Search

CN-122023619-A - Artificial intelligence elastic 2D animation generation method driven by continuous anatomical domain

CN122023619ACN 122023619 ACN122023619 ACN 122023619ACN-122023619-A

Abstract

The application discloses a continuous anatomical domain driven artificial intelligence elastic 2D animation generation method, which relates to the field of animation production and comprises the steps of obtaining multi-frame static images containing a target role, constructing a continuous anatomical domain of the target role based on the multi-frame static images, wherein the continuous anatomical domain comprises a density field, an elastic field and a target weight field, the target weight field comprises skeleton weights of all pixel points in the target role, the target weight field is obtained by optimizing an initial weight field generated by a Gaussian kernel function through an energy minimization equation, a driving field of the target role is generated based on the target weight field and a skeleton transformation matrix of the target role, and a deformation field of the target role is generated based on the driving field, the density field and the elastic field.

Inventors

  • ZHENG LIGUO
  • LI JUNYONG
  • ZHANG QIAN
  • ZHENG XINRAN

Assignees

  • 吉林动画学院
  • 吉林吉动盘古网络科技股份有限公司

Dates

Publication Date
20260512
Application Date
20260408

Claims (10)

  1. 1. The artificial intelligence elastic 2D animation generation method driven by the continuous anatomical domain is characterized by comprising an offline preprocessing stage and a real-time operation stage; the off-line preprocessing stage comprises the following steps: acquiring a multi-frame static image containing a target role; constructing a continuous anatomical domain of the target role based on a plurality of frames of the static images, wherein the continuous anatomical domain comprises a density field, an elastic field and a target weight field, the target weight field comprises skeleton weights of each pixel point in the target role, and the target weight field is obtained by optimizing an initial weight field generated by a Gaussian kernel function through an energy minimization equation; The real-time operation phase comprises: Generating a driving field of the target character based on the target weight field and a skeleton transformation matrix of the target character; And generating a deformation field of the target character based on the driving field, the density field and the elastic field, wherein the deformation field comprises the coordinate position of each pixel point in the target character in a next animation frame.
  2. 2. The continuous anatomic domain driven artificial intelligence elastic 2D animation generation method of claim 1, wherein the driving field generation formula is: F(x,y)=∑iW i (x,y)∙M i ∙(x,y,i) T Wherein F (x, y) is the driving field, W i (x, y) is the skeletal weight at pixel point i, M i is the skeletal transformation matrix at pixel point i, and (x, y, i) is the coordinates of pixel point i.
  3. 3. The continuous anatomic domain driven artificial intelligence elastic 2D animation generation method of claim 1, wherein generating a deformation field of the target character based on the drive field, the density field and the elastic field comprises: Estimating the thickness of each pixel point in the target character based on the gradient field of the density field to generate a volume field of the target character; Inputting the driving field, the volume field and the elastic field into a trained machine learning model, and predicting a dynamic elastic adjustment coefficient, a volume correction coefficient and a folding willingness field of the target role through the machine learning model; and generating a deformation field of the target role according to the driving field, the dynamic elastic adjustment coefficient and the gradient field of the folding willingness field.
  4. 4. The continuous anatomic domain driven artificial intelligence elastic 2D animation generation method of claim 3, wherein the deformation field is generated by iterative computation and has a generation formula: Wherein, the And The deformation fields before and after a round of iteration, For the purpose of the driving field, For the dynamic elastic adjustment coefficient, For the gradient field of the folding will field, For the gradient field of the volume field, alpha is, Gamma sum The weight is adjusted.
  5. 5. A continuous anatomic domain driven artificial intelligence resilient 2D animation generation method according to claim 3, characterized in that generating a target animation frame based on the deformation field and the current animation frame of the target character, in particular comprising: and deforming the current animation frame through the deformation field to obtain a candidate animation frame, and correcting the brightness of the candidate animation frame through the volume correction coefficient to obtain the target animation frame.
  6. 6. The continuous anatomic domain driven artificial intelligence elastic 2D animation generation method of claim 3 or 5, further comprising processing a hierarchical occlusion ordering in 2D rendering by a virtual depth field of the target character in generating an elastic 2D animation of the target character.
  7. 7. The continuous anatomic domain driven artificial intelligence resilient 2D animation generation method of claim 6, wherein the virtual depth field is generated from symmetry of the volume field, the folding willingness field and the target character.
  8. 8. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the continuous anatomical domain driven artificial intelligence resilient 2D animation generation method of any of claims 1-7.
  9. 9. A computer readable storage medium having stored thereon a computer program, which when executed by a processor implements the continuous anatomical domain driven artificial intelligence elastic 2D animation generation method of any of claims 1-7.
  10. 10. A computer program product comprising a computer program which, when executed by a processor, implements the continuous anatomical domain driven artificial intelligence resilient 2D animation generation method of any one of claims 1-7.

Description

Artificial intelligence elastic 2D animation generation method driven by continuous anatomical domain Technical Field The application relates to the field of animation production, in particular to a continuous anatomic domain driven artificial intelligent elastic 2D animation generation method. Background 2D animation is widely used in the fields of games, movies, virtual anchor (VTuber), interactive media, and the like. With the increasing demand of users for visual experience, the market is increasingly demanding 2D animation technologies that "have a real physical deformation effect close to 3D" and that "can run in real time on the mobile end". Current 2D character animations rely primarily on skeletal bindings (SKELETALRIGGING) and mesh transformations (MeshDeformation). The skeleton binding splits the 2D image into a plurality of paste blocks and controls the rotation and displacement of each component through a hierarchical skeleton matrix, and the grid deformation constructs a triangular grid on the image and drives grid vertexes to move through skeleton weight or free deformation control points. The above-mentioned conventional methods are based on discrete patches or grids, where joints are prone to folds, breaks or unnatural hard edges, lacking in biological soft tissue continuity. And, skeletal driving based on affine transformation is difficult to express volume changes such as muscle compression, skin stretching, etc., and the character dynamic appears stiff and scribbled. In order to achieve high-quality deformation effect, manual fine adjustment of grid topology, drawing of weight diagrams and setting of deformers are required, and the workload is large and the fine arts experience is relied on. Meanwhile, the 2D plane rotation is difficult to simulate perspective shielding and thickness variation in the process of large-angle surface rotation. Disclosure of Invention The application aims to provide an artificial intelligence elastic 2D animation generation method driven by a continuous anatomical domain, which adopts the continuous anatomical domain to replace a discrete grid so as to eliminate joint faults and simulate soft tissue characteristics. In order to achieve the above object, the present application provides the following solutions: in a first aspect, the present application provides a continuous anatomical domain driven artificial intelligence elastic 2D animation generation method, including an offline preprocessing stage and a real-time operation stage; the off-line preprocessing stage comprises the following steps: acquiring a multi-frame static image containing a target role; constructing a continuous anatomical domain of the target role based on a plurality of frames of the static images, wherein the continuous anatomical domain comprises a density field, an elastic field and a target weight field, the target weight field comprises skeleton weights of each pixel point in the target role, and the target weight field is obtained by optimizing an initial weight field generated by a Gaussian kernel function through an energy minimization equation; The real-time operation phase comprises: Generating a driving field of the target character based on the target weight field and a skeleton transformation matrix of the target character; And generating a deformation field of the target character based on the driving field, the density field and the elastic field, wherein the deformation field comprises the coordinate position of each pixel point in the target character in a next animation frame. In one embodiment, the driving field generation formula is: F(x,y)=∑iWi(x,y)∙Mi∙(x,y,i)T Wherein F (x, y) is the driving field, W i (x, y) is the skeletal weight at pixel point i, M i is the skeletal transformation matrix at pixel point i, and (x, y, i) is the coordinates of pixel point i. In an embodiment, generating the deformation field of the target character based on the driving field, the density field and the elastic field specifically includes: Estimating the thickness of each pixel point in the target character based on the gradient field of the density field to generate a volume field of the target character; Inputting the driving field, the volume field and the elastic field into a trained machine learning model, and predicting a dynamic elastic adjustment coefficient, a volume correction coefficient and a folding willingness field of the target role through the machine learning model; and generating a deformation field of the target role according to the driving field, the dynamic elastic adjustment coefficient and the gradient field of the folding willingness field. In an embodiment, the deformation field is generated by iterative calculation and the generation formula is: Wherein, the AndThe deformation fields before and after a round of iteration,For the purpose of the driving field,For the dynamic elastic adjustment coefficient,For the gradient field of the folding will field,F