Clothing simulation collision processing method based on SDF and triangle representation

Abstract

The invention discloses a clothing simulation collision processing method based on SDF and triangle representation, and belongs to the technical field of computer graphics, and the method comprises the steps: S1, inputting human body model parameters; s2, establishing a human body SDF through the MLP; s3, inputting a spatial position x of a clothing vertex, and obtaining penetration information of the clothing vertex; s4, estimating the penetration information of the triangle according to the penetration information of the top point of the garment; s5, predicting a displacement coefficient of the triangle according to the penetration information of the triangle; s6, mapping the predicted displacement coefficient of the triangle to the top point of the garment to obtain the displacement coefficient of the top point of the garment; s7, according to the displacement coefficient, the SDF value and the gradient, moving the clothing vertex to realize penetration processing; according to the clothing simulation collision processing method based on the SDF and the triangle representation, the human body SDF is used as a discrimination tool of clothing penetration conditions, and geometric elements with penetration can be quickly checked out. Displacement of geometric elements is efficiently and accurately solved through a neural network, so that penetration processing is completed.

Description

Technical Field

[0001] The present invention relates to the technical field of computer graphics, and in particular to a method for handling clothing simulation collisions based on SDF and triangle representation. Background Art

[0002] As the main body in scenarios such as the metaverse and virtual reality, the authenticity of digital humans plays an important role in enhancing the user experience. Clothing is an especially important aspect of enhancing the authenticity of digital humans: penetration of clothing with the human body or the scene will significantly reduce the authenticity of digital humans.

[0003] For the first prior art, reference can be made to the Chinese patent with the publication number CN116778092A, which discloses a high-fidelity virtual fitting method and device, using SDF to model the human body and output the volume density of spatial positions. This volume density will be used to optimize and obtain a nude human model. The human model is imported into clothing simulation software to complete clothing simulation; and the method and system for reconstructing a dynamic loose clothing human body based on monocular video with the publication number CN117557762A, which estimates the explicit and implicit representations of the human body in the canonical space - where the implicit representation takes the form of SDF - and trains a dynamic deformation field and a skinning deformation field to estimate the non-rigid deformation of the moving human body and the skinning weights at any position in space during the movement. Through the LBS technology, the position of the spatial point in the canonical space after transformation is obtained. However, the main purpose of using SDF in the above prior art solutions is to model the human body in a standard posture, and combine parametric human models and technologies such as LBS to deform the human body to obtain the human body in a motion posture. However, the penetration problem that may occur when the moving human body drives the clothing to move has received less attention. The first prior art may rely on physical simulation to solve the penetration problem. Such methods have the problem of high time complexity while having high accuracy, and may experience solution collapse under certain conditions, resulting in an inability to generate a reasonable solution. Another prior art uses a unified deformation field and skinning weight field to deform the human body and clothing simultaneously, and such solutions cannot guarantee that the clothing and the human body do not penetrate. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for handling clothing simulation collisions based on SDF and triangle representation to solve the problems in the above background art.

[0005] To achieve the above purpose, the present invention provides a method for handling clothing simulation collisions based on SDF and triangle representation, including the following steps:

[0006] Step S1, input human model parameters;

[0007] Step S2: Establish the Signed Distance Field (SDF) of the human body through a Multi-Layer Perceptron (MLP).

[0008] Step S3: Input the spatial position x of the clothing vertices to obtain the penetration information of the clothing vertices, including the SDF value s and the gradient.

[0009] Step S4: Estimate the penetration information of the triangle based on the penetration information of the clothing vertices.

[0010] Step S5: Predict the displacement coefficient of the triangle according to the penetration information of the triangle.

[0011] Step S6: Map the predicted displacement coefficient of the triangle to the clothing vertices to obtain the displacement coefficient of the clothing vertices.

[0012] Step S7: Move the clothing vertices according to the displacement coefficient, SDF value, and gradient to achieve penetration processing.

[0013] Preferably, in Step S3, the human body model parameters are divided into body shape parameters β and pose parameters θ. The process of establishing the human body Signed Distance Field (SDF) using MLP is expressed as:

[0014] s = γ(β, θ, x);

[0015] where s is the SDF value of the spatial position x, and γ is the MLP. Since the SDF is inferred through a neural network, the gradient of the SDF at x can be obtained through the automatic differentiation module of the deep learning framework (such as the autograd module of PyTorch). For the above formula, we only need to use the automatic differentiation module to obtain the gradient of the output s of the MLP with respect to the input vector {β, θ, x}, and take the partial gradient with respect to x. In PyTorch, this can be achieved through the grad() function.

[0016] Preferably, Step S4 is specifically calculated as follows:

[0017]

[0018] where, is the approximate penetration depth of the triangle, is the approximate direction vector from the triangle to the human body surface, i is the index of the clothing vertex, V represents the set of vertices belonging to the triangle, and Avg(·) represents the mean operation.

[0019] Preferably, in Step S5, after obtaining the penetration information of the triangle, a MLP η is used to predict the displacement coefficient of the triangle, specifically:

[0020]

[0021] Displacement coefficient α t means: Move each triangle along the direction vector approximated to the human body surface by the distance, and the penetrated triangle can be completely moved out of the human body.

[0022] Preferably, the specific calculation in step S6 is as follows:

[0023]

[0024] where T i is the set of triangles around vertex i, k is the serial number of a triangle in the set, is the displacement coefficient of the k-th triangle.

[0025] Preferably, the displacement coefficient of the vertex in step S7 is used to calculate the final movement distance of the vertex. By moving the vertex along the gradient direction of the SDF by the final movement distance, the penetration processing is realized, which is expressed as:

[0026]

[0027] where is the position of vertex i after correction, and |·| represents the operation of taking the modulus of the vector.

[0028] Therefore, the present invention adopts the above-mentioned clothing simulation collision processing method based on SDF and triangle representation, and has the following beneficial effects:

[0029] (1) Infer the displacement of the triangle in the triangle domain, and its semantics are: to completely solve the penetration and avoid damaging the clothing details generated by the clothing simulation method as much as possible, the displacement required for each triangle solves the problem that the existing methods for collision processing in the point domain generally have limitations;

[0030] (2) The pipeline used is simple, only including two sets of operations of "neural network propagation + mapping + taking the average", which can efficiently complete the collision processing and will not introduce a significant time burden on the clothing simulation process;

[0031] (3) It has a wide range of applications and has few requirements for the underlying clothing simulation method. Theoretically, this method can be applied to most clothing simulation methods that generate clothing with fixed topology. This property greatly improves the practical value of this method.

[0032] Next, through the drawings and embodiments, the technical solutions of the present invention will be further described in detail. Description of the Drawings

[0033] Figure 1This is a flowchart of a method for clothing simulation collision handling based on SDF and triangle representation in the present invention. Specific embodiments

[0034] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0035] Please refer to Figure 1 , a method for clothing simulation collision handling based on SDF and triangle representation, including the following steps:

[0036] Step S1: Input human body model parameters;

[0037] Step S2: Establish a human signed distance field SDF through a multi-layer perceptron (MLP). The MLP can receive human parameters, spatial positions, and optional clothing parameters of mainstream parameterized human body models as inputs and output the SDF value at that position.

[0038] Step S3: Input the spatial position x of the clothing vertices to obtain the penetration information of the clothing vertices, including the SDF value s and the gradient

[0039] Taking the parameterized human body model SMPL as an example. The human parameters of SMPL can be divided into body shape parameters β and pose parameters θ. The process of establishing the human signed distance field SDF using MLP is expressed as:

[0040] s = γ(β, θ, x);

[0041] where s is the SDF value at the spatial position x, and γ is the MLP; since the SDF is inferred through a neural network, the gradient of the SDF at x can be obtained through the automatic differentiation module of the deep learning framework (such as the autograd module of PyTorch). For the above formula, we only need to use the automatic differentiation module to obtain the gradient of the output s of the MLP with respect to the input vector {β, θ, x}, and take the partial gradient with respect to x. In PyTorch, this can be achieved through the grad() function.

[0042] The above SDF inference process is for clothing vertices. To achieve penetration processing, the penetration information needs to be mapped to the triangle domain, and the displacement information is inferred in the triangle domain. The mapping of point-triangle domain is essentially an array data structure in the computer. Each vertex and triangle on the clothing has a unique number. Through mapping, several triangles surrounding a vertex and the three vertices of a triangle can be quickly found.

[0043] Step S4: Estimate the penetration information of the triangle based on the penetration information of the clothing vertices. Specifically, the penetration depth of each triangle is characterized by the average value of the penetration depths of the three vertices; and the direction vector to the human body surface is characterized by the average value of the SDF gradients at the three vertices:

[0044]

[0045] Among them, is the approximate penetration depth of the triangle, is the approximate direction vector from the triangle to the human body surface, i is the index of the clothing vertex, V represents the set of vertices belonging to the triangle, and Avg(·) represents the operation of taking the average value.

[0046] Step S5: Predict the displacement coefficient of the triangle according to the penetration information of the triangle; after obtaining the penetration information of the triangle, the displacement coefficient of the triangle is predicted through an MLP η, specifically:

[0047]

[0048] The displacement coefficient α t means that each triangle is moved along the approximate direction vector to the human body surface by distance, and the penetrated triangle can be completely moved out of the human body.

[0049] Step S6: Map the predicted displacement coefficient of the triangle to the clothing vertex domain and perform an averaging operation to obtain the predicted displacement coefficient of the clothing vertex. The above process can be expressed as:

[0050]

[0051] Among them, T i is the set of triangles surrounding vertex i, k is the serial number of a triangle in the set, is the displacement coefficient of the k-th triangle.

[0052] Step S7: The displacement coefficient of the vertex is used to calculate the final movement distance of the vertex. By moving the vertex along the gradient direction of the SDF by the final movement distance, a more thorough penetration processing can be achieved, which is expressed as:

[0053]

[0054] Among them, is the position of vertex i after correction, and |·| represents the modulus operation of the vector.

[0055] Therefore, the present invention adopts the above-mentioned method for processing clothing simulation collision based on SDF and triangle representation, uses the human body SDF as a discriminant tool for clothing penetration, and quickly checks out the geometric elements that penetrate. The displacement of the geometric elements is efficiently and accurately solved through a neural network to complete the penetration processing.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A clothing simulation collision processing method based on SDF and triangle representation, characterized in that: The following steps are involved: Step S1, input human body model parameters; Step S2, establishing a human body signed distance field SDF through a multi-layer perceptron MLP; Step S3: Input the spatial position x of the clothing vertex to obtain the penetration information of the clothing vertex, including the SDF value s and gradient Step S4, estimating the penetration information of the triangle according to the penetration information of the clothing vertices; Step S5: predicting the displacement coefficient of the triangle according to the penetration information of the triangle; Step S6, mapping the predicted displacement coefficients of the triangles to the vertices of the clothing to obtain the displacement coefficients of the vertices of the clothing; Step S7: Move the clothing vertices according to the displacement coefficient, SDF value and gradient to achieve penetration processing.

2. The clothing simulation collision processing method based on SDF and triangle representation according to claim 1 is characterized in that: In step S3, the human body model parameters are divided into body shape parameters β and posture parameters θ. The process of establishing the human body signed distance field SDF using MLP is expressed as: s = γ(β,θ,x); Where s is the SDF value at spatial position x, and γ is the MLP; since the SDF is inferred through a neural network, the gradient of the SDF at x is It is obtained through the automatic differentiation module of the deep learning framework.

3. The clothing simulation collision processing method based on SDF and triangle representation according to claim 2 is characterized in that: The specific calculation of step S4 is as follows: in, is the approximate penetration depth of the triangle, is the approximate direction vector from the triangle to the human body surface, i is the index of the clothing vertex, V represents the set of vertices belonging to the triangle, and Avg(·) represents the averaging operation.

4. The clothing simulation collision processing method based on SDF and triangle representation according to claim 3 is characterized in that: In step S5, after obtaining the penetration information of the triangle, the displacement coefficient of the triangle is predicted by an MLPη, specifically: Displacement coefficient α t The meaning is: move each triangle along the direction vector approximating the human body surface move distance, move the penetrating triangle out of the human body.

5. The clothing simulation collision processing method based on SDF and triangle representation according to claim 4 is characterized in that: The specific calculation in step S6 is: Among them, T i is the set of triangles around vertex i, k is the index of a triangle in the set, is the displacement coefficient of the kth triangle.

6. The clothing simulation collision processing method based on SDF and triangle representation according to claim 5 is characterized in that: The displacement coefficient of the vertex in step S7 is used to calculate the final moving distance of the vertex. The penetration processing is achieved by moving the vertex along the gradient direction of the SDF by the final moving distance, which is expressed as: in, is the corrected position of vertex i, and |·| represents the modulus operation of the vector.

Images