CN115640714A - Method for generating finite element model of human body - Google Patents
Method for generating finite element model of human body Download PDFInfo
- Publication number
- CN115640714A CN115640714A CN202211159539.2A CN202211159539A CN115640714A CN 115640714 A CN115640714 A CN 115640714A CN 202211159539 A CN202211159539 A CN 202211159539A CN 115640714 A CN115640714 A CN 115640714A
- Authority
- CN
- China
- Prior art keywords
- finite element
- human body
- model
- element model
- fem
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Instructional Devices (AREA)
Abstract
The invention provides a method for generating a finite element model of a human body, which comprises the following steps: step 1, establishing a target human body finite element model characteristic acquisition module, firstly, acquiring human body three-dimensional point cloud data of different physical signs and different postures of a two-dimensional human body image through a human body three-dimensional reconstruction deep learning network model, and establishing a human body characteristic statistical model; step 2, establishing a deformation module of the basic finite element model, splitting the basic human finite element model, and acquiring human skeleton and body surface finite element models; and 3, establishing a target human body finite element model integration module, firstly inquiring a corresponding SMPL model according to the transformed human body finite element model, calculating joint points of the SMPL model, and finally integrating a soft tissue finite element model and a muscle unit to finish the generation of the target human body finite element model. The method realizes the positioning of the corresponding skeleton and solves the problem that the position of the skeleton needs to be manually adjusted in the prior human finite element model.
Description
The technical field is as follows:
the invention relates to the crossing field of automobile passive safety and computer science, in particular to a method for generating a finite element model of a human body.
Technical Field
The finite element human model is an important tool for evaluating the safety of the vehicle. The human body with different physical signs has different damage responses under different collision working conditions. The existing human body finite element model is only 5 th 、50 th 、95 th And the finite element models under several specific physical signs are used, so that the damage conditions of the human body under various physical signs cannot be reflected in the simulation calculation. The existing establishment of a human body finite element model based on multiple physical signs has three problems. Firstly, a large amount of human body surface point cloud data is needed for establishing a human body surface finite element model, and the prior art mainly obtains the point cloud data through human body scanning, and the process is time-consuming and labor-consuming. Secondly, the process of establishing the finite element model of the human body is very complicated, and the quality of the finite element grid needs to be ensured to carry out simulation calculation while the geometric precision is ensured. Finally, the existing grid deformation technology can realize the transformation of the existing human body finite element model according to the physical sign requirements, but only can transform the human body model under two postures of standing and driving, and cannot meet the requirement of real-time rapid generation of the human body model under multiple postures.
Disclosure of Invention
In view of the above, in order to solve the above problems, the present invention provides a method for generating a finite element model of a human body.
The invention specifically relates to a method for generating a finite element model of a human body, which is characterized by comprising the following steps: the method comprises the following steps:
step 2, establishing a basic finite element model deformation module, splitting the basic human body finite element model, obtaining human body bones and body surface finite element models, then extracting body surface and bone characteristic points of a target human body, generating a deep learning network model through the established characteristic points to generate deformation control points, and finally converting the basic body surface and bone finite element model into the target body surface and bone finite element model based on the RBF-TPS function;
and 3, establishing a target human body finite element model integration module, firstly inquiring a corresponding SMPL model according to the transformed human body finite element model, calculating joint points of the SMPL model, then corresponding the joint points to a corresponding body surface finite element model, positioning the transformed bone finite element model according to the joint points, and finally integrating a soft tissue finite element model and a muscle unit to finish the generation of the target human body finite element model.
Further, step 1 comprises the following steps:
step 1.1, obtaining human body two-dimensional RGB images under different physical sign parameters and postures, and extracting corresponding physical sign parameter matrixes and posture parameter matrixes, wherein the physical sign parameters comprise height, weight, gender and age, and the posture parameter labels comprise standing, driving, walking, riding and running;
step 1.2, a deep learning network model for human body three-dimensional reconstruction is realized through a single RGB image, and the network model regresses parameters of an SMPL model through extracting features in the human body RGB image, so that three-dimensional reconstruction from the human body RGB image to the human body SMPL model is realized;
the SMPL model determines a corresponding human body three-dimensional mesh model represented by a triangular patch through a 10-dimensional body type parameter beta and a 72-dimensional human body joint parameter theta, and the model expression is as follows:
wherein the content of the first and second substances,
the calculation formula is as follows:
in the formula (I), the compound is shown in the specification,
SMPL human model templates representing average body types,
and
representing the correction data of the body type parameters and the human body joint parameters on the average template,
representing the coordinates of each key point obtained from the surface vertex of the model, and w is the deformation weight;
step 1.3, extracting the top point of the SMPL human body model to obtain target human body point cloud data marked as PS;
step 1.4, establishing a statistical regression model by adopting an indirect statistical model mode through a physical sign parameter matrix, a behavior parameter matrix, an SMPL human body shape parameter matrix and an SMPL human body posture parameter matrix, then mapping to a corresponding human body model through SMPL model parameters, merging the physical sign parameter matrix and the behavior parameter matrix into a human body characteristic parameter matrix, and marking as F 1*5 The regression model can be expressed as:
in the formula, T pre Represents the predicted parameters of the SMPL model,
SMPL human model template parameters, M, representing average body type 5*82 Representing a regression parameter matrix.
Step 1.5, acquiring human skeleton point cloud data under different physical sign parameters based on a related human skeleton measurement database, wherein the human skeleton comprises a skull, a thoracic skeleton, a pelvis, a femur, a tibia, a fibula, a cervical vertebra, a lumbar vertebra, a humerus, a radius, an ulna, a scapula, a clavicle, a sacrum and a foot bone;
and step 1.6, matching the obtained human skeleton point cloud data with the same physical sign parameters with the obtained body surface point cloud data to obtain a target human finite element model characteristic point cloud set.
Further, step 2 comprises the following steps:
step 2.1, obtaining a THUMS human body finite element model, taking the THUMS human body finite element model as a basic finite element model, splitting the model, and respectively obtaining a body surface finite element model, a skeleton finite element model and a soft tissue finite element model;
step 2.2, extracting characteristic points of the finite element model of the body surface, selecting the characteristic points and adopting an ISS (Intrinsic shape signatures) algorithm, and marking the extracted characteristic points of the finite element model of the body surface as FEM (finite element model) S_F ；
Step 2.3, extracting characteristic points of the bone finite element model, selecting the characteristic points by adopting an ISS (Intrinsic shape signatures) algorithm, and marking the extracted characteristic points of the bone finite element model as FEM (finite element model) B_F ；
Step 2.4, establishing body surface feature points of the human body to generate a deep learning network model, and enabling feature points FEM of a finite element model of the body surface S_F Point cloud P on body surface of target human body S And simultaneously feeding the model for training, if the model accuracy is lower than the preset accuracy, adjusting the hyper-parameters for training again, wherein the loss function of the deep learning network model generated by the human body surface characteristic points in the step is as follows:
L S (FEM S_F ,P S ,R S )＝L rec_S (FEM S_F ,P S )+λ 1 L perm_S (R S )+λ 2 L mfd_S (R S )
in the formula, R S Is FEM S_F And P S A transformation matrix between, λ 1 And λ 2 Generating depths for body surface feature pointsRegularization coefficient of loss function of learning network model, L rec_S (FEM S_F ,P S ) Representing the error between the generated point cloud and the target body surface point cloud, L perm_S (R S ) Represents R S Error with permutation matrix, L mfd_S (R S ) Expressing the neighbor relation error between the point cloud of the body surface of the target human body and the corresponding point of the point cloud after transformation;
step 2.5, establishing human skeleton characteristic points to generate a deep learning network model, and enabling feature points FEM of a skeleton finite element model B_F And target human skeleton point cloud P B_X And simultaneously feeding the model for training, if the accuracy rate of the model is lower than the preset accuracy rate, adjusting the hyper-parameters for training again, wherein the loss function of the human skeleton feature point generated deep learning network model in the step is as follows:
L B (FEM B_F ,P B_X ,R B )＝L rec_B (FEM B_F ,P B_X )+λ 3 L perm_B (R B )+λ 4 L mfd_B (R B )
in the formula, R B Is a FEM B_F And P B_X A transformation matrix between, λ 3 And λ 4 Generating a regularization coefficient of a loss function of a deep learning network model for the human skeleton feature points, L rec_B (FEM B_F ,P B_X ) Representing the error between the generated point cloud and the target human skeleton point cloud, L perm_B (R B ) Represents R B Error with permutation matrix, L mfd_B (R B ) Expressing the neighbor relation error between the target human skeleton point cloud and the corresponding point of the transformed point cloud;
step 2.6, feature points FEM of the transformed body surface finite element model S_F Marking as a target human body surface control point cloud P C_S ；
Step 2.7, feature points FEM of the transformed bone finite element model B_F Marking as a target human skeletal control point cloud P C_B ；
Step 2.8, adopting RBF-TPS function to use the feature point FEM of the finite element model of the body surface S_F Is a source ofControl points, target body surface control point clouds P C_S Carrying out grid transformation for the target control point to obtain human body surface finite element models with different signs and different postures, and marking as FEM S_T ；
FEM (finite element model) for body surface finite element model by adopting RBF-TPS (radial basis function-TPS) function S Transforming to generate finite element models FEM of human body surface under different physical signs and different postures S_T Then, the following equation should be satisfied:
wherein, f s (x, y, z) represents a finite element model FEM of the body surface S Finite element model FEM of human body surface under different physical signs and different postures S_T Transforming a function based on feature points FEM of finite element model of body surface S_F Controlling point cloud P with target human body surface C_S The conversion relationship between them;
step 2.9, adopting RBF-TPS function and using feature points FEM of finite element model of skeleton B_F As source control point, target human skeleton control point cloud P C_B Carrying out grid transformation for the target control point to obtain human skeleton finite element models under different physical signs, and marking as FEM B_T ；
Specifically, a bone finite element model FEM is subjected to RBF-TPS function B Transforming to generate finite element model FEM of human skeleton under different physical signs B_T Then, the following equation should be satisfied:
wherein f is B (x, y, z) denotes a bone finite element model FEM B Finite element model FEM of human skeleton under different physical signs B_T Transformation function based on feature points FEM of finite element model of bone B_F Control point cloud P with target human skeleton C_B The conversion relationship between them;
further, step 3.1, transform the volumeHuman body surface finite element model FEM with same sign parameters S_T Finite element model FEM with skeleton B_T Carrying out pairing;
step 3.2, based on the transformed finite element model FEM of human body S_T Calculating the position of a joint point, wherein in the step, a corresponding SMPL human body model is inquired according to the human body finite element model, and the joint point is calculated according to the following formula:
wherein the content of the first and second substances,
for the transformation matrix, K is the number of human joint points, N is the number of SMPL human model vertexes,
for SMPL mannequin vertices after shape-blending,
calculating the position of the joint point;
3.3, positioning the skeleton according to the joint point position, wherein the transformed finite element model FEM of the human body surface S_T The spatial position, the shape, the human body posture and the like are completely consistent with those of the corresponding SMPL human body model, so the joint point positions are also the same, and the SMPL human body model joint obtained in the step 3.2 is corresponding to the finite element model FEM of the human body surface S_T Then transforming the bone finite element model FEM B_T Positioning according to the position of the joint point;
step 3.4, obtaining internal organs of the human body, including a heart, a kidney and a lung, by zooming through finite element models of thoracic bones and pelvic bones on the periphery of the organs;
step 3.5, inhibiting the generation of soft trunk tissues by internal organs and body surface geometry together;
and 3.6, adding muscle linear units, wherein the muscle linear units are attached to corresponding grid nodes of the human body finite element model and automatically change along with the grid changing process to finish the generation of the target human body finite element model.
The invention has the beneficial effects that:
(1) The invention provides a mode of generating a finite element model from a two-dimensional RGB image to a human body, which is an end-to-end generating method, and can automatically generate a corresponding human body finite element model according to any two-dimensional image with the human body, the generated finite element model has the characteristics of flexibility and accuracy, and the whole generating process does not need manual intervention;
(2) The target human body finite element model feature acquisition module provided by the invention can realize two different human body point cloud generation modes of generating according to pictures and inputting according to feature parameters according to needs, and the two modes can realize target human body point cloud generation under different physical sign parameters and different postures; the basic finite element model deformation module can carry out deep learning model training according to point clouds at different positions of a human body to obtain the accurate characteristic points at the corresponding positions, and solves the problem that the characteristic points need to be manually placed in the conventional finite element model deformation;
(3) The target human body finite element model integration module provided by the invention can automatically calculate the position of the joint point according to the transformed human body surface finite element model, realizes the positioning of the corresponding skeleton, and solves the problem that the previous human body finite element model needs to manually adjust the position of the skeleton.
Drawings
FIG. 1 is a schematic diagram of a finite element model generation system of a human body.
FIG. 2 is a schematic flow diagram of a finite element model feature acquisition module for a target human body.
FIG. 3 is a schematic flow chart of a basic finite element model deformation module.
FIG. 4 is a schematic diagram of finite element model deformation of bone and soft tissue.
FIG. 5 is a schematic diagram of a finite element model generation process of a human body based on image input.
FIG. 6 is a schematic diagram of a finite element model generation process of a human body based on characteristic parameter input.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present application more comprehensible, embodiments accompanying the present application are described in detail below with reference to the accompanying drawings.
As shown in fig. 1, the embodiment provides a system for generating a finite element model of a human body, which specifically comprises three modules, namely a target finite element model feature acquisition module, a basic finite element model deformation module and a target finite element model integration module; wherein:
the target human body finite element model characteristic acquisition module is used for acquiring a target human body surface and skeleton point cloud;
the basic finite element model deformation module is used for respectively transforming the basic human body surface finite element model and the skeleton finite element model into the target human body surface finite element model and the skeleton model
The target human body finite element model integration module is used for positioning the transformed target human body skeleton finite element model into the target human body surface finite element model, and simultaneously integrating the soft tissue finite element model, the muscle unit and the like to form a complete human body finite element model
The embodiment also provides a method for generating the human body finite element model, which specifically comprises the following steps:
This step is achieved by:
step 1.1, obtaining human body two-dimensional RGB images under different physical sign parameters and postures, and extracting corresponding physical sign parameter matrixes and posture parameter matrixes, wherein the physical sign parameters comprise height, weight, sex, age and the like, and the posture parameter labels comprise standing, driving, walking, riding, running and the like.
And step 1.2, realizing a deep learning network model of human body three-dimensional reconstruction through a single RGB image, wherein the network model realizes the three-dimensional reconstruction from the human body RGB image to the human body SMPL model by extracting the characteristics in the human body RGB image and regressing the parameters of the SMPL model.
The SMPL model is a parameterized human body model library proposed by Mapu in Germany, and can determine a corresponding human body three-dimensional mesh model represented by a triangular patch through a 10-dimensional body type parameter beta and a 72-dimensional human body joint point parameter theta, wherein the model expression is as follows:
wherein the content of the first and second substances,
the calculation formula is as follows:
in the formula (I), the compound is shown in the specification,
SMPL human model templates representing average body types,
and
representing the correction data of the body type parameters and the human body joint parameters on the average template,
the coordinates of each keypoint taken from the surface vertices of the model are represented, and w is the deformation weight.
In the step, a pre-trained SMPLify model is used as a basic framework to construct a human body model reconstruction deep learning network model, the two-dimensional RGB images of the human body under different physical signs and different postures obtained in the step 1 are fed into the model, the SMPL human body model, the corresponding body type parameter β and the corresponding joint point parameter θ under different physical signs and different posture parameters are obtained in the model, the SMPLify model is mainly characterized in that the minimum value of an objective function is calculated from the image, and the objective function is expressed as:
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)
where K denotes the camera parameters of the body image, J est Representing the two-dimensional coordinates, λ, of each keypoint θ 、λ a 、λ sp 、λ β Representing the weight coefficients.
Step 1.3, extracting the top point of the SMPL human body model to obtain target human body point cloud data marked as PS;
step 1.4, establishing a statistical regression model by adopting an indirect statistical model mode through a physical sign parameter matrix, a behavior parameter matrix, an SMPL human body shape parameter matrix and an SMPL human body posture parameter matrix, then mapping to a corresponding human body model through SMPL model parameters, merging the physical sign parameter matrix and the behavior parameter matrix into a human body characteristic parameter matrix, and marking as F 1*5 The regression model can be expressed as:
in the formula, T pre Represents the predicted parameters of the SMPL model,
SMPL human model template parameters, M, representing average body type 5*82 Representing a regression parameter matrix.
According to the formula, corresponding SMPL model parameters can be generated by the human body characteristic parameters which are combined randomly according to the formula, and then the corresponding human body surface three-dimensional point cloud is generated.
Step 1.5, based on a relevant human skeleton measurement database, acquiring human skeleton point cloud data under different physical sign parameters, wherein the human skeleton comprises a skull, a thoracic skeleton, a pelvis, a femur, a tibia, a fibula, a cervical vertebra, a lumbar vertebra, a humerus, a radius, an ulna, a scapula, a clavicle, a sacrum, a foot bone and the like (marked as P) B_X )；
And step 1.6, matching the obtained human skeleton point cloud data with the same physical sign parameters with the obtained body surface point cloud data to obtain a target human finite element model characteristic point cloud set.
And 2, establishing a basic finite element model deformation module, splitting the basic human body finite element model, acquiring human body skeleton and body surface finite element models, then extracting body surface and skeleton characteristic points of the target human body, generating a deep learning network model through the established characteristic points to generate deformation control points, and finally converting the basic body surface and skeleton finite element models into target body surface and skeleton finite element models based on the RBF-TPS function.
As shown in fig. 3, the process of establishing the target body surface finite element model and the target body skeleton finite element model block in the basic finite element model deformation module includes the following steps:
step 2.1, obtaining a THUMS human body finite element model, taking the THUMS human body finite element model as a basic finite element model, splitting the model, and respectively obtaining a body surface finite element model (marked as FEM) S ) Bone finite element model (labeled FEM) B ) Soft tissue finite element model (labeled FEM) R )；
Step 2.2, extracting characteristic points of the finite element model of the body surface, wherein the characteristic points are selected by adopting an ISS (internal shape descriptors) algorithm, and the extracted characteristic points of the finite element model of the body surface are marked as FEM (finite element modeling) S_F ；
Step 2.3, extracting bone finite element model characteristicsSelecting characteristic points in the step, adopting an ISS (internal shape descriptors) algorithm, and marking the characteristic points of the extracted bone finite element model as FEM (finite element model) B_F ；
Step 2.4, establishing a human body surface feature point to generate a deep learning network model, and carrying out FEM (finite element model) on the feature points of the finite element model of the body surface S_F Point cloud P of human body surface S And simultaneously feeding the model for training, if the accuracy of the model is lower than the preset accuracy, adjusting the hyper-parameters for training again, wherein in the step, the loss function of the deep learning network model generated by the human body surface feature points is as follows:
L S (FEM S_F ,P S ,R S )＝L rec_S (FEM S_F ,P S )+λ 1 L perm_S (R S )+λ 2 L mfd_S (R S )
in the formula, R S Is FEM S_F And P S A transformation matrix between, λ 1 And λ 2 Generating a loss function regularization coefficient L of a deep learning network model for the body surface characteristic points rec_S (FEM S_F ,P S ) Representing the error between the generated point cloud and the target body surface point cloud, L perm_S (R S ) Represents R S Error with permutation matrix, L mfd_S (R S ) And expressing the neighbor relation error between the point cloud of the body surface of the target human body and the corresponding point of the point cloud after transformation.
Step 2.5, establishing human skeleton feature points to generate a deep learning network model, and performing FEM (finite element model) on the skeleton finite element model feature points B_F And target human skeleton point cloud P B_X And simultaneously feeding the model for training, if the accuracy of the model is lower than the preset accuracy, adjusting the hyper-parameters for training again, wherein the loss function of the deep learning network model generated by the human skeleton characteristic points in the step is as follows:
L B (FEM B_F ,P B_X ,R B )＝L rec_B (FEM B_F ,P B_X )+λ 3 L perm_B (R B )+λ 4 L mfd_B (R B )
in the formula, R B Is FEM B_F And P B_X A transformation matrix between, λ 3 And λ 4 Generating a regularization coefficient of a loss function of a deep learning network model for the human skeleton feature points, L rec_B (FEM B_F ,P B_X ) Representing the error between the generated point cloud and the target human skeleton point cloud, L perm_B (R B ) Represents R B Error from permutation matrix, L mfd_B (R B ) And expressing the neighbor relation error between the target human skeleton point cloud and the corresponding point of the point cloud after transformation.
Step 2.6, feature points FEM of the transformed body surface finite element model S_F Marking as target human body surface control point cloud P C_S ；
Step 2.7, feature points FEM of the transformed bone finite element model B_F Marking as a target human skeletal control point cloud P C_B ；
Step 2.8, adopting RBF-TPS function to use the feature point FEM of the finite element model of the body surface S_F As a source control point, a target human body surface control point cloud P C_S Carrying out mesh transformation for the target control point to obtain human body surface finite element models (marked as FEM) with different physical signs and different postures S_T )；
Specifically, a finite element model FEM of the body surface is subjected to RBF-TPS function S Transforming to generate finite element models FEM of human body surface under different physical signs and different postures S_T Then, the following equation should be satisfied:
wherein f is s (x, y, z) representing a finite element model FEM of the body surface S Finite element model FEM of human body surface under different physical signs and different postures S_T Transformation function based on feature points FEM of finite element model of body surface S_F Controlling point cloud P with target human body surface C_S The conversion relationship between them;
step 2.9, adopting RBF-TPS function and using skeleton finite element modelFEM (finite element model) S_F As source control point, target human skeleton control point cloud P C_B Carrying out mesh transformation for target control points to obtain human skeleton finite element models (marked as FEM) under different physical signs B_T )；
Specifically, a RBF-TPS function is adopted to carry out FEM on a bone finite element model B Transforming to generate finite element model FEM of human skeleton under different physical signs B_T Then, the following equation should be satisfied:
wherein, f B (x, y, z) denotes a finite element model FEM of the bone B Finite element model FEM of human skeleton under different physical signs B_T Transformation function based on feature points FEM of finite element model of bone B_F Control point cloud P with target human skeleton C_B The conversion relationship between them;
and 3, establishing a target human body finite element model integration module, firstly inquiring a corresponding SMPL model according to the transformed human body finite element model, calculating joint points of the SMPL model, then corresponding the joint points to a corresponding body surface finite element model, positioning the transformed bone finite element model according to the joint points, and finally integrating a soft tissue finite element model and a muscle unit to complete the generation of the target human body finite element model.
As shown in fig. 4, the target human body finite element model integration module mainly comprises a human body surface finite element model, and positioning and assembling between a bone finite element model and a soft tissue finite element model are realized through the following steps:
step 3.1, transforming the human body surface finite element model FEM with the same physical sign parameters S_T FEM (finite element model) of bone and bone finite element model B_T Carrying out pairing;
step 3.2, based on the transformed finite element model FEM of the human body S_T Calculating the position of a joint point, in the step, inquiring a corresponding SMPL human body model according to the human body finite element model, and calculating the joint point according to the following formula:
wherein, the first and the second end of the pipe are connected with each other,
for the transformation matrix, K is the number of human body joint points, N is the number of SMPL human body model vertexes,
for SMPL mannequin vertices shaped based on shape blending,
calculating the position of the joint point;
3.3, positioning the skeleton according to the joint point position, wherein the transformed finite element model FEM of the human body surface S_T The spatial position, shape, body posture and the like are completely consistent with those of the corresponding SMPL body model, so the joint positions are also the same, and the joint of the SMPL body model obtained in the step 3.2 is corresponding to the finite element model FEM of the body surface of the human body S_T Then, the transformed bone finite element model FEM is used B_T Positioning according to the position of the joint point;
step 3.4, obtaining internal organs of the human body, including a heart, a kidney and a lung, by zooming through finite element models of thoracic bones and pelvic bones on the periphery of the organs;
step 3.5, inhibiting the generation of the soft trunk tissues by internal organs and body surface geometry;
and 3.6, adding muscle linear units, wherein the muscle linear units are attached to corresponding grid nodes of the human body finite element model and automatically change along with the grid change process to complete the generation of the target human body finite element model.
The method for generating the human body finite element model established through the steps can realize two different modes of generating the human body finite element model according to the picture and inputting the human body finite element model according to the characteristic parameters according to the requirements.
Specifically, the generating manner according to the picture is shown in fig. 5, and the process includes: inputting a human body picture into a target human body finite element model characteristic acquisition module, reconstructing an SMPL (simple Markov simulation) human body model through an SMPLfy network model, extracting surface point clouds, and then sequentially inputting the surface point clouds and corresponding skeleton point clouds into a basic finite element model deformation module and a target human body finite element model integration module to complete generation of a human body finite element model;
specifically, the method for inputting the characteristic parameters is shown in fig. 6, and the process is as follows: inputting the physical characteristics parameters (height, weight, sex and age) of the human body model to be established and the posture parameters (standing, driving, walking, riding and running) of the human body model to be established into a human body characteristic statistical regression model to obtain the point cloud of the body surface of the human body, and then sequentially inputting the point cloud of the human body and the corresponding skeleton point cloud into a basic finite element model deformation module and a target human body finite element model integration module to complete the generation of the human body finite element model.
Although the present application has been disclosed in detail with reference to the accompanying drawings, it is to be understood that such description is merely illustrative and is not intended to limit the application of the present application. The scope of the present application is defined by the appended claims and may include various modifications, adaptations, and equivalents of the subject invention without departing from the scope and spirit of the present application.
Claims (4)
1. A method for generating a finite element model of a human body is characterized by comprising the following steps: the method comprises the following steps:
step 1, establishing a target human body finite element model characteristic acquisition module, firstly, acquiring human body three-dimensional point cloud data of different physical signs and different postures of a two-dimensional human body image through a human body three-dimensional reconstruction deep learning network model, then establishing a human body characteristic statistical model between characteristic parameters and the point cloud data, and finally acquiring bone point cloud data of corresponding physical sign parameters based on a related bone point cloud database and matching the bone point cloud data with body surface point cloud data;
step 2, establishing a basic finite element model deformation module, splitting the basic human body finite element model, obtaining human body bones and body surface finite element models, then extracting body surface and bone characteristic points of a target human body, generating a deep learning network model through the established characteristic points to generate deformation control points, and finally converting the basic body surface and bone finite element model into the target body surface and bone finite element model based on the RBF-TPS function;
and 3, establishing a target human body finite element model integration module, firstly inquiring a corresponding SMPL model according to the transformed human body finite element model, calculating joint points of the SMPL model, then corresponding the joint points to a corresponding body surface finite element model, positioning the transformed bone finite element model according to the joint points, and finally integrating a soft tissue finite element model and a muscle unit to complete the generation of the target human body finite element model.
2. A finite element model generation method of human body as claimed in claim 1, wherein: the step 1 comprises the following steps:
step 1.1, obtaining human body two-dimensional RGB images under different physical sign parameters and postures, and extracting corresponding physical sign parameter matrixes and posture parameter matrixes, wherein the physical sign parameters comprise height, weight, gender and age, and the posture parameter labels comprise standing, driving, walking, riding and running;
step 1.2, realizing a deep learning network model of human body three-dimensional reconstruction through a single RGB image, wherein the network model realizes the three-dimensional reconstruction from the human body RGB image to the human body SMPL model by extracting the characteristics in the human body RGB image and regressing the parameters of the SMPL model;
the SMPL model determines a corresponding human body three-dimensional mesh model represented by a triangular patch through a 10-dimensional body type parameter beta and a 72-dimensional human body joint parameter theta, and the model expression is as follows:
wherein the content of the first and second substances,
the calculation formula is as follows:
in the formula (I), the compound is shown in the specification,
SMPL human model templates representing average body types,
and with
Representing the correction data of the body type parameters and the human body joint parameters on the average template,
representing the coordinates of each key point obtained from the surface vertex of the model, and w is the deformation weight;
step 1.3, extracting the top point of the SMPL human body model to obtain target human body point cloud data marked as P S ；
Step 1.4, establishing a statistical regression model by adopting an indirect statistical model mode through a physical sign parameter matrix, a behavior parameter matrix, an SMPL human body shape parameter matrix and an SMPL human body posture parameter matrix, then mapping to a corresponding human body model through SMPL model parameters, merging the physical sign parameter matrix and the behavior parameter matrix into a human body characteristic parameter matrix, and marking as F 1*5 The regression model can be expressed as:
in the formula, T pre Represents the predicted parameters of the SMPL model,
SMPL human model template parameters, M, representing average body type 5*82 Representing a regression parameter matrix.
Step 1.5, acquiring human skeleton point cloud data under different physical sign parameters based on a related human skeleton measurement database, wherein the human skeleton comprises a skull, a thoracic skeleton, a pelvis, a femur, a tibia, a fibula, a cervical vertebra, a lumbar vertebra, a humerus, a radius, an ulna, a scapula, a clavicle, a sacrum and a foot bone;
and step 1.6, matching the obtained human skeleton point cloud data with the same physical sign parameters with the obtained body surface point cloud data to obtain a target human finite element model characteristic point cloud set.
3. A finite element model generation method of human body as claimed in claim 1, wherein: the step 2 comprises the following steps:
step 2.1, obtaining a THUMS human body finite element model, taking the THUMS human body finite element model as a basic finite element model, splitting the model, and respectively obtaining a body surface finite element model, a skeleton finite element model and a soft tissue finite element model;
step 2.2, extracting characteristic points of the finite element model of the body surface, selecting the characteristic points and adopting an ISS (Intrinsic shape signatures) algorithm, and marking the extracted characteristic points of the finite element model of the body surface as FEM (finite element model) S_F ；
Step 2.3, extracting characteristic points of the bone finite element model, selecting the characteristic points by adopting an ISS (Intrinsic shape signatures) algorithm, and marking the extracted characteristic points of the bone finite element model as FEM (finite element model) B_F ；
Step 2.4, establishing a human body surface feature point to generate a deep learning network model, and carrying out FEM (finite element model) on the feature points of the finite element model of the body surface S_F Point cloud P of human body surface S And simultaneously feeding the model for training, if the accuracy of the model is lower than the preset accuracy, adjusting the hyper-parameters for training again, wherein in the step, the loss function of the deep learning network model generated by the human body surface feature points is as follows:
L S (FEM S_F ，P S ，R S )＝L rec_S (FEM S_F ，P S )+λ 1 L perm_S (R S )+λ 2 L mfd_S (R S )
in the formula, R S Is FEM S_F And P S A transformation matrix between, λ 1 And λ 2 Generating a loss function regularization coefficient L of a deep learning network model for the body surface characteristic points rec_S (FEM S_F ，P S ) Representing the error between the generated point cloud and the target body surface point cloud, L perm_S (R S ) Represents R S Error with permutation matrix, L mfd_S (R S ) Expressing the neighbor relation error between the point cloud of the body surface of the target human body and the corresponding point of the point cloud after transformation;
step 2.5, establishing human skeleton feature points to generate a deep learning network model, and performing FEM (finite element model) on the skeleton finite element model feature points B_F And target human skeleton point cloud P B_X And simultaneously feeding the model for training, if the accuracy rate of the model is lower than the preset accuracy rate, adjusting the hyper-parameters for training again, wherein the loss function of the human skeleton feature point generated deep learning network model in the step is as follows:
in the formula, R B Is a FEM B_F And P B_X A transformation matrix between, λ 3 And λ 4 Generating a loss function regularization coefficient L of a deep learning network model for human skeleton feature points rec_B (FEM B_F ，P B_X ) Representing the error between the generated point cloud and the target human skeleton point cloud, L perm_B (R B ) Represents R B Error with permutation matrix, L mfd_B (R B ) Expressing the neighbor relation error between the target human skeleton point cloud and the corresponding point of the transformed point cloud;
step 2.6, feature points FEM of the transformed body surface finite element model S_F Marking as target human body surface control point cloud P C_S ；
Step 2.7, feature points FEM of the transformed bone finite element model B_F Marking as a target human skeletal control point cloud P C_B ；
Step 2.8, adopting RBF-TPS function and using body surface finite element model characteristic point FEM S_F As source control point, target human body surface control point cloud P C_S Carrying out grid transformation for the target control point to obtain human body surface finite element models with different physical signs and different postures, and marking as FEM S_T ；
FEM (finite element model) for body surface finite element model by adopting RBF-TPS (radial basis function-TPS) function S Transforming to generate finite element models FEM of human body surface under different physical signs and different postures S_T Then, the following equation should be satisfied:
wherein f is s (x, y, z) represents a finite element model FEM of the body surface S Finite element model FEM of human body surface under different physical signs and different postures S_T Transforming a function based on feature points FEM of finite element model of body surface S_F Controlling point cloud P with target human body surface C_S The conversion relationship between them;
step 2.9, adopting RBF-TPS function and using FEM as characteristic point of skeleton finite element model B_F As source control point, target human skeleton control point cloud P C_B Carrying out mesh transformation for the target control point to obtain human skeleton finite element models under different physical signs, and marking as FEM B_T ；
Specifically, a bone finite element model FEM is subjected to RBF-TPS function B Transforming to generate finite element model FEM of human skeleton under different physical signs B_T Then, the following equation should be satisfied:
wherein, f B (x, y, z) represents a boneFinite element model FEM B Finite element model FEM of human skeleton under different physical signs B_T Transformation function based on feature points FEM of finite element model of bone B_F Control point cloud P with target human skeleton C_B A conversion relationship therebetween;
4. a finite element model generation method of human body as claimed in claim 1, wherein:
step 3.1, transforming the human body surface finite element model FEM with the same physical sign parameters S_T Finite element model FEM with skeleton B_T Carrying out pairing;
step 3.2, based on the transformed finite element model FEM of human body S_T Calculating the position of a joint point, wherein in the step, a corresponding SMPL human body model is inquired according to the human body finite element model, and the joint point is calculated according to the following formula:
wherein the content of the first and second substances,
for the transformation matrix, K is the number of human joint points, N is the number of SMPL human model vertexes,
for SMPL mannequin vertices shaped based on shape blending,
calculating the position of the joint point;
step 3.3, positioning the skeleton according to the joint point position, wherein in the step, the transformed finite element model FEM of the human body surface S_T The spatial position, shape, body posture, etc. are exactly the same as the corresponding SMPL mannequin, so their joint positions are also the same, corresponding the SMPL mannequin joint obtained in step 3.2 to the personFinite element model FEM of body surface S_T Then, the transformed bone finite element model FEM is used B_T Positioning according to the position of the joint point;
step 3.4, obtaining internal organs of the human body including the heart, the kidney and the lung by zooming through finite element models of thoracic bones and pelvic bones at the periphery of the organs;
step 3.5, inhibiting the generation of the soft trunk tissues by internal organs and body surface geometry;
and 3.6, adding muscle linear units, wherein the muscle linear units are attached to corresponding grid nodes of the human body finite element model and automatically change along with the grid changing process to finish the generation of the target human body finite element model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211159539.2A CN115640714A (en) | 2022-09-22 | 2022-09-22 | Method for generating finite element model of human body |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211159539.2A CN115640714A (en) | 2022-09-22 | 2022-09-22 | Method for generating finite element model of human body |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115640714A true CN115640714A (en) | 2023-01-24 |
Family
ID=84941738
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211159539.2A Pending CN115640714A (en) | 2022-09-22 | 2022-09-22 | Method for generating finite element model of human body |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115640714A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117725706A (en) * | 2024-02-08 | 2024-03-19 | 中国汽车技术研究中心有限公司 | Tibia shape prediction method, device, equipment and storage medium |
| CN117725706B (en) * | 2024-02-08 | 2024-05-31 | 中国汽车技术研究中心有限公司 | Tibia shape prediction method, device, equipment and storage medium |
-
2022
- 2022-09-22 CN CN202211159539.2A patent/CN115640714A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117725706A (en) * | 2024-02-08 | 2024-03-19 | 中国汽车技术研究中心有限公司 | Tibia shape prediction method, device, equipment and storage medium |
| CN117725706B (en) * | 2024-02-08 | 2024-05-31 | 中国汽车技术研究中心有限公司 | Tibia shape prediction method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110599528B (en) | Unsupervised three-dimensional medical image registration method and system based on neural network | |
| CN110570515B (en) | Method for three-dimensional modeling of human skeleton by using CT image | |
| JP4272621B2 (en) | Skeletal fitting based on human body measurements | |
| Sigal et al. | Mesh-morphing algorithms for specimen-specific finite element modeling | |
| CN106920228B (en) | The method for registering and device of brain map and brain image | |
| Dong et al. | An anatomy-based approach to human muscle modeling and deformation | |
| CN1864074B (en) | Method for determining patient-related information, control apparatus and magnetic resonance tomography instrument | |
| Qin et al. | A novel modeling framework for multilayered soft tissue deformation in virtual orthopedic surgery | |
| KR20230004837A (en) | Generative nonlinear human shape model | |
| de Buhan et al. | A facial reconstruction method based on new mesh deformation techniques | |
| Shui et al. | A computerized craniofacial reconstruction method for an unidentified skull based on statistical shape models | |
| CN111260702B (en) | Laser three-dimensional point cloud and CT three-dimensional point cloud registration method | |
| Tajdari et al. | Next-generation prognosis framework for pediatric spinal deformities using bio-informed deep learning networks | |
| US11798232B2 (en) | Method and system for generating a new anatomy | |
| CN108898568A (en) | Image composition method and device | |
| CN115116586A (en) | Deformable statistical atlas construction method based on joint registration | |
| CN102034272A (en) | Generating method of individualized maxillofacial soft tissue hexahedral mesh | |
| Fotsin et al. | Shape, pose and density statistical model for 3D reconstruction of articulated structures from X-ray images | |
| CN115640714A (en) | Method for generating finite element model of human body | |
| CN112562070A (en) | Craniosynostosis operation cutting coordinate generation system based on template matching | |
| CN111105502A (en) | Biological rib nose and lower jaw simulation plastic technology based on artificial bone repair material | |
| CN114511561B (en) | Spine sagittal position form prediction method, device, equipment and storage medium | |
| Subsol et al. | Automatic 3D facial reconstruction by feature-based registration of a reference head | |
| CN110322491B (en) | Algorithm for registering deformable mouse whole-body atlas and mouse image | |
| Cheng et al. | Facial morphology prediction after complete denture restoration based on principal component analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |