Three-dimensional face animation generation method and system based on physiological prior

Abstract

The invention provides a three-dimensional face animation generation method and system based on physiological priori, and belongs to the field of computer vision and computer simulation, and the method comprises the steps: S1, obtaining an initial three-dimensional face model based on CT and MRI scanned cross-sectional images; carrying out correction, grid processing and filling on the human face physiological model to obtain a three-dimensional human face physiological model; s2, an unstructured Delaunay algorithm is used for achieving unstructured tetrahedral subdivision on the three-dimensional face physiological model, and a mesh subdivision result of the three-dimensional face physiological model is obtained; s3, carrying out biomechanical modeling on a mesh generation result of the three-dimensional face physiological model; s4, adding contact constraints and boundary conditions among different tissues of the three-dimensional human face physiological model, and constructing a finite element analysis model to simulate facial deformation; and S5, solving and analyzing the finite element model to obtain a three-dimensional face animation. According to the method, the high-quality three-dimensional face animation can be generated based on a small number of samples.

Description

Technical Field

[0001] The invention belongs to the field of computer vision and computer simulation, and in particular relates to a method and system for generating three-dimensional human face animation based on physiological priors. Background Art

[0002] In recent years, with the development of artificial intelligence technology and computer hardware, virtual digital humans have entered the public eye and have shown great application potential in the fields of human-computer interaction, game and film industry, and virtual live broadcast. The modeling and driving of human faces are key tasks in digital human technology. Humans are very sensitive to changes in faces. High-quality facial animation can not only quickly convey information, but also express emotions and psychological states, giving people a sense of face-to-face communication. Low-quality facial animation will make people suspicious or even uncomfortable with the content. This puts forward requirements for the generation quality of facial animation. At present, the research on facial animation generation methods in academia and industry has reached the stage of high-fidelity hyper-realism, requiring each frame of facial animation to have photo-level accuracy and a level of fineness that cannot be distinguished by the human eye.

[0003] According to the different types of facial animation, facial animation can be divided into two categories: two-dimensional facial animation and three-dimensional facial animation. Compared with two-dimensional facial animation, three-dimensional facial animation provides comprehensive depth information and spatial dimensions, can show more realistic and vivid visual effects, accurately simulate the dynamics of muscles and skin, and support complex lighting effects. There are two types of three-dimensional facial animation from the perspective of driving methods: data-driven and physiological prior-driven. Although data-driven methods represented by video or motion capture can achieve highly realistic animation effects, they are costly, highly dependent on equipment, and have a relatively complex processing process. The biggest limitation at this stage is that high-quality open source three-dimensional facial animation data is scarce and cannot support data-driven model training, so overly smooth facial animations will be generated.

[0004] "FaceFormer: Speech-Driven 3D Facial Animation with Transformers" is a paper from CVPR 2022. This paper proposes a deep learning-based 3D facial animation generation method and a data-driven deep learning network model Faceformer. The network model consists of an encoder and a decoder. Based on the FLAME 3D face model, it extracts audio features from the input audio and maps them into 3D face motion, thereby realizing the generation of 3D facial animation. SOTA results have been achieved on multiple datasets. However, due to the difficulty of collection and processing, the scale of the 3D facial animation database used is small, and the facial animation generated by Faceformer is over-smoothed, and the capture and prediction of high-frequency details in the face area are not accurate enough. Therefore, how to achieve the generation of high-quality 3D facial animation based on a small number of samples has become an urgent problem to be solved. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method for generating three-dimensional facial animation based on physiological priors, comprising the following steps:

[0006] Step S1: obtaining an initial three-dimensional human face model based on tomographic images scanned by CT and MRI; simplifying the initial three-dimensional human face model into a three-layer structure of exoskeleton-muscle-soft tissue, and performing correction, mesh processing and filling on the model to obtain a three-dimensional human face physiological model;

[0007] Step S2: using an unstructured Delaunay algorithm to perform unstructured tetrahedral division on the three-dimensional human face physiological model to obtain a mesh division result of the three-dimensional human face physiological model;

[0008] Step S3: performing biomechanical modeling on the mesh generation result of the three-dimensional human face physiological model, defining material properties at each level, using rigid bodies to describe bones, and using a hyperelastic model Mooney-Rivlin to describe muscles and soft tissues;

[0009] Step S4: constructing finite elements to simulate facial deformation, adding contact constraints between different tissues of the three-dimensional human face physiological model to simulate pressure and friction between different tissues, and adding boundary conditions between bones and facial soft tissues to constrain the mutual movement between rigid bodies and soft bodies;

[0010] Step S5: solving and analyzing the finite element to obtain a three-dimensional facial animation.

[0011] Beneficial effects:

[0012] 1. The present invention generates high-quality three-dimensional facial animation in a short time with a small amount of samples, does not rely on massive data sets, and reduces the generation cost of high-quality three-dimensional facial animation.

[0013] 2. The present invention focuses on the movement of internal organs of the human face. The animation conforms to anatomical principles and has guiding significance for computer-assisted language teaching, facial surgery and other fields.

[0014] 3. The animation quality is closely related to the modeling accuracy of bones, muscles and soft tissues. The present invention can be gradually improved with the advancement of modeling and simulation methods, and has good practicality and development prospects in the long run.

[0015] 4. The method of the present invention has strong transferability and can be quickly generated when a new role or a new virtual character is produced. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic flow chart of a method for generating three-dimensional facial animation based on physiological priors according to the present invention;

[0017] Figure 2 This is a schematic diagram of the modeling results of the three-dimensional human face physiological model;

[0018] Figure 3 This is a schematic diagram of the mesh generation results of the 3D human face physiological model;

[0019] Figure 4 It is a schematic diagram of the hinge point;

[0020] Figure 5 It is the flow chart of finite element solution;

[0021] Figure 6 This is a key frame animation diagram of the synthesis result of the vowel / ao / ;

[0022] Figure 7 This is a key frame animation diagram of the synthesis result of the pursed lips action;

[0023] Figure 8 The present invention is a structural block diagram of a three-dimensional facial animation generation system based on physiological priors. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0025] Embodiment 1

[0026] like Figure 1 As shown, a method for generating three-dimensional facial animation based on physiological priors provided by an embodiment of the present invention includes the following steps:

[0027] Step S1: obtaining an initial three-dimensional face model based on the tomographic images of CT and MRI scans; simplifying the initial three-dimensional face model into a three-layer structure of exoskeleton-muscle-soft tissue, and performing correction, mesh processing and filling on the structure to obtain a three-dimensional face physiological model;

[0028] Step S2: using an unstructured Delaunay algorithm to implement unstructured tetrahedron partitioning of the three-dimensional human face physiological model, and obtaining a mesh partitioning result of the three-dimensional human face physiological model;

[0029] Step S3: biomechanical modeling is performed on the mesh generation result of the 3D human face physiological model, and material properties of each layer are defined. Rigid bodies are used to describe bones, and hyperelastic model Mooney-Rivlin is used to describe muscles and soft tissues.

[0030] Step S4: constructing finite elements to simulate facial deformation, adding contact constraints between different tissues in the three-dimensional human face physiological model to simulate pressure and friction between different tissues, and adding boundary conditions between bones and facial soft tissues to constrain the mutual movement between rigid bodies and soft bodies;

[0031] Step S5: Solve and analyze the finite element to obtain a three-dimensional face animation.

[0032] In one embodiment, the above step S1: obtaining an initial three-dimensional face model based on the tomographic images scanned by CT and MRI; simplifying the initial three-dimensional face model into a three-layer structure of exoskeleton-muscle-soft tissue, and performing correction, mesh processing and filling to obtain a three-dimensional face physiological model, specifically includes:

[0033] At present, conventional technology can create an accurate three-dimensional physiological model of the face from the tomographic images of CT and MRI scans through Minics medical image processing software. However, the physiological structure of the face is relatively complex, with muscles, fat, fascia, blood vessels and other organs intertwined. The results of Minics reconstruction often show that the negative Jacobian matrix cannot be solved when the finite element analysis method is used to simulate the animation. Considering the difficulty of analyzing the actual face animation generation, the calculation difficulty of the overly complex face modeling structure increases sharply when using physical methods for analysis. The embodiment of the present invention simplifies the static three-dimensional face model into a three-layer structure from the inside to the outside of the skeleton-muscle-soft tissue. With Minic reconstruction and the shape of the textbook as a reference, it is corrected by carving through professional three-dimensional modeling software Zbrush. The anatomical accuracy of the face model and the rules of the structure are suitable for physical calculation as much as possible, and then the engineering file is imported into Blender for automatic mesh processing. First, the three-dimensional model is automatically deformed and matched to fully adapt the skin, muscles and bones. The mesh model is retopologically optimized using the Instant Mesh and QuadriFlow algorithms. Finally, the gap between the skin and the skeletal muscle model is filled by Boolean operations to generate a soft tissue mesh. Figure 2 A schematic diagram showing the modeling results of the three-dimensional human face physiological model.

[0034] In one embodiment, the above step S2: using the unstructured Delaunay algorithm to implement unstructured tetrahedron partitioning on the three-dimensional human face physiological model to obtain the mesh partitioning result of the three-dimensional human face physiological model specifically includes:

[0035] High-quality tetrahedrons can be obtained by using the empty circle properties and maximum and minimum angle properties of the Delaunay algorithm, and the unstructured subdivision can adapt to the complex face shape. Therefore, the unstructured Delaunay algorithm is used to take into account the mesh shape and the computational complexity of the finite element analysis stage.

[0036] When using physical methods to simulate and analyze an object, a volume mesh is required. The three-dimensional human face physiological model obtained in step S1 is a surface mesh, and its interior is hollow. Therefore, a meshing algorithm is required to implement meshing, and a specified space is decomposed into a set of simple geometric shapes through meshing. In three-dimensional space, the triangular surface mesh is decomposed into a volume mesh model composed of tetrahedral units. Meshing is a key step in building a finite element model and performing finite element analysis. Different types of mesh units and meshing algorithms will significantly affect the scale, results and accuracy of the calculation. Meshing is not only a basic link in the finite element analysis process, but also crucial to ensuring the efficiency of calculation and the accuracy of the results. According to the meshing type, it can be divided into structured meshing and unstructured meshing. Unstructured meshing is suitable for adapting to complex collective structures. The structure of the human face is relatively complex. The present invention uses an unstructured Delaunay algorithm to implement unstructured tetrahedral meshing of the three-dimensional human face model. The Delaunay algorithm can avoid the appearance of low-quality tetrahedrons. Figure 3 This is a schematic diagram of the mesh generation results of the 3D human face physiological model, where the blue line on the surface represents the surface structure of the 3D human face model, and the black line on the section represents the internal tetrahedral mesh distribution of the 3D human face model after unstructured tetrahedral generation using the Delaunay algorithm.

[0037] In one embodiment, the above step S3: performing biomechanical modeling on the mesh generation result of the three-dimensional human face physiological model, defining the material properties of each layer, using rigid body to describe bones, and using hyperelastic Mooney-Rivlin model to describe muscles and soft tissues, specifically includes:

[0038] The stress-strain characteristics of soft tissue and muscle are defined as nonlinear, incompressible and isotropic. Since Mooney-Rivlin material is a hyperelastic material with nonlinear and isotropic stress-strain characteristics, a strain energy function is constructed based on Mooney-Rivlin to describe the stress-strain characteristics:

[0039] ;

[0040] in, represents the strain energy function, and represents the first and second invariants of the strain tensor, and is a model material parameter that can be determined experimentally; It represents the elastic volume ratio; Represents temperature-related parameters; when used to represent incompressible materials, the Poisson's ratio is 0.5. .

[0041] Biomechanical modeling of the three-dimensional human face physiological model is to determine the constitutive equation, that is, to define the material properties of each level of the three-dimensional human face physiological model in step S2. In an embodiment of the present invention, for the task of facial animation synthesis, the skull will not deform unless it is hit hard or in rare circumstances, so it is defined as a rigid body. The biomechanical properties of human face soft tissue, namely muscles and other soft tissues, are very complex. Soft tissue exhibits complex biomechanical properties with its hyperelasticity, mainly nonlinearity, anisotropy, quasi-incompressibility, and also plasticity and viscosity. The present invention simplifies it into nonlinearity, incompressibility and isotropy.

[0042] First is nonlinearity. Under small deformation, muscles and soft tissues show approximately linear elastic behavior, but the slope of strain with increasing stress is not constant. The stress-strain curve is initially flat and then becomes steeper and steeper. Quasi-incompressibility refers to the very small volume change during deformation, which can be almost ignored. The quasi-incompressible condition is achieved by setting the Poisson's ratio of the material to 0.5, which is the ratio of the lateral strain to the longitudinal strain of the material. For a material with a Poisson's ratio of 0.5, it is theoretically completely incompressible, that is, stretching or compression in any direction will not cause a volume change. Anisotropic properties are reflected in its complex internal structure. Due to the combination of cells, fibers and fine structures, soft tissues exhibit different mechanical properties in different parts. In particular, for tissues with rich fiber structures such as muscles, their mechanical behavior and response will be significantly different in local areas, which is crucial for setting local parameters. In contrast, isotropic materials have the same mechanical properties in all directions. Metal is a typical material that behaves isotropically within a certain stress range. The present invention simplifies muscles and soft tissues into isotropic materials.

[0043] Muscles and soft tissues exhibit the material properties of hyperelastic materials, with an elastic potential energy function W, which is a scalar function of the strain tensor, and its derivative with respect to the strain component is the corresponding stress component. When unloading, the strain of the material can be automatically restored, and the relationship between the mechanical response and the deformation does not simply follow Hooke's law but corresponds one to one through the elastic energy function. When subjected to external force and deformed, it can return to its initial state after the external force is removed.

[0044] Mooney-Rivlin material is a hyperelastic material that describes material properties based on three main invariants of the strain tensor. The first invariant is related to the volume change of the material, and the last two invariants are related to the shape change of the material. This model is suitable for describing nonlinear deformation of soft tissues, and the present invention uses it to describe muscles and soft tissues. Mooney-Rivlin materials contain terms with a finite number of coefficients, usually including linear and quadratic terms. The model considers the linear approximation of the material stress-strain curve in the early stage, and simulates the material behavior in the nonlinear region by including higher-order terms.

[0045] In one embodiment, the above step S4: constructing finite elements to simulate facial deformation, adding contact constraints between different tissues in the three-dimensional human face physiological model to simulate pressure and friction between different tissues, and adding boundary conditions between bones and facial soft tissues to constrain the mutual movement between rigid bodies and soft bodies, specifically includes:

[0046] Step S41: setting a rigid body constraint as a boundary condition to constrain the interaction between bones and muscles and soft tissues, and simulating the constraint relationship between soft tissues and their bones;

[0047] In the present invention, facial deformation is simulated using the finite element method. The structural stability of the facial model is mainly provided by the skull. In the model setting, the skull is appropriately constrained. The movement of the mandible is defined as rotation and freedom to simulate the opening and closing of the mouth. Muscles and some soft tissues are fixed to the bones by attachment. Muscle tissues and muscles and adjacent soft tissues contact each other during movement. Contact constraints are added to the model. These constraints will simulate the pressure and friction between different tissues to ensure the correct transmission of force and the authenticity of biomechanical behavior. The experiments of the embodiments of the present invention are carried out in the FEBIO open source finite element solver. FEBIO uses a global Cartesian coordinate system to model and simulate the human face model. The anatomically accurate three-dimensional human face physiological model established in step S1 is meshed in step S2. The meshed model can be directly imported into FEBIO for biomechanical modeling in step S3. The definition of materials is achieved by writing material plug-ins. Skull fixation, that is, in No rotation or displacement is allowed in the coordinate system. The mandible can be displaced in the y-axis and z-axis directions, but not in the x-axis, that is, the mandible has the freedom to rotate along the x-axis. This is achieved by defining the skull and mandible as rigid bodies with an articulated structure. Figure 4 Shown is a schematic diagram of the hinge point, where the blue part represents the mandible, and the circled and arrowed area indicates the location of the hinge point between the mandible and the skull.

[0048] Step S42: Adopt the binding contact method to construct contact constraints for the contact surfaces between muscles and soft tissues and between muscles; the contact surfaces are not allowed to slide or separate relative to each other, ensuring the continuous transmission of force and displacement in the contact area; the binding contact relies on the enhanced Lagrangian method, and the relative displacement and rotation between the contact interfaces are accurately controlled by setting the gap function, the gap function is defined as the distance between the nearest point of the slave surface and the master surface, and it is ensured that it is calculated only once in the entire material reference frame; the gap function considers displacement control in all directions, not just in the normal direction, to ensure that the contact surface is fully fitted in all directions; through the gap function, the virtual work expression of the suture contact is derived, in which the reaction force is combined with the Lagrangian multiplier and the additional penalty factor to implement the constraint of zero gap; the gap function is used to define the distance between the master surface and the slave surface, and the gap function is defined as:

[0049] ;

[0050] in, represents the main surface; represents the deformation function of the main surface; represents the deformation function of the corresponding slave surface; represents the projection of the slave node onto the master surface;

[0051] The work equation for the reaction force of the bonded contact is defined as:

[0052] ;

[0053] in, Represents the contact boundary between the master surface and the slave surface; represents the Lagrange multiplier; Represents the change of constraint function; represents the differential of the boundary.

[0054] In the present invention, the mutual movement between the rigid body and the soft body is constrained for the boundary conditions between the skull and the facial soft tissue. The skull provides support and attachment points for the facial soft tissue, which is achieved through rigid body constraints. Specifically, the interaction between the rigid body (skull) and the soft body (muscles and other soft tissues) is achieved through rigid body constraints, ensuring that the soft tissue can accurately reflect its constraint relationship with the skeleton during the simulation process. The rigid body constraint does not allocate any degrees of freedom at the end of the rigid body, and the motion or static position of the rigid body will directly affect the soft body connected to it. When the soft body receives external forces or stress and deformation caused by bone movement, it can closely follow the motion trajectory of the rigid body. In addition, the contact surface between the rigid body and the soft body is usually realized by point connection elements in the numerical model. These methods allow the rigid body and the soft body to maintain an appropriate relative position when subjected to force, while allowing a certain range of sliding or separation to simulate real physiological movements. The constraints between soft bodies include direct connections between soft tissues and muscles and between muscles. The present invention adopts a tied-elastic method to describe the contact surface between two or more objects. These contact surfaces are not allowed to slide or separate relative to each other. This contact type ensures the continuous transfer of forces and displacements within the contact region. Bound contact relies on an augmented Lagrangian approach to precisely control the relative displacements and rotations between the contact interfaces by setting a gap function. The gap function is defined as the closest point distance from the slave surface to the master surface and is calculated only once in the entire material reference frame. This function considers displacement control in all directions, not just the normal direction, to ensure that the contact surfaces fit perfectly in all directions. From the gap function, an expression for the virtual work of stitched contact can be derived, where the reaction forces are combined with Lagrangian multipliers and additional penalty factors to enforce the constraint of zero gap.

[0055] In one embodiment, the above step S5: solving and analyzing the finite element to obtain the three-dimensional face animation specifically includes:

[0056] The finite element method (FEM) simulates the geometry and load conditions of real physical systems through mathematical approximation. It uses simple elements to connect with each other to approximate infinite real systems with finite unknowns. By simplifying complex problems into simple problems, the solution domain is divided into multiple interconnected small subdomains (finite elements), an approximate solution is assumed for each element, and then the approximate solution of the entire domain is derived. When using finite element analysis, there are three main steps: pre-processing, assembly solution, and post-processing. Pre-processing involves establishing a mathematical model to approximate the actual problem. It mainly includes the following contents:

[0057] 1) Define the geometric area and determine the solution domain / given solid model;

[0058] 2) Select the appropriate unit type and tetrahedron model;

[0059] 3) Specify material properties, such as elastic modulus and density;

[0060] 4) Determine the geometric dimensions of the unit, such as length and area;

[0061] 5) Set the connection mode between units to ensure the consistency of the overall structure;

[0062] 6) Select basis functions for expressing approximate solutions;

[0063] 7) Apply boundary conditions and external loads to simulate actual conditions.

[0064] Figure 5 The finite element solution flow chart is shown.

[0065] The data obtained from the above steps are stored in the Febio platform as XML files. The data of the final assembly solution stage is stored as log files, and the final results are stored in xplt files and can be post-processed and analyzed.

[0066] The following examples illustrate how to use the method proposed in the present invention to synthesize visual animations corresponding to Chinese phonemes on the face and facial action animations. By setting a traction force of a specific direction and magnitude on the muscle surface, the experiment was conducted in FEBIO open source finite element. The experimental environment is shown in Table 1:

[0067] Table 1 Experimental environment configuration table

[0068] Figure 6 This is the key frame animation of the final vowel / ao / , with a total of 12 frames of animation and a total time of 3 minutes.

[0069] Figure 7 This is the key frame animation of the synthesis result of the pursed lips action, with a total of 15 frames of animation and a total time of 5 minutes.

[0070] Embodiment 2

[0071] like Figure 8 As shown, the embodiment of the present invention provides a three-dimensional face animation generation system based on physiological priors, including the following modules:

[0072] A 3D face physiological model building module 61 is used to obtain an initial 3D face model based on the tomographic images scanned by CT and MRI; the initial 3D face model is simplified into a three-layer structure of exoskeleton-muscle-soft tissue, and the three-layer structure is corrected, meshed and filled to obtain a 3D face physiological model;

[0073] A mesh generation module 62 is used to use an unstructured Delaunay algorithm to perform unstructured tetrahedral generation on the three-dimensional human face physiological model to obtain a mesh generation result of the three-dimensional human face physiological model;

[0074] A face biomechanical modeling module 63 is used to perform biomechanical modeling on the mesh generation result of the three-dimensional face physiological model, define the material properties of each layer, use rigid body to describe bones, and use the hyperelastic model Mooney-Rivlin to describe muscles and soft tissues;

[0075] Setting a boundary condition and constraint module 64, for constructing finite elements to simulate facial deformation, adding contact constraints between different tissues in the three-dimensional human face physiological model, for simulating pressure and friction between different tissues, and adding boundary conditions between bones and facial soft tissues, for constraining the mutual movement between rigid bodies and soft bodies;

[0076] The finite element analysis module 65 is used to solve and analyze the finite element to obtain a three-dimensional face animation.

[0077] A three-dimensional facial animation generation device based on physiological priors includes one or more electronic devices, wherein the one or more electronic devices are used to implement a three-dimensional facial animation generation method, system and device based on physiological priors.

[0078] An electronic device includes: one or more processors; a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement a method, system and device for generating three-dimensional facial animation based on physiological priors.

[0079] A computer-readable storage medium stores executable instructions, which, when executed by a processor, enable the processor to implement a method, system and device for generating three-dimensional facial animation based on physiological priors.

[0080] A non-transitory computer-readable storage medium stores a computer program, which, when executed by a processor, implements a method, system and device for generating three-dimensional facial animation based on physiological priors.

[0081] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present invention will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features applied herein.

Claims

1. A method for generating three-dimensional facial animation based on physiological priors, characterized in that: include: Step S1: obtaining an initial three-dimensional face model based on tomographic images scanned by CT and MRI; Simplifying the initial three-dimensional human face model into a three-layer structure of exoskeleton-muscle-soft tissue, and correcting, meshing and filling the three-dimensional human face physiological model; Step S2: using an unstructured Delaunay algorithm to perform unstructured tetrahedral division on the three-dimensional human face physiological model to obtain a mesh division result of the three-dimensional human face physiological model; Step S3: performing biomechanical modeling on the mesh generation result of the three-dimensional human face physiological model, defining material properties at each level, using rigid bodies to describe bones, and using a hyperelastic model Mooney-Rivlin to describe muscles and soft tissues; Step S4: constructing finite elements to simulate facial deformation, adding contact constraints between different tissues in the three-dimensional human face physiological model to simulate pressure and friction between different tissues, and adding boundary conditions between bones and facial soft tissues to constrain the mutual movement between rigid bodies and soft bodies; Step S5: solving and analyzing the finite element to obtain a three-dimensional facial animation.

2. The method for generating three-dimensional facial animation based on physiological priors according to claim 1, characterized in that: The step S2: using an unstructured Delaunay algorithm to implement unstructured tetrahedron partitioning on the three-dimensional human face physiological model to obtain a mesh partitioning result of the three-dimensional human face physiological model, specifically includes: High-quality tetrahedrons can be obtained by using the empty circle properties and maximum and minimum angle properties of the Delaunay algorithm, and the unstructured subdivision can adapt to the complex face shape. Therefore, the unstructured Delaunay algorithm is used to take into account the mesh shape and the computational complexity of the finite element analysis stage.

3. The method for generating three-dimensional facial animation based on physiological priors according to claim 1, characterized in that: The step S3: performing biomechanical modeling on the mesh generation result of the three-dimensional human face physiological model, defining the material properties of each layer, using rigid bodies to describe bones, and using the hyperelastic model Mooney-Rivlin to describe muscles and soft tissues, specifically includes: The stress-strain characteristics of soft tissue and muscle are defined as nonlinear, incompressible and isotropic. Since Mooney-Rivlin material is a hyperelastic material with nonlinear and isotropic stress-strain characteristics, a strain energy function is constructed based on Mooney-Rivlin to describe the stress-strain characteristics: ； in, represents the strain energy function, and represents the first and second invariants of the strain tensor, and is a model material parameter that can be determined experimentally; represents the elastic volume ratio; Represents temperature-related parameters; when used to represent incompressible materials, the Poisson's ratio is 0.

5. .

4. The method for generating three-dimensional facial animation based on physiological priors according to claim 1, characterized in that: The step S4: constructing finite elements to simulate facial deformation, adding contact constraints between different tissues in the three-dimensional human face physiological model to simulate pressure and friction between different tissues, and adding boundary conditions between bones and facial soft tissues to constrain the mutual movement between rigid bodies and soft bodies, specifically includes: Step S41: setting a rigid body constraint as a boundary condition to constrain the interaction between bones and muscles and soft tissues, and simulating the constraint relationship between soft tissues and their bones; Step S42: Adopt the binding contact method to construct contact constraints for the contact surface between muscle and soft tissue and between muscle and muscle; the contact surface does not allow relative sliding or separation, ensuring the continuous transmission of force and displacement in the contact area; the binding contact relies on the enhanced Lagrangian method to accurately control the relative displacement and rotation between the contact interfaces by setting the gap function, the gap function is defined as the closest point distance from the slave surface to the master surface, and ensures that it is only calculated once in the entire material reference frame; the gap function considers displacement control in all directions, not limited to the normal direction, to ensure that the contact surface is fully fitted in all directions; through the gap function, the virtual work expression of the suture contact is derived, in which the reaction force is combined with the Lagrangian multiplier and the additional penalty factor to implement the constraint of zero gap; the gap function is used to define the distance between the master surface and the slave surface, and the gap function is defined as: ； in, represents the main surface; represents the deformation function of the main surface; represents the deformation function of the corresponding slave surface; represents the projection of the slave node onto the master surface; The work equation for the reaction force of the bonded contact is defined as: ； in, Represents the contact boundary between the master surface and the slave surface; represents the Lagrange multiplier; Represents the change of constraint function; represents the differential of the boundary.

5. A three-dimensional face animation generation system based on physiological priors, characterized in that: Includes the following modules: Constructing a 3D face physiological model module for obtaining an initial 3D face model based on CT and MRI scanned tomographic images; simplifying the initial 3D face model into a three-layer structure of exoskeleton-muscle-soft tissue, and performing correction, mesh processing and filling on the structure to obtain a 3D face physiological model; A mesh generation module, used for implementing unstructured tetrahedral generation of the three-dimensional human face physiological model using an unstructured Delaunay algorithm to obtain a mesh generation result of the three-dimensional human face physiological model; A face biomechanical modeling module is used to perform biomechanical modeling on the mesh generation result of the three-dimensional face physiological model, define material properties at each level, use rigid bodies to describe bones, and use a hyperelastic model Mooney-Rivlin to describe muscles and soft tissues; Setting a boundary condition and constraint module for constructing a finite element to simulate facial deformation, adding contact constraints between different tissues of the three-dimensional human face physiological model to simulate pressure and friction between different tissues, and adding boundary conditions between bones and facial soft tissues to constrain mutual movement between rigid bodies and soft bodies; The finite element analysis module is used to solve and analyze the finite element to obtain a three-dimensional face animation.

6. A three-dimensional facial animation generation device based on physiological priors, characterized in that: The method comprises one or more electronic devices, wherein the one or more electronic devices are used to implement the method according to any one of claims 1 to 4.

7. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 4.

8. A computer-readable storage medium, characterized in that: Executable instructions are stored thereon, and when the instructions are executed by a processor, the processor implements the method according to any one of claims 1 to 4.

9. A non-transitory computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images