Virtual image expression generation method and system based on real feeling technology

Abstract

The invention relates to a virtual image expression generation method and system based on a real feeling technology. The method comprises the steps of collecting facial expression data in a multi-angle mode, and performing triangulation and facial region marking through the collected facial expression data; component analysis is carried out on the facial region, main features of facial expression changes are extracted, and a facial model is established through weighted fusion; establishing a mapping relation between the expression features and the facial displacement, and calculating a geometric constraint function of the facial displacement; correspondingly generating different expression models according to the geometric constraint function, optimizing details of a specific area and realizing dynamic face rendering on the virtual image; by introducing multi-view data acquisition, a multi-level feature extraction and fusion strategy, expression displacement regression analysis, geometric constraint and detail optimization, the authenticity and personalized effect of virtual expression generation are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual image expression processing, and in particular to a virtual image expression generation method and system based on real sensing technology. Background Art

[0002] With the development and popularization of 3D motion capture and digital virtual human technology, its application has expanded to the fields of film, games and social networking. However, despite technological advances, the cost and complexity of full-body motion capture still limit its widespread adoption, especially in scenes that pursue a high degree of realism. Currently, many digital humans on the market are still cartoon-like images, and they visually lack depth and details compared to real humans. The production of high-quality digital humans, especially those based on precise motion capture, still requires expensive equipment and professional technical support, which increases the cost and technical difficulty of creation; Chinese patent publication number CN202010699705.2 discloses a method and device for generating virtual expressions, which includes: performing key point detection on a user image to obtain facial key points; determining initial facial expression information and head posture information based on facial key points and preset expression bases; and driving an animation model to make virtual expressions based on user images, facial key points, initial facial expression information, and head posture information. It is used to improve the accuracy of real-time facial expression motion capture technology, thereby reducing the difference between virtual expressions and user facial expressions. It relies on image acquisition methods with a single perspective or a small number of perspectives, and only obtains facial expression information through key point detection. It is difficult to fully obtain subtle expression changes and overall facial information, and it is difficult to fully display corresponding expression features in virtual characters. Therefore, this application proposes a method and system for generating virtual image expressions based on real-sensing technology. Summary of the invention

[0003] In view of the technical problems existing in the prior art, the present invention provides a method and system for generating virtual image expressions based on real-sense technology. By using the facial capture technology based on the Intel RealSense depth camera, a 3D virtual human solution combined with a facial restoration algorithm is proposed. This technology can realistically simulate human facial expressions and accurately restore human facial features to promote the development of virtual social interaction. Our vision is to create a digital human that is as real as if it were real, with a strong sense of presence, as if it were right next to you. This digital human can interact with humans using a facial restoration algorithm, and its authenticity is difficult to distinguish from a real person. The focus of research and development is to enhance the personalized expression of digital humans, such as accurately reproducing the subtle changes in facial wrinkles and expressions, and capturing the unique characteristics of each person; in terms of natural interaction, it is also committed to optimizing the eye contact and head movement of digital humans to achieve more natural user interaction.

[0004] The technical solution of the present invention to solve the above technical problems is as follows: a method and system for generating virtual image expressions based on real-feeling technology; the method comprises the following steps: S1: Collect facial expression data from multiple angles, triangulate the collected facial expression data and mark the facial regions; S2: Perform component analysis on the facial region, extract the main features of facial expression changes, and establish a facial model through weighted fusion; S3: Establish the mapping relationship between expression features and facial displacement, and calculate the geometric constraint function of facial displacement; S4: Generate different expression models according to the corresponding geometric constraint functions, optimize the details of specific areas and realize dynamic facial rendering on the virtual image.

[0005] Furthermore, the virtual image expression generation method based on real-sense technology includes: capturing real-person facial expression data through multi-angle distributed cameras: Fuse 3D depth data from multiple angles into a 3D point cloud model: ; in, is the depth data obtained by the i-th camera, is the weight.

[0006] Furthermore, the virtual image expression generation method based on real-sense technology performs triangulation on the 3D point cloud model and marks key areas: ;in are the triangulation vertices in the point cloud, defining the triangulation mesh in the 3D point cloud model Furthermore, the method for generating virtual image expressions based on real-sense technology includes: performing component analysis on the facial area and extracting the main features of facial expression changes, including: As input, the feature vector is extracted through PCA : ; in, is the triangulated mesh of the face, is the PCA feature matrix.

[0007] Furthermore, the virtual image expression generation method based on real-sense technology obtains a stable facial model by fusing the extracted feature vectors: ; in, are the weights of different perspectives.

[0008] Furthermore, the virtual image expression generation method based on real-sense technology establishes a mapping relationship between expression features and facial displacement to simulate expression deformation and establish a regression model between expression features and facial key point displacement: ; Where Y represents the facial displacement, is the bias term, is the regression coefficient, represents the jth expression feature vector in the facial model, is the error term.

[0009] Furthermore, in the virtual image expression generation method based on real sensing technology, the geometric constraint function for calculating the facial displacement includes: ; in, Indicates the current point deformation position, is the output of the expression displacement model, represents the point position of the i-th expression in the facial model, represents the smoothing parameter, represents pairs of adjacent points in the grid, is the smoothing term.

[0010] Furthermore, the method for generating virtual image expressions based on real-sense technology includes generating different expression models according to geometric constraint functions: ;in, is a displacement model of different expressions, is the bias term, is the regression coefficient of expression, The jth feature in the geometric model.

[0011] A virtual image expression generation system based on real-sense technology, applied to any one of the virtual image expression generation methods based on real-sense technology, the system comprising: Acquisition and segmentation unit: collects facial expression data from multiple angles, performs triangulation on the collected facial expression data, and marks facial regions; Analysis and extraction unit: Perform component analysis on the facial area, extract the main features of facial expression changes, and establish a facial model through weighted fusion; Mapping calculation unit: establishes the mapping relationship between expression features and facial displacement, and calculates the geometric constraint function of facial displacement; Generate rendering unit: Generate different expression models according to the corresponding geometric constraint functions, optimize the details of specific areas and realize dynamic facial rendering on the virtual image.

[0012] The beneficial effects of the present invention are: Through multi-view data collection and weighted fusion methods, the generated facial model is more stable and reduces the impact of occlusion. Compared with the existing single-view data collection method, this solution significantly improves the integrity and quality of the data and helps to accurately restore complex facial expressions; The geometric constraints and detail optimization steps can ensure smooth and natural transitions of facial deformation, especially in the detail processing of key areas (such as lips, eyes, etc.), achieving more realistic expression effects. Compared with traditional methods, this solution can avoid distortion caused by deformation and make the generated expressions more natural. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A schematic diagram of a flow chart of a method for generating virtual image expressions based on real-sense technology according to the present invention; Figure 2 The present invention is a schematic diagram of a virtual image expression generation system based on real sensing technology. DETAILED DESCRIPTION

[0014] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0015] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, "plurality" means two or more, unless otherwise clearly and specifically defined.

[0016] In the description of the present application, the term "for example" is used to mean "used as an example, illustration or description". Any embodiment described as "for example" in the present application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid unnecessary details to obscure the description of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present application.

[0017] like Figure 1 As shown: In one embodiment, a method and system for generating virtual image expressions based on real-sense technology, the method comprises the following steps: S1: Collect facial expression data from multiple angles, triangulate the collected facial expression data and mark the facial regions; S2: Perform component analysis on the facial region, extract the main features of facial expression changes, and establish a facial model through weighted fusion; S3: Establish the mapping relationship between expression features and facial displacement, and calculate the geometric constraint function of facial displacement; S4: Generate different expression models according to the corresponding geometric constraint functions, optimize the details of specific areas and realize dynamic facial rendering on the virtual image.

[0018] In one embodiment, S1 includes: capturing real-person facial expression data through cameras deployed in a distributed manner at multiple angles: Fuse 3D depth data from multiple angles into a 3D point cloud model: ; in, is the depth data obtained by the i-th camera, is the weight.

[0019] In this embodiment, multiple Intel RealSense cameras are distributed at different angles on the face to ensure that the cameras can capture facial features from various perspectives, and the cameras are calibrated to ensure that their relative positions and angles are accurate, thereby avoiding errors during data fusion; the calibration can be completed through a calibration plate or a preset calibration algorithm.

[0020] By synchronizing multiple cameras to start collecting data, you can obtain 3D depth data of the face from different perspectives at the same time to avoid changes in expression or position due to time differences. In the Intel RealSense camera SDK, you can use synchronous triggering or external clock synchronization to achieve multi-camera synchronization.

[0021] Depth data collection: The depth data collected by each camera can be represented as a two-dimensional matrix ,in is the image pixel coordinate, value Represents the depth from the camera and converts these depth data into 3D point cloud coordinates The method is as follows: , , ;in, is the optical center of the camera, and is the focal length of the camera; The 3D point cloud data captured by each camera may be in its own coordinate system. In order to merge the data of all cameras into a unified coordinate system, it is necessary to perform coordinate transformation on the data of each camera. Assume that the transformation matrix of each camera is (obtained by the calibration process), the global coordinates of each point can be obtained through the transformation matrix: ; For each pixel position, the depth data of all cameras are fused to reduce noise and enhance the accuracy of the data. Here, the weighted average method can be used, with the camera's viewing angle, distance or data quality as the weight. The fusion formula is as follows: ; in, is the weight of the i-th camera, satisfying , weight It can be set according to the camera's depth data quality, distance or viewing angle. After weighted fusion, the point cloud data obtained is Represents a complete facial 3D model, which can be used for subsequent triangulation and expression feature extraction.

[0022] In one embodiment, triangulation is performed on the 3D point cloud model to mark key areas: ; in are the triangulation vertices in the point cloud, defining the triangulation mesh T in the 3D point cloud model.

[0023] In this embodiment, triangulation is used to organize irregular point cloud data into a structured grid to provide support for subsequent geometric deformation and expression extraction; first, the point cloud data generated in the previous step needs to be Preprocessing, including denoising and smoothing, is performed to ensure the quality of point cloud data.

[0024] The key points of the face are detected in the 3D point cloud data, usually including the eyes, nose, lips and other feature areas. These key points can be marked using a pre-trained facial feature detection model; the key point set marked on the 3D data is , where each Represents the coordinates of key points.

[0025] The point cloud model is triangulated using the Delaunay triangulation algorithm, dividing the entire facial point cloud into several triangular meshes; the characteristic of Delaunay triangulation is that it can avoid generating triangles that are too long or too thin, making the mesh more uniform; for the point cloud collection , the goal of Delaunay triangulation is to find a set of triangles T, so that the minimum angle of the triangle interior angle is as large as possible (avoiding narrow triangles) and meets the following conditions: here, are the triangulation vertices in the point cloud, which satisfy the Delaunay empty circle rule, that is, the circumscribed circle of any triangle does not contain other points.

[0026] Based on the triangle set T obtained by Delaunay triangulation, the triangular mesh of the face is generated; these triangles constitute the basic geometric structure of the face, making the point cloud a continuous surface, which is expressed as follows: .

[0027] For some key point sets such as eyes, nose and lips, mark the corresponding areas in the triangular mesh. For example: Representing the key points of the eye area, we can extract the triangle subset containing these key points , the generated triangular mesh T and the marked key areas (such as etc.) are stored as part of the facial model for subsequent component analysis.

[0028] In one embodiment, the facial region is subjected to component analysis to extract the main features of facial expression changes, including: taking the triangular mesh T generated in the previous step as input, extracting the feature vector by PCA : ; Where T is the triangulated mesh of the face and W is the PCA feature matrix.

[0029] In this embodiment, principal component analysis (PCA) is used to extract the main features of facial expression changes to reduce the dimension of the data and simplify subsequent processing; PCA can extract the key change patterns of expressions from the point cloud data of the triangular mesh and use these patterns to describe complex facial changes instead of directly using the original high-dimensional data.

[0030] The triangular network structure obtained in the previous step , extract the 3D coordinates of each vertex , the 3D coordinates of all vertices are arranged in order to form a high-dimensional feature vector. Assuming there are N vertices, the feature vector of the face can be expressed as: ; Each expression state (such as smiling, frowning, etc.) corresponds to a feature vector X. If there are M different expression samples, a feature matrix can be obtained : ;in, is the feature vector under the i-th expression state.

[0031] Data decentralization: Calculating the feature matrix The mean vector of : ; Decentralize all expression sample vectors, that is, in each sample Subtract the mean vector from , and get the decentralized data matrix: ; Calculate the centralized data matrix The covariance matrix C of is: ; The covariance matrix C describes the changing relationship of facial expression features; the covariance matrix is ​​decomposed to obtain eigenvalues ​​and eigenvectors. Assume that the eigenvalue is (sorted from largest to smallest), the corresponding eigenvector is , select the eigenvectors corresponding to the first k largest traits to form the PCA feature matrix W. These eigenvectors represent the main directions of expression changes. Use the PCA feature matrix to transform the decentralized data Projecting into the principal component space yields the simplified expression feature vector: .

[0032] In one embodiment, a stable facial model is obtained by fusing the extracted feature vectors: ; in, are the weights of different perspectives.

[0033] In this embodiment, expression feature vectors are obtained from different perspectives (cameras). It is assumed that each perspective corresponds to a feature vector, which is a simplified expression feature obtained by PCA in step S3. For example, the feature vector of perspective 1 is , the eigenvector of view 2 is And so on; according to the reliability, clarity, occlusion and other factors of each perspective, weights are assigned to different perspectives. The weight assignment rules can be based on the following criteria: The more positive the perspective (e.g., the perspective directly in front of you), the higher the weight; The less obstructed the view is, the higher the weight is; The higher the data quality (e.g., less noise), the higher the weight; And the weights satisfy the normalization condition, that is, the sum of the weights of all perspectives is 1.

[0034] In one embodiment, a mapping relationship between expression features and facial displacement is established to simulate expression deformation, and a regression model between expression features and facial key point displacement is established: ; Where Y represents the facial displacement, is the bias term, is the regression coefficient, represents the jth expression feature vector in the facial model, is the error term.

[0035] In this embodiment, the samples in the expression dataset are used to fit the regression model, and the regression coefficient is solved by the least squares method. The goal of the least squares method is to minimize the error between the predicted value and the actual displacement, that is, to minimize the following loss function L: ; in, is the actual displacement of the ith sample, is the feature vector of the i-th sample in the j-th expression change.

[0036] By taking the derivative of the loss function L and setting it to 0, the optimal solution of the regression coefficient can be obtained. The trained regression model is used to predict the displacement Y of the facial key points according to the input expression feature vector to simulate facial deformation. The predicted output of the regression model is used as a parameter to control the geometric deformation of the facial triangular mesh, thereby generating dynamic expressions.

[0037] In one embodiment, the geometric constraint function for calculating facial displacement includes: ;in, Indicates the current point deformation position, is the output of the expression displacement model, represents the point position of the i-th expression in the facial model, represents the smoothing parameter, represents pairs of adjacent points in the grid, is the smoothing term.

[0038] In this embodiment, the geometric constraint function consists of two parts: Data item: Constrain each key point To the target location That is, the sum of the expression displacement and the facial model; by minimizing the geometric constraint energy function , get the optimal position of each key point , to ensure the smoothness and naturalness of expression deformation, the minimization process can be achieved through gradient descent method, conjugate gradient method or other optimization algorithms.

[0039] Smoothing item ensures that the relative positions between adjacent key points remain smooth to avoid excessive stretching or shrinkage; smoothing parameter The size of affects the degree of smoothness; if Larger, the model pays more attention to smoothness; if The smaller the value, the more attention the model pays to the consistency with the target displacement.

[0040] Minimize the energy function for each key point The positions are used as the final expression deformation positions and applied to the facial triangle mesh model for rendering; through these geometrically constrained point positions, a natural facial expression deformation effect is generated.

[0041] For example: Suppose there are three key points , and , their target locations are: , and ; The constraints constrain each point to be as close to the target position as possible: ; The smoothing term constrains the distance variation of adjacent points: ; By optimizing the geometric constraint function, the optimal position of each point is finally obtained, so that the entire facial expression deformation conforms to the expression displacement while maintaining a natural smooth effect.

[0042] In one embodiment, generating different expression models according to the geometric constraint function includes: ; in, is a displacement model of different expressions, is the bias term, is the regression coefficient of expression, The jth feature in the geometric model.

[0043] In this embodiment, in order to adapt to the user's personality characteristics, multiple samples of different user expressions are collected, such as smile, anger, surprise, etc. Each sample contains the face model data after geometric constraints and the personalized displacement of the actual expression effect. Assume that there are three expression sample data, and the geometric constraint characteristics and displacement of each sample are: Sample 1: Geometric feature constraints , the displacement is ; Sample 2: Geometric feature constraints , the displacement is ; Sample 3: Geometric feature constraints , the displacement is ; The regression coefficient can be obtained through multi-sample training , use this model to generate user-specific expression effects.

[0044] In a specific embodiment, a smiling expression generation process example is used to illustrate: Assume that 4 Intel RealSense cameras are used to synchronously collect 3D depth data of a smiling expression. The depth data captured by each camera is converted into 3D coordinates, generating the following point cloud data (simplified representation): Camera 1: ; Camera 2: ; Camera 3: ; Camera 4: ; Through the weighted average method (weights are all set to 0.25), a unified point cloud model is generated by fusion: ; Then, the Delaunay triangulation algorithm is used to divide the point cloud data into multiple triangular meshes and mark feature areas such as lips and eyes; Extract the feature vector X of the smiling expression from the triangular mesh and reduce the feature dimension through PCA. Assume that the feature vector is: ; A regression model is established to generate a smiling effect with upturned lips; assuming that when smiling, the displacement of the mouth corner is 0.3 and the displacement of the eye slight change is 0.05; the regression model is as follows: ; Assuming the regression coefficient , , , ,but: The generated smile displacement is 0.949, which controls the geometric deformation of the lips and mouth corners.

[0045] In the geometric constraint function, set the smoothing parameter , smoothing constraints are imposed on the corners of the mouth and eyes. Assume that the adjustment amounts of the corners of the mouth and eyes are: Mouth corner: 0.1; Corner of eye: 0.05; The final deformation is: .

[0046] By applying all geometry and detail adjustments to the rendering of the avatar, a smiling expression effect is generated. The final rendered smiling image achieves the upturned corners of the mouth and changes in eye details, making the avatar's smile more realistic and natural.

[0047] like Figure 2 As shown in one embodiment, a virtual image expression generation system based on real-sense technology, the system includes: Acquisition and triangulation unit: collects facial expression data from multiple angles, triangulates the collected facial expression data and marks the facial regions; specifically: collects facial expression data from multiple angles, generates a 3D point cloud model, and triangulates to mark facial regions (such as eyes, lips, etc.). Suppose we collect smiling expressions and capture the upward movement of the mouth corners and eye changes from multiple angles. The system uses Delaunay triangulation to form a structured grid of these data and define the feature points of each facial region.

[0048] Analysis and extraction unit: Perform component analysis on the facial area, extract the main features of facial expression changes, and build a facial model through weighted fusion; specifically: Use PCA to perform component analysis on the triangulated facial data, extract the main features of the expression, and build a stable facial model through weighted fusion; For smiling expressions, PCA extracts the main change trends of the corners of the mouth, such as the direction and amplitude of the upward movement of the corners of the mouth. Weighted fusion of data from different perspectives ensures the stability of the expression model, making the smiling features complete and coherent.

[0049] Mapping calculation unit: establish the mapping relationship between expression features and facial displacement, and calculate the geometric constraint function of facial displacement; specifically: establish the mapping relationship between expression features and facial displacement, and calculate the geometric constraint function to ensure natural facial deformation. Assume that the corner of the mouth is raised by 0.3 units and the corner of the eye shrinks slightly. The system uses the regression model to calculate the displacement of these feature points, and through the geometric constraint function, the deformation of the corner of the mouth and the corner of the eye is kept smooth without distortion.

[0050] Generate rendering unit: Generate different expression models according to the geometric constraint function, optimize the details of specific areas and realize dynamic facial rendering on the virtual image. Specifically: Generate different expression models according to geometric constraints, optimize the details (such as the curvature of the lips), and finally dynamically render the expression on the virtual image. The system renders the upward corners of the mouth and the subtle changes of the corners of the eyes on the virtual character to generate a natural smile. Through texture mapping and lighting adjustment, the expression looks vivid and realistic.

[0051] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0052] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for generating virtual image expressions based on real-sense technology, characterized in that: The following steps are involved: S1: Collect facial expression data from multiple angles, triangulate the collected facial expression data and mark the facial regions; S2: Perform component analysis on the facial region, extract the main features of facial expression changes, and establish a facial model through weighted fusion; S3: Establish the mapping relationship between expression features and facial displacement, and calculate the geometric constraint function of facial displacement; S4: Generate different expression models according to the corresponding geometric constraint functions, optimize the details of specific areas and realize dynamic facial rendering on the virtual image.

2. The method for generating virtual image expressions based on real-sense technology according to claim 1, characterized in that: S1 includes: capturing real-person facial expression data through distributed cameras deployed at multiple angles: Fuse 3D depth data from multiple angles into a 3D point cloud model: ； in, is the depth data obtained by the i-th camera, is the weight.

3. The method for generating virtual image expressions based on real-sense technology according to claim 2, characterized in that: Perform triangulation on the 3D point cloud model and mark the key areas: ； in are the triangulation vertices in the point cloud, defining the triangulation mesh in the 3D point cloud model .

4. The method for generating virtual image expressions based on real-sense technology according to claim 3, characterized in that: Perform component analysis on the facial area to extract the main features of facial expression changes, including: As input, the feature vector is extracted through PCA : ； in, is the triangulated mesh of the face, is the PCA feature matrix.

5. The method for generating virtual image expressions based on real-sense technology according to claim 4, characterized in that: A stable facial model is obtained by fusing the extracted feature vectors: ； in, are the weights of different perspectives.

6. The method for generating virtual image expressions based on real-sense technology according to claim 5, characterized in that: Establish a mapping relationship between expression features and facial displacement to simulate expression deformation and establish a regression model between expression features and facial key point displacement: ； in, represents the facial displacement, is the bias term, is the regression coefficient, represents the jth expression feature vector in the facial model, is the error term.

7. The method for generating virtual image expressions based on real-sense technology according to claim 1, characterized in that: The geometric constraint functions for calculating facial displacement include: ； in, Indicates the current point deformation position, is the output of the expression displacement model, represents the point position of the i-th expression in the facial model, represents the smoothing parameter, represents pairs of adjacent points in the grid, is the smoothing term.

8. The method for generating virtual image expressions based on real-sense technology according to claim 7, characterized in that: Different expression models are generated according to the geometric constraint functions, including: ； in, is a displacement model of different expressions, is the bias term, is the regression coefficient of expression, The jth feature in the geometric model.

9. A virtual image expression generation system based on real-sense technology, characterized in that: A method for generating virtual image expressions based on real-sense technology as described in any one of claims 1 to 8, wherein the system comprises: Acquisition and segmentation unit: collects facial expression data from multiple angles, performs triangulation on the collected facial expression data, and marks facial regions; Analysis and extraction unit: Perform component analysis on the facial area, extract the main features of facial expression changes, and establish a facial model through weighted fusion; Mapping calculation unit: establishes the mapping relationship between expression features and facial displacement, and calculates the geometric constraint function of facial displacement; Generate rendering unit: Generate different expression models according to the corresponding geometric constraint functions, optimize the details of specific areas and realize dynamic facial rendering on the virtual image.

Images