Method for building three-dimensional human head triangular mesh model based on statistical shape analysis

Abstract

The invention discloses a method for building a three-dimensional human head triangular mesh model based on statistical shape analysis. The method comprises the steps that i, a training set is created; ii, template models are selected; iii, the template models are aligned at head point cloud models respectively to obtain corresponding homogeneous head triangular mesh models; iv, shape transformation parameters are calculated; v, a human head statistical model is built; vi, the three-dimensional human head triangular mesh model is obtained through shape coefficients and the human head statistical model; or iv, the human head statistical model and a regression model are built; v, the three-dimensional human head triangular mesh model is obtained through the human head statistical model and the regression model; or iv, a manifold interpolation method is adopted for the shape transformation parameters to obtain the three-dimensional human head triangular mesh model. A model shape is expressed through the triangle transformation parameters, the fact that parameter space is smooth manifold can be strictly proved with the expression, and the more accurate statistical model can be built through a manifold statistical analysis method.

Description

technical field [0001] The invention relates to a method for establishing a three-dimensional human head triangular mesh model based on statistical shape analysis, and belongs to the fields of computer-aided design and manufacture and ergonomics. Background technique [0002] Today, with the continuous development of computer technology, especially digital geometry processing technology, the synthesis and editing technology of 3D models has been widely used, such as various virtual scenes with a strong sense of reality, gorgeous 3D video animation, computer-aided geometric design, etc. The synthesis and editing of 3D human head models can be widely used in the following aspects: the generation of virtual people can serve for the composition of 3D virtual scenes; in the network fitting, according to the size of the customer's own size, the 3D head that is consistent with the shape of the customer's head can be generated Models, the display of head and face accessories such as...

AI Technical Summary

Problems solved by technology

If this method is used to synthesize and edit the human head, it has the following disadvantages: 1) The experimental results show that the parameter space represented by the coordinates of the human head is also called the shape space, not the entire vector space with a dimension of 3N (N is the network The number of lattice vertices), a more reasonable assumption is to regard the shape space as a nonlinear manifold in a 3N-dimensional vector space, so the new point obtained by linear interpolation may not be on this manifold, that is, it does not correspond to a human head shape of the head, which would produce an unrealistic human head model

2) Since the shape space has a manifold structure, the Euclidean distance cannot be used to accurately measure the distance between two points on the manifold, and the geodesic distance should be used; PCA analysis is a statistical analysis method based on Euclidean distance, so the PCA method is used Statistical models are also inaccurate

3) This method cannot effectively deal with three-dimensional shapes with rich posture changes, such as human bodies and animals, and cannot simulate non-rigid changes in muscles, which largely hinders the universality of the method

The disadvantage of this method is that the degree of freedom of the transformation matrix is ​​9, and the unique solution cannot be obtained due to insufficient constraints. All three types of matrices need to be obtained by solving optimization problems, and the solution process is complicated.

At the same time, using the PCA method for statistical shape modeling still cannot establish an accurate statistical model

Images