Vol.1 No.1 2008

43/76

Research paper : Technologies for the design and retail service of well-fitting eyeglass frames (M. Mochimaru, et al.)−40−Synthesiology - English edition Vol.1 No.1 (2008) study, we developed a computerized recommendation system based on explicit knowledge. Three elemental technologies (easy measurements, size recommendation and style recommendation) were developed based on a database of 3-D face and head shapes. These technologies may be applied to the business through a scenario, as shown in the right side of Figure 2. With this scenario, the database can be expanded continuously through the business cycle. In the present paper, we describe the database of 3-D face and head shapes and the three elemental technologies. Subsequently, integration of these technologies and validity for actual retail services are discussed. 3 Elemental technologies3.1 Homologous human body shape modeling and database of 3-D face and head shapesThe database of 3-D face and head shapes was the technological foundation of the three elemental technologies. In this section, general methods for human body measurement and homologous shape modeling are described. Subsequently, concrete technologies for applications such as head scanning technologies and head modeling methods are described.Optical scanning technologies are used commonly for human body shape measurement. A patterned light source is projected onto the body surface, and the reflected light is captured by camera from a direction different from the light source. The 3-D coordinates of the surface are calculated by triangulation [2]. The human body shape obtained by scanning consists of 3-D coordinates of a large number of data points (many millions) and 3-D coordinates of dozens of anatomical landmarks. Anatomical landmarks are feature points based on anatomical correspondence. Many landmarks are defined based on specific positions on bones. An expert anthropometrist palpates and determines the landmark locations, and places markers on the skin surface before scanning. Most body scanners can automatically detect the markers on landmarks and calculate their 3-D coordinates. Landmark locations have the same anatomical meaning in all subjects, and all subjects have the same number of landmarks. However, the number of landmarks is too small to represent the 3-D body shape.The number of data points on the body surface is large enough to represent the 3-D body shape, but the number differs between subjects. It is difficult to compare 3-D body shape in different subjects, as the number of landmarks is too small to describe body shape, and the number of data points on the body surface differs between subjects. Therefore, we proposed homologous body shape modeling. With homologous body shape modeling, the body surface shape of any subject is described with the same number of data points and the same topology. We proposed polygonal representation. Body shape is described with k piece of 3-D coordinate vector T, which has k × 3 components. The average body shape of m persons can be calculated easily from multiple shape vectors T1,T2,..., and Tm. By analyzing the matrix M=[T1,T2,...,Tm] using the principal component analysis (PCA), eigenvector P is obtained.Body shapes for m persons can be represented in the eigenspace by using principal components A1 - Am. When n principal components explain sufficient variance, human body variations can be described in the eigenspace of n dimensions (n

元のページ 

10秒後に元のページに移動します