PARAMETRIC HUMAN BODY MODELING SYSTEM FOR VIRTUAL GARMENT FITTING

Modeling customer-tailored 3D human models that have a realistic body shape is an important problem in computer graphics, especially in the application of virtual garment fitting systems. However, the time, cost, and labor required to generate such models using conventional modeling methods are often high. Moreover, such processes are considerably difficult for common users who lack the knowledge of computer graphics. To solve these problems, we propose a parametric modeling method for the human body shape. In our study, the 3D body scan data of 250 different individuals are gathered, and for each model, 128 feature points are specified to constitute the body shape statistics. The modeling parameters are then extracted by applying a principal component analysis method to the statistics of the feature points. Using these parameters, we generate a new set of feature points that satisfies the given body measures, and we create a surface by interpolating these feature points using a deformation technique based on a radial basis function network. As a result, realistic 3D human models that satisfy the given constraints are obtained. Since our method uses only a few modeling parameters, a user can generate or modify a human model conveniently even if he/she lacks expertise in computer graphics. Further, since the method involves only simple matrix calculations, the computational cost is remarkably reduced.