Human body and human face mesh template fitting method based on deformation graph

Abstract

The invention discloses a human body and human face mesh template fitting method based on deformation graph, which can quickly transform the standard human body or the human face template into the scanning data and obtain a good fitting effect without any preparation in advance. The method comprises the following steps that firstly, feature point detection is carried out on the target scanning data, and manual correction is added, and then two sparse registration methods including rigid alignment and deformation image registration are adopted to realize the unification of the standard templateand the scanning data in the space and the transformation of the attitude, and finally non-rigid dense registration is conducted to perform shape fitting and surface detail reduction. According to the method, the deformation image registration is adopted for taking the frame driving technology to carry out preliminary fitting on the human body posture or the facial expression, and a non-rigid dense registration with a corresponding point relation is used for surface detail reduction under the condition that the posture or the expression is basically consistent. The method has the advantages of being simple in advance preparation, good in fitting effect and high in efficiency, and has good popularization and utilization prospect.

Description

technical field [0001] The invention relates to the technical field of computer graphics and three-dimensional grid model reconstruction, in particular to a method for fitting a grid template of a human body and a face based on a deformation graph. Background technique [0002] In recent years, the application of deep learning and data-driven methods has become more and more popular. The algorithm behind it requires a large number of databases to support it. The model data obtained by scanning with Kinect or RGB-D cameras is noisy and there will be a lot of gaps, so it cannot be directly used for construction. The database generally uses a standard human body or face template model to fit the scanned data to obtain a batch of 3D mesh models with the same topological structure but different poses, so that the standard human body and face model can be quickly realized. The method of fitting deformation is increasingly urgent. [0003] Research on 3D model modeling has been de...

AI Technical Summary

Problems solved by technology

Skeleton-driven technology improves the efficiency of pose transformation, but the restoration of the skin surface shape is not ideal; in order to describe the pose and shape of the model at the same time, two well-known parametric models, SCAPE and SMPL, have emerged. A large number of human body and face models are subjected to principal component analysis to construct a parameterized space, and the posture and shape are expressed by high-dimensional vectors. However, this method also requires a large number of databases for support, and it takes a long time to analyze a large number of models. Therefore, The prior preparation required by the method of parameterizing the model is too cumbersome

Images