A human body finite element model generation method

Abstract

This invention provides a method for generating a human finite element model, comprising the following steps: Step 1, establishing a target human finite element model feature acquisition module, firstly, obtaining 3D point cloud data of the human body under different vital signs and postures from a 2D human image through a deep learning network model for 3D human reconstruction, and establishing a statistical model of human features; Step 2, establishing a basic finite element model deformation module, decomposing the basic human finite element model to obtain the human skeleton and body surface finite element models; Step 3, establishing a target human finite element model integration module, firstly, querying the corresponding SMPL model based on the transformed human finite element model, calculating the joints of the SMPL model, and finally integrating the soft tissue finite element model and muscle units to complete the generation of the target human finite element model. This method achieves the localization of the corresponding skeleton, solving the problem of manually adjusting the skeleton position in previous human finite element models.

Description

Technical fields:

[0001] This invention relates to the interdisciplinary field of automotive passive safety and computer science, specifically a method for generating a human body finite element model. Technical Background

[0002] The finite element method (FEM) model of the human body is an important tool for assessing vehicle safety. Different human body characteristics result in different damage responses under different collision conditions. Currently, there are only 5 existing finite element models of the human body. th 50 th 95 th Finite element models based on specific physiological characteristics, such as those for upright and driving postures, cannot reflect the extent of injury in the human body under multiple physiological conditions during simulation. There are three main problems with existing methods for establishing multi-physical-characteristic finite element models of the human body. First, establishing a finite element model of the human body surface requires a large amount of human body surface point cloud data, which current technologies primarily obtain through human body scanning, a very time-consuming and labor-intensive process. Second, the process of establishing a human finite element model is extremely complex, requiring both geometric accuracy and the quality of the finite element mesh to be suitable for simulation calculations. Finally, while existing mesh deformation techniques can transform existing human finite element models according to physiological requirements, they can only transform models in upright and driving postures, failing to meet the real-time and rapid generation requirements for multi-posture human body models. Summary of the Invention

[0003] In view of this, in order to solve the above problems, the present invention proposes a method for generating a finite element model of the human body.

[0004] This invention specifically relates to a method for generating a finite element model of the human body, characterized by the following steps:

[0005] Step 1: Establish a feature acquisition module for the target human body finite element model. First, obtain three-dimensional point cloud data of the human body under different physical signs and postures by using a three-dimensional human body reconstruction deep learning network model. Then, establish a statistical model of human body features between feature parameters and point cloud data. Finally, obtain the skeletal point cloud data of the corresponding physical sign parameters based on the relevant skeletal point cloud database and match it with the body surface point cloud data.

[0006] Step 2: Establish a basic finite element model deformation module. Decompose the basic human finite element model to obtain the human skeleton and body surface finite element models. Then extract the body surface and skeleton feature points of the target human body. Generate deformation control points through the established feature points to generate a deep learning network model. Finally, transform the basic body surface and skeleton finite element model into the target body surface and skeleton finite element model based on the RBF-TPS function.

[0007] Step 3: Establish the target human finite element model integration module. First, query the corresponding SMPL model based on the transformed human finite element model, calculate the joints of the SMPL model, then map the joints to the corresponding body surface finite element model, and locate the transformed skeletal finite element model according to the joints. Finally, integrate the soft tissue finite element model and muscle units to complete the generation of the target human finite element model.

[0008] Furthermore, step 1 includes the following steps:

[0009] Step 1.1: Obtain two-dimensional RGB images of the human body under different vital signs and postures, and extract the corresponding vital signs parameter matrix and posture parameter matrix. The vital signs parameters include height, weight, gender, and age, and the posture parameter labels include standing, driving, walking, cycling, and running.

[0010] Step 1.2: A deep learning network model for 3D reconstruction of the human body using a single RGB image. The network model extracts features from the human body RGB image and regresses the parameters of the SMPL model to achieve 3D reconstruction from the human body RGB image to the human body SMPL model.

[0011] The SMPL model uses a 10-dimensional body shape parameter β and a 72-dimensional human joint parameter θ to determine the corresponding 3D human body mesh model represented by triangular facets. The model expression is as follows:

[0012]

[0013] in, The calculation formula is:

[0014]

[0015] In the formula, An SMPL human body model template representing average body size. and This represents the corrected data for body shape parameters and human joint parameters on the average template. This represents the coordinates of each keypoint obtained from the surface vertices of the model, where w is the deformation weight;

[0016] Step 1.3: Extract the vertices of the SMPL human body model to obtain the target human body point cloud data, and label it as PS;

[0017] Step 1.4: Using an indirect statistical model, establish a statistical regression model by combining the vital sign parameter matrix, behavioral parameter matrix, SMPL human body shape parameter matrix, and SMPL human body posture parameter matrix. Then, map the SMPL model parameters to the corresponding human body model, merging the vital sign parameter matrix and behavioral parameter matrix into a human body feature parameter matrix, and label it as F.1*5 The regression model can be expressed as:

[0018]

[0019] In the formula, T pre This represents the predicted SMPL model parameters. M represents the parameters of the SMPL human body template representing the average body size. 5*82 This represents the regression parameter matrix.

[0020] Step 1.5: Based on the relevant human skeleton measurement database, obtain human skeleton point cloud data under different vital signs parameters. The human skeleton includes the skull, thoracic skeleton, pelvis, femur, tibia, fibula, cervical vertebrae, lumbar vertebrae, humerus, radius, ulna, scapula, clavicle, sacrum, and foot bones.

[0021] Step 1.6: Match the obtained human skeleton point cloud data with the same vital signs parameters with the body surface point cloud data to obtain the feature point cloud set of the target human finite element model.

[0022] Furthermore, step 2 includes the following steps:

[0023] Step 2.1: Obtain the THUMS human finite element model and use it as the basic finite element model. Then, decompose the model to obtain the body surface finite element model, the bone finite element model, and the soft tissue finite element model.

[0024] Step 2.2: Extract feature points from the finite element model of the body surface. The feature point selection uses the ISS (Intrinsic Shape Signatures) algorithm, and the extracted feature points of the finite element model of the body surface are marked as FEM. S_F ;

[0025] Step 2.3: Extract feature points from the finite element model of the skeleton. The feature point selection uses the ISS (Intrinsic Shape Signatures) algorithm, and the extracted feature points of the finite element model of the skeleton are marked as FEM. B_F ;

[0026] Step 2.4: Establish a deep learning network model for generating human body surface feature points, and convert the feature points of the finite element model of the body surface into FEM (Feature Element Model). S_F P and the target human body surface point cloud S Simultaneously, the model is fed in for training. If the model's accuracy is lower than the preset accuracy, the hyperparameters are adjusted and training is repeated. The loss function of the deep learning network model for generating human body surface feature points in this step is shown below:

[0027] L S (FEM S_F ,P S ,RS ) = L rec_S (FEM S_F ,P S )+λ1L perm_S (R S )+λ2L mfd_S (R S )

[0028] In the formula, R S For FEM S_F With P S The transformation matrix between them, λ1 and λ2 are the regularization coefficients of the loss function of the deep learning network model for generating human body surface feature points, L rec_S (FEM S_F ,P S The ) represents the error between the generated point cloud and the point cloud on the target human body surface, L perm_S (R S ) represents R S The error between L and the permutation matrix mfd_S (R S The nearest neighbor error is represented by ) between the point cloud on the target human body surface and the corresponding points in the transformed point cloud.

[0029] Step 2.5: Establish a deep learning network model for generating human skeleton feature points, and convert the feature points of the skeleton finite element model into FEM (Feature Element Model). B_F P with the target human skeleton point cloud B_X Simultaneously, the model is fed in for training. If the model's accuracy is lower than the preset accuracy, the hyperparameters are adjusted and training is repeated. The loss function of the deep learning network model for generating human skeletal feature points in this step is shown below:

[0030] L B (FEM B_F ,P B_X ,R B ) = L rec_B (FEM B_F ,P B_X )+λ3L perm_B (R B )+λ4L mfd_B (R B )

[0031] In the formula, R B For FEM B_F With P B_X The transformation matrix between them, λ3 and λ4 are the regularization coefficients of the loss function of the deep learning network model for generating human skeletal feature points, and L rec_B (FEM B_F ,P B_X The ) represents the error between the generated point cloud and the target human skeleton point cloud, L perm_B(R B ) represents R B The error between L and the permutation matrix mfd_B (R B The nearest neighbor error is represented by ) between the target human skeleton point cloud and the corresponding points in the transformed point cloud.

[0032] Step 2.6, transform the feature points of the finite element model of the body surface. S_F P is marked as the control point cloud of the target human body surface. C_S ;

[0033] Step 2.7, transform the feature points of the finite element model of the skeleton into FEM. B_F P is marked as the target human skeleton control point cloud. C_B ;

[0034] Step 2.8: Use the RBF-TPS function to perform FEM on the feature points of the finite element model of the body surface. S_F P is the source control point and the target human body surface control point cloud. C_S Using the target control points, a mesh transformation is performed to obtain finite element models of the human body surface under different vital signs and postures, labeled as FEM. S_T ;

[0035] The RBF-TPS function was used to analyze the finite element model (FEM) of the body surface. S Transformation is performed to generate finite element models (FEMs) of the human body surface under different vital signs and postures. S_T When, the following formula should be satisfied:

[0036]

[0037] Among them, f s (x,y,z) represents the finite element model (FEM) of the body surface. S Finite element model (FEM) of human body surface under different physical signs and postures S_T Transformation function, the function is based on the feature points of the finite element model of the body surface (FEM). S_F P and the control point cloud of the target human body surface C_S The conversion relationship between them;

[0038] Step 2.9: Use the RBF-TPS function to perform FEM on the feature points of the skeletal finite element model. B_F Source control point, target human skeleton control point cloud P C_B Using the target control points, a mesh transformation is performed to obtain finite element models of the human skeleton under different physical characteristics, labeled as FEM. B_T ;

[0039] Specifically, the RBF-TPS function is used to analyze the finite element model (FEM) of the skeleton. BTransformation is performed to generate finite element models (FEMs) of the human skeleton under different physical characteristics. B_T When, the following formula should be satisfied:

[0040]

[0041] Among them, f B (x,y,z) represents the finite element model (FEM) of the skeleton. B Finite element model of human skeleton under different physical signs B_T Transformation function, which is based on the feature points FEM of the skeletal finite element model. B_F P with the target human skeleton control point cloud C_B The conversion relationship between them;

[0042] Further, in step 3.1, the transformed human body surface finite element model (FEM) with the same vital signs parameters is... S_T With finite element model of skeleton (FEM) B_T To pair up;

[0043] Step 3.2, based on the transformed human finite element model (FEM) S_T In calculating the joint point locations, the corresponding SMPL human body model is queried based on the finite element model of the human body, and the joint points are calculated using the following formula:

[0044]

[0045] in, Here, K is the transformation matrix, K is the number of human joints, and N is the number of vertices in the SMPL human model. The vertices of the SMPL human body model are based on the shape blending process. The calculated joint positions;

[0046] Step 3.3: Locate the bones based on the joint positions. In this step, the transformed human body surface finite element model (FEM) is used. S_T The spatial position, shape, and human posture are completely identical to the corresponding SMPL human body model, therefore the joint point positions are also the same. The joint points of the SMPL human body model obtained in step 3.2 are mapped to the finite element model (FEM) of the human body surface. S_T Then, the transformed skeletal finite element model (FEM) is processed. B_T Positioning is based on the location of the joints;

[0047] Step 3.4: The internal organs of the human body, including the heart, kidneys and lungs, are obtained by scaling up the finite element models of the thoracic skeleton and pelvis surrounding the organs.

[0048] Step 3.5: The soft tissues of the trunk are produced by the combined inhibition of internal organs and body surface geometry;

[0049] Step 3.6: Add muscle linear elements. In this step, the muscle linear elements are attached to the corresponding human finite element model mesh nodes and automatically transform during the mesh transformation process to complete the generation of the target human finite element model.

[0050] The beneficial effects of this invention are as follows:

[0051] (1) This invention proposes a method for generating a human finite element model from a two-dimensional RGB image. This is an end-to-end generation method that can automatically generate a corresponding human finite element model based on any two-dimensional image containing a human body. The generated finite element model is flexible and accurate, and the entire generation process does not require manual intervention.

[0052] (2) The target human body finite element model feature acquisition module proposed in this invention can realize two different human body point cloud generation methods as needed: generating from images and generating from feature parameters. Both methods can generate target human body point clouds under different physical parameters and different postures. The basic finite element model deformation module proposed in this invention can train a deep learning model based on the point cloud at different positions of the human body to obtain the feature points at the corresponding positions, which solves the problem of manually placing feature points in the previous finite element model deformation.

[0053] (3) The target human body finite element model integration module proposed in this invention can automatically calculate the joint position based on the transformed human body surface finite element model, realize the positioning of the corresponding bones, and solve the problem that the human body finite element model previously required manual adjustment of bone position. Attached Figure Description

[0054] Figure 1 Schematic diagram of the human body finite element model generation system architecture.

[0055] Figure 2 Schematic diagram of the feature acquisition module for the target human body finite element model.

[0056] Figure 3 Schematic diagram of the deformation module of the basic finite element model.

[0057] Figure 4 Schematic diagram of deformation of finite element model of bone and soft tissue.

[0058] Figure 5 A schematic diagram of the process of generating a human finite element model based on image input.

[0059] Figure 6 A schematic diagram of the process of generating a human finite element model based on feature parameter input. Detailed Implementation

[0060] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0061] like Figure 1 As shown, this embodiment provides a human body finite element model generation system, specifically comprising three modules: a target human body finite element model feature acquisition module, a basic finite element model deformation module, and a target human body finite element model integration module; wherein:

[0062] The target human body finite element model feature acquisition module is used to acquire point clouds of the target human body surface and skeleton.

[0063] The basic finite element model deformation module is used to transform the basic human body surface finite element model and skeletal finite element model into the target human body surface model and skeletal model, respectively.

[0064] The target human finite element model integration module is used to locate the transformed target human skeleton finite element model into the target human body surface finite element model, while also integrating soft tissue finite element models and muscle elements to form a complete human finite element model.

[0065] This embodiment also provides a method for generating a human finite element model, which specifically includes the following steps:

[0066] Step 1: Establish a feature acquisition module for the target human body finite element model. First, obtain three-dimensional point cloud data of the human body under different physical signs and postures by using a three-dimensional human body reconstruction deep learning network model. Then, establish a statistical model of human body features between feature parameters and point cloud data. Finally, obtain skeletal point cloud data of corresponding physical sign parameters based on relevant skeletal point cloud databases and match them with body surface point cloud data.

[0067] This step is achieved in the following way:

[0068] Step 1.1: Obtain two-dimensional RGB images of the human body under different vital signs and postures, and extract the corresponding vital signs and postures matrix. The vital signs include height, weight, gender, age, etc., and the postures label includes standing, driving, walking, cycling, running, etc.

[0069] Step 1.2: A deep learning network model for 3D reconstruction of the human body using a single RGB image. This network model extracts features from the human body RGB image and regresses the parameters of the SMPL model to achieve 3D reconstruction from the human body RGB image to the human body SMPL model.

[0070] Among them, the SMPL model is a parametric human body model library proposed by the Max Planck Institute in Germany. This model can determine the corresponding three-dimensional human body mesh model represented by triangular patches using 10-dimensional body shape parameters β and 72-dimensional human body joint parameters θ. The model expression is as follows:

[0071]

[0072] in, The calculation formula is:

[0073]

[0074] In the formula, An SMPL human body model template representing average body size. and This represents the corrected data for body shape parameters and human joint parameters on the average template. This represents the coordinates of each keypoint obtained from the surface vertices of the model, where w is the deformation weight.

[0075] In this step, a deep learning network model for human body model reconstruction is built based on the pre-trained SMPLify model. Two-dimensional RGB images of the human body under different vital signs and poses obtained in step 1 are fed into the model. This model acquires the SMPL human body model with different vital signs and pose parameters, along with the corresponding body shape parameter β and joint point parameter θ. The main method of the SMPLify model is to calculate the minimum value of the objective function from the image. The objective function is expressed as:

[0076] E(β,θ)=E J (β,θ;K,J est )+λ θ E θ (θ)+λ a E a (θ)+λ sp E sp (θ;β)+λ β E β (β) (3)

[0077] In the formula, K represents the camera parameters of the human body image, and J est λ represents the two-dimensional coordinates of each key point. θ , λ a , λ sp , λ β This represents the weighting coefficient.

[0078] Step 1.3: Extract the vertices of the SMPL human body model to obtain the target human body point cloud data, and label it as PS;

[0079] Step 1.4: Using an indirect statistical model, establish a statistical regression model by combining the vital sign parameter matrix, behavioral parameter matrix, SMPL human body shape parameter matrix, and SMPL human body posture parameter matrix. Then, map the SMPL model parameters to the corresponding human body model, merging the vital sign parameter matrix and behavioral parameter matrix into a human body feature parameter matrix, and label it as F. 1*5 The regression model can be expressed as:

[0080]

[0081] In the formula, T pre This represents the predicted SMPL model parameters. M represents the parameters of the SMPL human body template representing the average body size. 5*82 This represents the regression parameter matrix.

[0082] Traditionally, statistical regression models between human feature parameters and human point clouds are established by mapping three-dimensional point coordinates on the human body surface using several human feature parameters. After these point coordinates are unfolded into one dimension, the matrix length often reaches hundreds of thousands, which can cause large regression errors. This step, based on the above formula, can generate corresponding SMPL model parameters from any combination of human feature parameters, and then generate the corresponding three-dimensional point cloud of the human body surface.

[0083] Step 1.5: Based on relevant human skeletal measurement databases, obtain human skeletal point cloud data under different vital signs parameters. The human skeleton includes the skull, thoracic skeleton, pelvis, femur, tibia, fibula, cervical vertebrae, lumbar vertebrae, humerus, radius, ulna, scapula, clavicle, sacrum, and foot bones (labeled as P). B_X );

[0084] Step 1.6: Match the obtained human skeleton point cloud data with the same vital signs parameters with the body surface point cloud data to obtain the feature point cloud set of the target human finite element model.

[0085] Step 2: Establish a basic finite element model deformation module. Decompose the basic human body finite element model to obtain the human skeleton and body surface finite element models. Then, extract the surface and skeleton feature points of the target human body. Generate deformation control points through a deep learning network model based on the established feature points. Finally, transform the basic body surface and skeleton finite element model into the target body surface and skeleton finite element model based on the RBF-TPS function.

[0086] like Figure 3 As shown, the process of establishing the target human body surface finite element model and the target human skeleton finite element model block in the basic finite element model deformation module includes the following steps:

[0087] Step 2.1: Obtain the THUMS human finite element model and use it as the base finite element model. Then, decompose the model to obtain the body surface finite element model (labeled as FEM). S ), finite element model of skeleton (labeled as FEM) B ), finite element model of soft tissue (labeled as FEM) R );

[0088] Step 2.2: Extract feature points from the finite element model of the body surface. In this step, the feature point selection uses the ISS (Intrinsic Shape Signatures) algorithm. The extracted feature points of the finite element model of the body surface are marked as FEM. S_F ;

[0089] Step 2.3: Extract feature points from the finite element model of the skeleton. In this step, the feature point selection uses the ISS (Intrinsic Shape Signatures) algorithm. The extracted feature points of the finite element model of the skeleton are labeled as FEMs. B_F ;

[0090] Step 2.4: Establish a deep learning network model for generating human body surface feature points, and convert the feature points of the finite element model of the body surface into FEM (Feature Element Model). S_F P and the target human body surface point cloud S Simultaneously, the model is fed in for training. If the model's accuracy is lower than the preset accuracy, the hyperparameters are adjusted and training is repeated. The loss function of the deep learning network model for generating human body surface feature points in this step is shown below:

[0091] L S (FEM S_F ,P S ,R S ) = L rec_S (FEM S_F ,P S )+λ1L perm_S (R S )+λ2L mfd_S (R S )

[0092] In the formula, R S For FEM S_F With P S The transformation matrix between them, λ1 and λ2 are the regularization coefficients of the loss function of the deep learning network model for generating human body surface feature points, L rec_S (FEM S_F ,P S The ) represents the error between the generated point cloud and the point cloud on the target human body surface, L perm_S (R S ) represents RS The error between L and the permutation matrix mfd_S (R S The nearest neighbor error is represented by the point cloud on the target human body surface and the corresponding points in the transformed point cloud.

[0093] Step 2.5: Establish a deep learning network model for generating human skeleton feature points, and convert the feature points of the skeleton finite element model into FEM (Feature Element Model). B_F P with the target human skeleton point cloud B_X Simultaneously, the model is fed in for training. If the model's accuracy is lower than the preset accuracy, the hyperparameters are adjusted and training is repeated. The loss function of the deep learning network model for generating human skeletal feature points in this step is shown below:

[0094] L B (FEM B_F ,P B_X ,R B ) = L rec_B (FEM B_F ,P B_X )+λ3L perm_B (R B )+λ4L mfd_B (R B )

[0095] In the formula, R B For FEM B_F With P B_X The transformation matrix between them, λ3 and λ4 are the regularization coefficients of the loss function of the deep learning network model for generating human skeletal feature points, and L rec_B (FEM B_F ,P B_X The ) represents the error between the generated point cloud and the target human skeleton point cloud, L perm_B (R B ) represents R B The error between L and the permutation matrix mfd_B (R B The nearest neighbor error between the target human skeleton point cloud and the corresponding points in the transformed point cloud is represented by ).

[0096] Step 2.6, transform the feature points of the finite element model of the body surface. S_F P is marked as the control point cloud of the target human body surface. C_S ;

[0097] Step 2.7, transform the feature points of the finite element model of the skeleton into FEM. B_F P is marked as the target human skeleton control point cloud. C_B ;

[0098] Step 2.8: Use the RBF-TPS function to perform FEM on the feature points of the finite element model of the body surface.S_F P is the source control point and the target human body surface control point cloud. C_S Using the target control points, a mesh transformation is performed to obtain a finite element model (labeled FEM) of the human body surface under different vital signs and postures. S_T );

[0099] Specifically, the RBF-TPS function is used to analyze the finite element model (FEM) of the body surface. S Transformation is performed to generate finite element models (FEMs) of the human body surface under different vital signs and postures. S_T When, the following formula should be satisfied:

[0100]

[0101] Among them, f s (x,y,z) represents the finite element model (FEM) of the body surface. S Finite element model (FEM) of human body surface under different physical signs and postures S_T Transformation function, which is based on the feature points of the finite element model of the body surface (FEM). S_F P and the control point cloud of the target human body surface C_S The conversion relationship between them;

[0102] Step 2.9: Use the RBF-TPS function to perform FEM on the feature points of the skeletal finite element model. S_F Source control point, target human skeleton control point cloud P C_B Using the target control points, a mesh transformation is performed to obtain finite element models of the human skeleton under different physical characteristics (labeled as FEM). B_T );

[0103] Specifically, the RBF-TPS function is used to analyze the finite element model (FEM) of the skeleton. B Transformation is performed to generate finite element models (FEMs) of the human skeleton under different physical characteristics. B_T When, the following formula should be satisfied:

[0104]

[0105] Among them, f B (x,y,z) represents the finite element model (FEM) of the skeleton. B Finite element model of human skeleton under different physical signs B_T Transformation function, which is based on the feature points FEM of the skeletal finite element model. B_F P with the target human skeleton control point cloud C_B The conversion relationship between them;

[0106] Step 3: Establish the target human finite element model integration module. First, query the corresponding SMPL model based on the transformed human finite element model, calculate the joints of the SMPL model, then map the joints to the corresponding body surface finite element model, and locate the transformed skeletal finite element model according to the joints. Finally, integrate the soft tissue finite element model and muscle units to complete the generation of the target human finite element model.

[0107] like Figure 4 As shown, the target human finite element model integration module mainly includes the positioning and assembly of the human body surface finite element model, bone finite element model and soft tissue finite element model through the following steps:

[0108] Step 3.1, transform the human body surface finite element model (FEM) with the same vital signs parameters. S_T With finite element model of skeleton (FEM) B_T To pair up;

[0109] Step 3.2, based on the transformed human finite element model (FEM) S_T In calculating the joint point locations, the corresponding SMPL human body model is queried based on the finite element model of the human body, and the joint points are calculated using the following formula:

[0110]

[0111] in, Here, K is the transformation matrix, K is the number of human joints, and N is the number of vertices in the SMPL human model. The vertices of the SMPL human body model are based on the shape blending process. The calculated joint positions;

[0112] Step 3.3: Locate the bones based on the joint positions. In this step, the transformed human body surface finite element model (FEM) is used. S_T The spatial position, shape, and human posture are completely identical to the corresponding SMPL human body model, therefore the joint point positions are also the same. The joint points of the SMPL human body model obtained in step 3.2 are mapped to the finite element model (FEM) of the human body surface. S_T Then, the transformed skeletal finite element model (FEM) is processed. B_T Positioning is based on the location of the joints;

[0113] Step 3.4: The internal organs of the human body, including the heart, kidneys and lungs, are obtained by scaling up the finite element models of the thoracic skeleton and pelvis surrounding the organs.

[0114] Step 3.5: The soft tissues of the trunk are produced by the combined inhibition of internal organs and body surface geometry;

[0115] Step 3.6: Add muscle linear elements. In this step, the muscle linear elements are attached to the corresponding human finite element model mesh nodes and automatically transform during the mesh transformation process to complete the generation of the target human finite element model.

[0116] The human finite element model generation method established through the above steps can realize two different human finite element model generation methods according to needs: generating from images and generating from feature parameter input.

[0117] Specifically, the method of generating images is as follows: Figure 5 As shown, the process is as follows: input the human body image into the target human body finite element model feature acquisition module, reconstruct the SMPL human body model through the SMPLify network model, extract the surface point cloud, and then input it and the corresponding bone point cloud into the basic finite element model deformation module and the target human body finite element model integration module in sequence to complete the generation of the human body finite element model.

[0118] Specifically, the input method based on feature parameters is as follows: Figure 6 As shown, the process is as follows: input the human body model's physical characteristics parameters (height, weight, gender, age) and the human body model's posture parameters (standing, driving, walking, cycling, running) into the human feature statistical regression model to obtain the human body surface point cloud. Then, input the point cloud along with the corresponding skeletal point cloud into the basic finite element model deformation module and the target human body finite element model integration module to complete the generation of the human body finite element model.

[0119] Although this application has been disclosed in detail with reference to the accompanying drawings, it should be understood that these descriptions are merely exemplary and not intended to limit the application of this application. The scope of protection of this application is defined by the appended claims and may include various variations, modifications, and equivalents of the invention without departing from the scope and spirit of this application.

Claims

1. A method for generating a human finite element model, characterized by: Comprise the following steps: Step 1, the establishment of target human finite element model feature acquisition module, first two-dimensional human image through human three-dimensional reconstruction deep learning network model obtains the human three-dimensional point cloud data under different signs, different postures, then the establishment of characteristic parameters and point cloud data between human feature statistical model, finally based on the relevant bone point cloud database obtains the corresponding sign parameter bone point cloud data, and with the surface point cloud data matching, obtain the target human finite element model feature point cloud set; Step 2, the establishment of basic finite element model deformation module; Step 3, the establishment of target human finite element model integration module, first according to the transformed human surface finite element model query corresponding SMPL model, calculate the SMPL model joint, then the joint is corresponding to the corresponding surface finite element model, and the transformed bone finite element model is positioned according to the joint, finally integrate soft tissue finite element model and muscle unit, complete the generation of target human finite element model; Step 2 in the following steps: Step 2.1, obtain THUMS human finite element model, and take it as the basic finite element model, and split the model to obtain the surface finite element model, bone finite element model and soft tissue finite element model respectively; Step 2.2, extract feature points of the body surface finite element model, the feature points are selected by using ISS algorithm, and the extracted feature points of the body surface finite element model are marked as ; Step 2.3, extract the feature points of the bone finite element model, the feature points are selected by using the ISS algorithm, and the extracted feature points of the bone finite element model are marked as ; Step 2.4, a deep learning network model for generating human body surface feature points is established, and the feature points of the surface finite element model are input into the deep learning network model and the target human body surface point cloud The deep learning network model is trained at the same time, and if the accuracy of the model is lower than the preset accuracy, the hyperparameters are adjusted for retraining; Step 2.5, a deep learning network model for generating human skeleton feature points is established, and the feature points of the bone finite element model are input into the deep learning network model for training and the target human skeleton point cloud Meanwhile, the deep learning network model is trained, and if the accuracy of the model is lower than the preset accuracy, the hyperparameters are adjusted for retraining; Step 2.

6. Transform the feature points of the surface finite element model Label the target human body surface control point cloud ; Step 2.

7. Feature points of the transformed bone finite element model Labeling target human bone control point cloud ; Step 2.8, using RBF-TPS function, with feature points of surface finite element model as source control points, target human body surface control point cloud as target control points, perform mesh transformation, obtain human body surface finite element model under different signs and different postures, marked as ; Step 2.9, using RBF-TPS function, with the feature points of the bone finite element model as the source control points, the target human bone control point cloud as the target control points, perform mesh transformation, obtain the human bone finite element model under different signs, marked as .

2. The method of claim 1, wherein: Step 1 in the following steps: Step 1.1, obtain human two-dimensional RGB image under different sign parameters and postures, and extract corresponding sign parameter matrix and posture parameter matrix, the sign parameters include height, weight, gender, age, and posture parameter labels include standing, driving, walking, cycling and running; Step 1.2, realize human three-dimensional reconstruction through deep learning network model of single RGB image, the network model extracts features from human RGB image, regresses SMPL model parameters, realizes three-dimensional reconstruction from human RGB image to human SMPL model; where the SMPL model is determined by 10-dimensional body shape parameters and 72-dimensional human joint parameters to determine a corresponding human three-dimensional mesh model represented by triangular facets, and the model expression is: （1） wherein The calculation is: （2） In the formula, SMPL human model template representing an average body shape, and represent correction data of body shape parameters and human joint parameters on the average template, represents the coordinates of each key point obtained from the surface vertices of the model, is a deformation weight; Step 1.3, extract the SMPL human model vertex to obtain the target human point cloud data, marked as ; Step 1.

4. In the manner of indirect statistical model, the sign parameter matrix and the behavior parameter matrix are combined into the human feature parameter matrix by mapping the SMPL model parameters to the corresponding human body model, and are marked as The regression model can be expressed as: ； wherein denotes predicted SMPL model parameters, denotes SMPL human model template parameters of an average body shape, denotes a regression parameter matrix; Step 1.5, based on the relevant human bone measurement database, obtain human bone point cloud data under different sign parameters, the human bones include skull, chest bone, pelvis, femur, tibia, fibula, cervical vertebra, lumbar vertebra, humerus, radius, ulna, scapula, clavicle, sacrum and foot bone; Step 1.6, match the obtained human bone point cloud data with the surface point cloud data under the same sign parameters to obtain the target human finite element model feature point cloud set.

3. The method of claim 1, wherein: The loss function of the human surface feature point generation deep learning network model is as follows: ； In the formula, is and the conversion matrix between and is a regularization coefficient of the loss function of the deep learning network model generated for the human body surface feature points, represents the error between the feature points of the surface finite element model and the target human body surface point cloud, represents the error between represents the nearest neighbor relationship error between the target human body surface point cloud and the corresponding points of the transformed point cloud; The loss function of the human bone feature point generation deep learning network model is as follows: ； In the formula, is and the conversion matrix between and is a regularization coefficient of the loss function of the deep learning network model for human skeleton feature points, represents the error between the feature points of the skeletal finite element model and the target human skeleton point cloud, represents the error between the transformation matrix and the permutation matrix, represents the near-neighbor relationship error between the target human skeleton point cloud and the corresponding points of the transformed point cloud; The RBF-TPS function is used to transform the surface finite element model to generate the surface finite element model of the human body in different signs and different postures The following formula should be met: ； wherein, representing a body surface finite element model to a human body surface finite element model in different signs, different postures transform function, the function is based on the feature points of the body surface finite element model with the target human body surface control point cloud between the conversion relationship; Adopt RBF-TPS function to the bone finite element model Transform to generate the human bone finite element model under different signs When, should satisfy the following formula: ； wherein, representing a skeleton finite element model to a human skeleton finite element model under different signs transform function, which is based on the feature points of the skeleton finite element model and the target human skeleton control point cloud between the conversion relationship.

4. The human finite element model generation method according to claim 1, characterized in that: Step 3.1, the same body surface finite element model of human body with transformed sign parameters with the bone finite element model pairing; Step 3.2, based on the transformed human body surface finite element model Calculate the joint positions, in this step, the corresponding SMPL human model is queried according to the human body finite element model, and the joint positions are calculated according to the following formula: ； wherein, is a transformation matrix, is the number of human body joints, is the number of SMPL human model vertices, is the SMPL human model vertices after shape blending based shaping, is the computed joint positions; Step 3.

3. Position the skeleton according to the joint locations, in this step, the transformed human body surface finite element model The spatial position and shape, the human pose and the corresponding SMPL human model are completely consistent, so the joint locations are also the same. The joint locations of the SMPL human model obtained in step 3.2 are corresponded to the human body surface finite element model Then the transformed skeleton finite element model is positioned according to the joint locations; Step 3.4, the human internal organs including heart, kidney and lung are scaled through the chest bone and pelvic finite element model of the organ periphery; Step 3.5, the trunk soft tissue is generated by the internal organs and the surface geometry together; Step 3.6, add muscle linear unit, in this step, the muscle linear unit is attached to the corresponding human finite element model grid node, and automatically transforms with the grid transformation process to complete the generation of the target human finite element model.

Images