Human body rendering method and system based on 3D Gaussian distribution function
Through the human body rendering method based on 3D Gaussian distribution function, a deformable template grid is constructed and dual quaternary skinning technology is used to solve the problem of difficult to simulate human body posture changes and high computing resources in traditional 3D human body rendering technology, and realizes the Efficient and accurate rendering of human body.
Patent Information
- Application Number
- CN202410965494.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-12
AI Technical Summary
Traditional 3D human body rendering technology is difficult to accurately simulate the complex changes of the human body under different postures, and consumes a lot of computing resources, resulting in performance degradation.
Using the human body rendering method based on 3D Gaussian distribution function, the deformable template grid is constructed, the 3D Gaussian distribution function is attached, and the human body mesh deformation caused by bone movement is obtained using dual quaternary skinning technology, and input it into the 2D texture space for Predictions are made and rendered based on the deformation appearance and geometry.
It realizes the flexibility of efficiently simulating the shape and structure of the human body, ensures the smoothness and nature of the animation, improves the rendering accuracy and computing efficiency, and provides a more realistic visual experience.
Smart Images

Figure CN118941692A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of human body rendering, and in particular to a human body rendering method and system based on a 3D Gaussian distribution function. Background Art
[0002] In traditional 3D human body rendering, how to accurately simulate the shape, structure and deformation caused by skeletal movement of the human body is a key challenge. Traditional rendering technology may not be able to accurately capture and present the complex changes of the human body in different postures, especially when dealing with mesh deformation caused by skeletal movement, distortion or unevenness may occur. In addition, traditional rendering technology often consumes a lot of computing resources, including memory, CPU and GPU, when dealing with complex 3D models and fine rendering effects, which may lead to system performance degradation, especially when dealing with large-scale scenes or high-resolution images. Summary of the invention
[0003] In view of this, the present invention proposes a human body rendering method and system based on 3D Gaussian distribution function, which can solve the defects of the prior art that it cannot accurately capture and present the complex changes of the human body in different postures, and requires a large amount of computing resources.
[0004] The technical solution of the present invention is achieved in this way:
[0005] A human body rendering method based on 3D Gaussian distribution function specifically includes:
[0006] Constructing a deformable template grid, wherein the deformable template grid is used to simulate the shape and structure of a human body to obtain a human body grid;
[0007] Attach a 3D Gaussian distribution function to the deformable template mesh;
[0008] Simulate human skeleton movement based on a deformable template mesh with a 3D Gaussian distribution function attached;
[0009] Based on the dual quaternion skinning technology, the human body mesh deformation caused by human skeleton movement is obtained;
[0010] The human body mesh deformation is input into the 2D texture space for prediction, and the appearance and geometric shape of the human body mesh after deformation are predicted;
[0011] The human body is rendered based on the deformed appearance and geometric shape of the human body mesh, thereby achieving human body rendering.
[0012] As a further optional solution of the human body rendering method based on 3D Gaussian distribution function, the constructing of a deformable template grid specifically includes:
[0013] Get a 3D human body model as a base mesh;
[0014] Bind the association between mesh vertices and bones on the base mesh to obtain an initial deformable template mesh;
[0015] Add the expression blend shape to the initial deformable template mesh to get the final deformable template mesh.
[0016] As a further optional solution of the human body rendering method based on 3D Gaussian distribution function, the expression blending shape is added to the initial deformable template mesh to obtain the final deformable template mesh, which specifically includes:
[0017] Use 3D modeling software to create basic expression shapes on the base mesh;
[0018] Create a Black Shape deformer to associate the base mesh with the base expression shape.
[0019] Sets the blending weight for each base expression shape.
[0020] As a further optional solution of the human body rendering method based on 3D Gaussian distribution function, attaching the 3D Gaussian distribution function to the deformable template grid specifically includes:
[0021] Define the mathematical model of the 3D Gaussian distribution function;
[0022] Associating the position of the mathematical model with the vertices of the deformable template mesh to obtain a 3D Gaussian distribution function attached to the deformable template mesh;
[0023] Parameterize a 3D Gaussian distribution function attached to a deformable template mesh.
[0024] As a further optional solution of the human body rendering method based on 3D Gaussian distribution function, the method of obtaining the human body mesh deformation caused by human body skeleton movement based on dual quaternion skinning technology specifically includes:
[0025] Defining a skin for a human body mesh, wherein the skin is used to determine the association between vertices and bones in the human body mesh;
[0026] Obtain skeletal animation data based on human skeletal movement;
[0027] According to the rotation and translation information of the bones, calculate the dual quaternion representation of each bone in each frame;
[0028] Based on the relationship between vertices and bones in the human body mesh and the dual quaternion representation of each bone in each frame, the new position of the vertex is calculated using the dual quaternion skinning technology;
[0029] The positions of the vertices are updated according to the calculated position results to obtain the human body mesh deformation.
[0030] As a further optional solution of the human body rendering method based on 3D Gaussian distribution function, the new position of the vertex is calculated using the dual quaternion skinning technology based on the association relationship between the vertices and bones in the human body mesh and the dual quaternion representation of each bone in each frame, which specifically includes:
[0031] Based on the bone hierarchy, the bind transformation matrix is calculated for each bone;
[0032] For each vertex, use the bind transformation matrix of its associated bone to transform it from model space to the corresponding bone space;
[0033] Calculate the weight of each vertex in bone space;
[0034] Use dual quaternions to represent the rotation and translation of each bone at each frame;
[0035] For each vertex, calculate the new position of the vertex in model space based on its position and weight in bone space and the dual quaternion of each bone.
[0036] As a further optional solution of the human body rendering method based on 3D Gaussian distribution function, the construction of the 2D texture space specifically includes:
[0037] Collect appearance and geometry data of human meshes containing different skeletal motion states as training data;
[0038] Design 2D convolutional network structure;
[0039] The 2D convolutional network structure is trained according to the training data to generate a 2D texture space.
[0040] A human body rendering system based on 3D Gaussian distribution function, comprising:
[0041] A construction module, used to construct a deformable template grid, wherein the deformable template grid is used to simulate the shape and structure of a human body to obtain a human body grid;
[0042] An attachment module for attaching a 3D Gaussian distribution function to a deformable template mesh;
[0043] A simulation module for simulating human skeleton motion based on a deformable template mesh with a 3D Gaussian distribution function attached;
[0044] An acquisition module, used for acquiring the human body mesh deformation caused by human body skeleton movement based on dual quaternion skinning technology;
[0045] A prediction module is used to input the human body mesh deformation into the 2D texture space for prediction, and predict the appearance and geometric shape of the human body mesh after deformation;
[0046] The rendering module is used to render according to the appearance and geometric shape of the human body mesh after deformation, thereby realizing the rendering of the human body.
[0047] A computing device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the processor implements any one of the steps of the above-mentioned human body rendering method based on a 3D Gaussian distribution function.
[0048] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the computer program implements any one of the steps of the above-mentioned human body rendering method based on a 3D Gaussian distribution function.
[0049] The beneficial effects of the present invention are as follows: by constructing a deformable template grid, the technical solution can efficiently simulate the shape and structure of the human body; the deformable template grid has sufficient flexibility and can adapt to human body models of different postures and shapes, providing a basis for subsequent rendering; the dual quaternion skinning technology is used to obtain the human body grid deformation caused by skeletal movement, and the grid deformation caused by skeletal animation can be simulated in real time and accurately, ensuring the fluency and naturalness of human body animation and improving user experience; the 3D Gaussian function is parameterized and attached to the deformable template grid, so that these parameters can be efficiently learned and processed in the 2D texture space, simplifying the management of the Gaussian function parameters, improving the calculation efficiency, and ensuring the rendering accuracy; the human body grid deformation is input into the 2D texture space for prediction, and the appearance and geometric shape of the human body grid after deformation can be accurately predicted; this prediction method takes into account the influence of skeletal movement on the grid deformation, making the rendering result more realistic; at the same time, rendering is performed based on the appearance and geometric shape of the human body grid after deformation, so that high-quality human body rendering effect can be achieved; the rendering result can accurately reflect the deformation and appearance change of the human body under skeletal movement, and provides a more realistic visual experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0051] Figure 1 A schematic diagram of a flow chart of a human body rendering method based on a 3D Gaussian distribution function according to the present invention;
[0052] Figure 2 The figure is a schematic diagram of the composition of a human body rendering system based on a 3D Gaussian distribution function according to the present invention. DETAILED DESCRIPTION
[0053] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0054] refer to Figure 1-2 , a human body rendering method based on 3D Gaussian distribution function, specifically comprising:
[0055] Constructing a deformable template grid, wherein the deformable template grid is used to simulate the shape and structure of a human body to obtain a human body grid;
[0056] Attach a 3D Gaussian distribution function to the deformable template mesh;
[0057] Simulate human skeleton movement based on a deformable template mesh with a 3D Gaussian distribution function attached;
[0058] Based on the dual quaternion skinning technology, the human body mesh deformation caused by human skeleton movement is obtained;
[0059] The human body mesh deformation is input into the 2D texture space for prediction, and the appearance and geometric shape of the human body mesh after deformation are predicted;
[0060] The human body is rendered based on the deformed appearance and geometric shape of the human body mesh, thereby achieving human body rendering.
[0061] In this embodiment, by constructing a deformable template grid, the technical solution can efficiently simulate the shape and structure of the human body. This deformable template grid has sufficient flexibility and can adapt to human body models of different postures and shapes, providing a basis for subsequent rendering. The dual quaternion skinning technology is used to obtain the human body grid deformation caused by skeletal movement, which can simulate the grid deformation caused by skeletal animation in real time and accurately, ensuring the smoothness and naturalness of human body animation and improving user experience. The 3D Gaussian function is parameterized and attached to the deformable template grid, so that these parameters can be efficiently learned and processed in the 2D texture space, simplifying the management of Gaussian function parameters, improving calculation efficiency, and ensuring the rendering accuracy. The human body grid deformation is input into the 2D texture space for prediction, and the appearance and geometric shape of the human body grid after deformation can be accurately predicted. This prediction method takes into account the influence of skeletal movement on grid deformation, making the rendering result more realistic. At the same time, rendering based on the appearance and geometric shape of the human body grid after deformation can achieve high-quality human body rendering effect. The rendering result can accurately reflect the deformation and appearance changes of the human body under skeletal movement, providing a more realistic visual experience.
[0062] Preferably, the step of constructing a deformable template grid specifically includes:
[0063] Get a 3D human body model as a base mesh;
[0064] Bind the association between mesh vertices and bones on the base mesh to obtain an initial deformable template mesh;
[0065] Add the expression blend shape to the initial deformable template mesh to get the final deformable template mesh.
[0066] In this embodiment, using a ready-made 3D human body model as a basic mesh can greatly save the time and cost of modeling from scratch. At the same time, a high-quality 3D human body model can provide rich details and accurate morphology, providing a solid foundation for deformation and animation processing in subsequent steps; by binding the association between mesh vertices and bones, the driving of the mesh by skeletal animation can be realized. This binding relationship enables the mesh vertices associated with it to move accordingly when the bones move, thereby simulating the morphological changes of the human body in motion; expression blend shapes are a technology for simulating changes in facial expressions. By adding expression blend shapes to the initial deformable template mesh, the avatar can show rich facial expressions. This technical effect is of great significance for enhancing the expressiveness and emotional expression ability of the avatar. At the same time, the use of expression blend shapes also makes the production of expression animation more flexible and efficient, because different expressions can be quickly generated by mixing different shapes.
[0067] Preferably, the adding of the expression blending shape to the initial deformable template mesh to obtain the final deformable template mesh specifically includes:
[0068] Use 3D modeling software to create basic expression shapes on the base mesh;
[0069] Create a Black Shape deformer to associate the base mesh with the base expression shape.
[0070] Sets the blending weight for each base expression shape.
[0071] In this embodiment, by creating multiple basic expression shapes (such as happy, angry, surprised, etc.) on the basic mesh, the avatar can show a variety of different facial expressions. This diversity makes the avatar's expression more real and vivid, and can convey emotions more accurately; using the Black Shape deformer to associate the basic mesh with the basic expression shape, a smooth transition between expressions can be achieved. This transition effect is very natural, avoiding the abruptness when switching expressions, and improving the overall expressiveness and credibility of the avatar; by setting the blending weight of each basic expression shape, the blending ratio between different expressions can be adjusted as needed, thereby creating a new expression effect. This flexibility and customizability make the avatar's expression system more adaptable to different application scenarios and user needs; once the basic expression shape is created and the blending weight is set, these resources can be reused. Whether on the same avatar or on other avatars, the desired expression effect can be quickly generated by adjusting the blending weight. This efficiency and reusability greatly improve the production efficiency of expression animation; due to the use of 3D modeling software and Black Shape, the avatar can be more easily generated. Shape deformer, users can edit and expand the avatar's expression system at any time according to their needs. Whether adding new expression shapes or adjusting the blending weights of existing expressions, it can be achieved through simple operations. This easy-to-edit and expandable feature makes the avatar's expression system more flexible and adaptable.
[0072] Preferably, attaching the 3D Gaussian distribution function to the deformable template grid specifically includes:
[0073] Define the mathematical model of the 3D Gaussian distribution function;
[0074] Associating the position of the mathematical model with the vertices of the deformable template mesh to obtain a 3D Gaussian distribution function attached to the deformable template mesh;
[0075] Parameterize a 3D Gaussian distribution function attached to a deformable template mesh.
[0076] In this embodiment, by defining a mathematical model of a 3D Gaussian distribution function, a precise mathematical description can be provided for subsequent steps to ensure the accuracy and reliability of the data; by associating the position of the 3D Gaussian distribution function with the vertices of the deformable template mesh, the precise positioning of the Gaussian distribution function on the mesh can be achieved. This association method enables the Gaussian distribution function to change accordingly with the deformation of the mesh, thereby maintaining consistency with the mesh morphology. In addition, by associating the Gaussian distribution function with the mesh vertices, the entire mesh can be conveniently processed and analyzed in a unified manner; parameterization is an important means to adjust and optimize model performance. By parameterizing the 3D Gaussian distribution function attached to the deformable template mesh, the shape, size, position and other properties of the Gaussian distribution can be easily adjusted to adapt to different application scenarios and requirements. Through parameterization, the Gaussian distribution function can be flexibly controlled and optimized, thereby improving its performance in shape matching, feature extraction and data analysis.
[0077] Preferably, the method of obtaining the human body mesh deformation caused by human body skeleton movement based on the dual quaternion skinning technology specifically includes:
[0078] Defining a skin for a human body mesh, wherein the skin is used to determine the association between vertices and bones in the human body mesh;
[0079] Obtain skeletal animation data based on human skeletal movement;
[0080] According to the rotation and translation information of the bones, calculate the dual quaternion representation of each bone in each frame;
[0081] Based on the relationship between vertices and bones in the human body mesh and the dual quaternion representation of each bone in each frame, the new position of the vertex is calculated using the dual quaternion skinning technology;
[0082] The positions of the vertices are updated according to the calculated position results to obtain the human body mesh deformation.
[0083] In this embodiment, by defining a skin for the human body mesh and determining the association between the vertices and bones in the human body mesh, it can be ensured that the deformation of the mesh is accurate and in accordance with the laws of physics when the bones move, and this association ensures the accuracy of the deformation effect; dual quaternions are used to represent the rotation and translation information of the bones. Compared with the traditional matrix representation method, dual quaternions have less calculation amount and higher calculation efficiency, which makes it possible to obtain results faster when calculating the new position of the vertex in each frame, thereby improving the rendering speed of the animation; the dual quaternion skinning technology can smoothly handle the transition and blending between bones, avoiding the breakage or distortion that may occur in the traditional skinning technology, which enables the human body mesh to maintain a smooth and natural appearance during the deformation process, thereby improving the visual effect of the animation; because the dual quaternion skinning technology is based on the skeletal animation data, the deformation effect of the human body mesh can be flexibly controlled by adjusting the skeletal animation data, and this flexibility makes the technology suitable for animation production with different styles and different needs.
[0084] Preferably, the method of calculating the new position of the vertex using the dual quaternion skinning technique based on the association between the vertices and bones in the human body mesh and the dual quaternion representation of each bone in each frame specifically includes:
[0085] Based on the bone hierarchy, the bind transformation matrix is calculated for each bone;
[0086] For each vertex, use the bind transformation matrix of its associated bone to transform it from model space to the corresponding bone space;
[0087] Calculate the weight of each vertex in bone space;
[0088] Use dual quaternions to represent the rotation and translation of each bone at each frame;
[0089] For each vertex, calculate the new position of the vertex in model space based on its position and weight in bone space and the dual quaternion of each bone.
[0090] In this embodiment, a binding transformation matrix is calculated for each bone according to the hierarchical structure of the bones, thereby ensuring an accurate binding relationship between the bones and their associated mesh vertices. This binding relationship enables the movement of the bones to be accurately transmitted to the mesh vertices, thereby producing an accurate deformation effect; the vertices are transformed from the model space to the corresponding bone space, so that the position change of the vertices is closely related to the movement of the bones. Calculation is performed in the bone space, which can avoid deformation distortion or breakage caused by complex transformations generated by bone movement, and achieve smooth deformation transition; by calculating the weight of each vertex in the bone space, the natural effect of the vertex being affected by multiple bones can be simulated, and this weight distribution method makes the deformation effect more natural and in line with the actual situation of human body movement; using dual quaternions to represent the rotation and translation of the bones not only simplifies the calculation process, but also improves the calculation efficiency. The dual quaternion has a compact representation and efficient operation properties, and can quickly perform interpolation and mixing operations, thereby accelerating the update process of the vertex position; since the dual quaternion skinning technology is based on bone animation data, the deformation effect of the human body mesh can be flexibly controlled by adjusting the bone animation data.
[0091] Preferably, the construction of the 2D texture space specifically includes:
[0092] Collect appearance and geometry data of human meshes containing different skeletal motion states as training data;
[0093] Design 2D convolutional network structure;
[0094] The 2D convolutional network structure is trained according to the training data to generate a 2D texture space.
[0095] In this embodiment, the geometric and appearance data of human avatars with different skeletal motion states are collected. These data may include mesh vertex coordinates, texture coordinates, lighting conditions, colors, textures, etc. The data are preprocessed, such as normalization and enhancement, to improve the generalization ability of the model. A motion-aware 2D CNN structure, this network should be able to handle mesh deformation caused by skeletal movement and predict the appearance and geometry of the deformed Avatar. The network structure can include multiple convolutional layers, pooling layers, activation functions, and fully connected layers. It is possible to consider using residual connections, batch normalization, and other techniques to improve the performance of the model. Add processing of skeletal movement information to the network, such as using skeletal movement data as additional input features, or integrating skeletal movement information at different levels of the network; divide the preprocessed data into training sets, validation sets, and test sets, and define a loss function to measure the difference between the model prediction results and the actual data. The loss function can be designed according to the needs of specific tasks, such as mean square error, cross entropy loss, etc. Select a suitable optimization algorithm, such as gradient descent, Adam, etc., to train the model. During the training process, the gradient is calculated and the network parameters are updated through the back propagation algorithm. The validation set is used to monitor the training process of the model to avoid overfitting or underfitting. The network structure, hyperparameters, etc. can be adjusted according to the performance of the validation set; when designing the network structure, it is necessary to fully consider the impact of skeletal movement on the appearance and geometry of the Avatar to ensure that the network can accurately capture these changes.
[0096] A human body rendering system based on 3D Gaussian distribution function, comprising:
[0097] A construction module, used to construct a deformable template grid, wherein the deformable template grid is used to simulate the shape and structure of a human body to obtain a human body grid;
[0098] An attachment module for attaching a 3D Gaussian distribution function to a deformable template mesh;
[0099] A simulation module for simulating human skeleton motion based on a deformable template mesh with a 3D Gaussian distribution function attached;
[0100] An acquisition module, used for acquiring the human body mesh deformation caused by human body skeleton movement based on dual quaternion skinning technology;
[0101] A prediction module is used to input the human body mesh deformation into the 2D texture space for prediction, and predict the appearance and geometric shape of the human body mesh after deformation;
[0102] The rendering module is used to render according to the appearance and geometric shape of the human body mesh after deformation, thereby realizing the rendering of the human body.
[0103] A computing device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the processor implements any one of the steps of the above-mentioned human body rendering method based on a 3D Gaussian distribution function.
[0104] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the computer program implements any one of the steps of the above-mentioned human body rendering method based on a 3D Gaussian distribution function.
[0105] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A human body rendering method based on 3D Gaussian distribution function, characterized in that: Specifically include: Constructing a deformable template grid, wherein the deformable template grid is used to simulate the shape and structure of a human body to obtain a human body grid; Attach a 3D Gaussian distribution function to the deformable template mesh; Simulate human skeleton movement based on a deformable template mesh with a 3D Gaussian distribution function attached; Based on the dual quaternion skinning technology, the human body mesh deformation caused by human skeleton movement is obtained; The human body mesh deformation is input into the 2D texture space for prediction, and the appearance and geometric shape of the human body mesh after deformation are predicted; The human body is rendered based on the deformed appearance and geometric shape of the human body mesh, thereby achieving human body rendering.
2. A human body rendering method based on 3D Gaussian distribution function according to claim 1, characterized in that: The constructing of the deformable template grid specifically includes: Get a 3D human body model as a base mesh; Bind the association between mesh vertices and bones on the base mesh to obtain an initial deformable template mesh; Add the expression blend shape to the initial deformable template mesh to get the final deformable template mesh.
3. The human body rendering method based on 3D Gaussian distribution function according to claim 2, characterized in that: The step of adding the expression blend shape to the initial deformable template mesh to obtain the final deformable template mesh specifically includes: Use 3D modeling software to create basic expression shapes on the base mesh; Create a Blend Shape deformer to associate the base mesh with the base expression shape. Sets the blending weight for each base expression shape.
4. The human body rendering method based on 3D Gaussian distribution function according to claim 3, characterized in that: The attaching of the 3D Gaussian distribution function to the deformable template grid specifically includes: Define the mathematical model of the 3D Gaussian distribution function; Associating the position of the mathematical model with the vertices of the deformable template mesh to obtain a 3D Gaussian distribution function attached to the deformable template mesh; Parameterize a 3D Gaussian distribution function attached to a deformable template mesh.
5. The human body rendering method based on 3D Gaussian distribution function according to claim 4, characterized in that: The method of obtaining the human body mesh deformation caused by human body skeleton movement based on the dual quaternion skinning technology specifically includes: Defining a skin for a human body mesh, wherein the skin is used to determine the association between vertices and bones in the human body mesh; Obtain skeletal animation data based on human skeletal movement; According to the rotation and translation information of the bones, calculate the dual quaternion representation of each bone in each frame; Based on the relationship between vertices and bones in the human body mesh and the dual quaternion representation of each bone in each frame, the new position of the vertex is calculated using the dual quaternion skinning technology; The positions of the vertices are updated according to the calculated position results to obtain the human body mesh deformation.
6. The human body rendering method based on 3D Gaussian distribution function according to claim 5, characterized in that: The method uses the dual quaternion skinning technology to calculate the new position of the vertex based on the association between the vertices and bones in the human body mesh and the dual quaternion representation of each bone in each frame, specifically including: Based on the bone hierarchy, the bind transformation matrix is calculated for each bone; For each vertex, use the bind transformation matrix of its associated bone to transform it from model space to the corresponding bone space; Calculate the weight of each vertex in bone space; Use dual quaternions to represent the rotation and translation of each bone at each frame; For each vertex, calculate the new position of the vertex in model space based on its position and weight in bone space and the dual quaternion of each bone.
7. The human body rendering method based on 3D Gaussian distribution function according to claim 6, characterized in that: The construction of the 2D texture space specifically includes: Collect appearance and geometry data of human meshes containing different skeletal motion states as training data; Design 2D convolutional network structure; The 2D convolutional network structure is trained according to the training data to generate a 2D texture space.
8. A human body rendering system based on 3D Gaussian distribution function, characterized in that: include: A construction module, used to construct a deformable template grid, wherein the deformable template grid is used to simulate the shape and structure of a human body to obtain a human body grid; An attachment module for attaching a 3D Gaussian distribution function to a deformable template mesh; A simulation module for simulating human skeleton motion based on a deformable template mesh with a 3D Gaussian distribution function attached; An acquisition module, used for acquiring the human body mesh deformation caused by human body skeleton movement based on dual quaternion skinning technology; A prediction module is used to input the human body mesh deformation into the 2D texture space for prediction, and predict the appearance and geometric shape of the human body mesh after deformation; The rendering module is used to render according to the appearance and geometric shape of the human body mesh after deformation, thereby realizing the rendering of the human body.
9. A computing device, characterized in that The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the human body rendering methods based on a 3D Gaussian distribution function in claims 1-7 when executing the computer program.
10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the steps of any one of the human body rendering methods based on 3D Gaussian distribution functions in claims 1-7 are implemented.
Citation Information
Patent Citations
Animating objects using the human body
CN104508709A
Whole body anatomical structure personalized modeling and posture deformation algorithm for human body biomechanical simulation
CN115954105A
Multi-view neural body prediction using implicit differentiatable renderers for facial expressions, body posture morphology, and garment performance capture
CN116134491A
Mapping a geological parameter on an unstructured grid
WO2016083861A1
Cited By
High-precision dynamic digital human generation method
CN120976379A
Digital human facial expression generation method and system based on generative adversarial network
CN121353482A
Digital character animation control method, modeling method, equipment and storage medium
CN121746555A
Digital character animation control method, modeling method, device and storage medium
CN121746555B
