corresponding feature matrix is ​​copied through the first residual processing; The copied feature matrix is ​​subjected to a second residual processing to obtain the corresponding residual; The residual is added to the feature matrix of the same resolution in subsequent upsampling processes.

4. An automatic attitude positioning and quantitative analysis system, characterized in that, The system for implementing the automatic attitude localization and quantitative analysis method according to any one of claims 1-3, the system comprising: The first acquisition unit is used to acquire a dataset of individual model performance poses. A construction unit is used to construct a deep neural network framework including a feature extraction network, an automatic localization subnetwork, and a quantitative analysis subnetwork, wherein the feature extraction network is connected to the automatic localization subnetwork and the quantitative analysis subnetwork, respectively. Training unit, used to train a deep neural network framework using the dataset; The second acquisition unit is used to acquire images of the single model to be analyzed. The automatic positioning and quantitative analysis unit is used to input the single model image to be analyzed into a trained deep neural network framework to achieve automatic positioning and quantitative analysis of the model's posture.

5. A computer device, comprising a processor and a memory for storing a processor-executable program, characterized in that, When the processor executes the program stored in the memory, it implements the attitude automatic positioning and quantitative analysis method according to any one of claims 1-3.

6. A storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the automatic attitude positioning and quantitative analysis method according to any one of claims 1-3.

Images