Dynamic Three-Dimensional Human Model

Abstract

Three-dimensional (3-D) content is widely used in movies and video games, thanks to advances in computer graphics, motion capture, 3-D scanners, and so on. However, producing high quality 3-D content is expensive and involves many different technologies. The dynamic 3-D model discussed in this chapter contains shape information about a moving subject. Thus, it can express the motions of humans in detail, including movements of clothes. It is useful as a component in video synthesis, because of its flexibility in viewpoint and view direction. In addition, it can be transformed into stereoscopic images such as binocular, lenticular, integral photographic, etc. To generate a dynamic 3-D model, 10 or more cameras are set up in a dome surrounding a subject to synchronously take video images. The model is calculated frame by frame from these video images. There are many choices from which to choose a shape restoration algorithm. Of these, the popular volume-intersection method is steady and has low calculation cost [1]. However, it cannot produce a concave surface of a subject. To make up for this fault, various methods such as stereo matching [2, 3], voxel coloring [4, 5], photo hull [6], deformable mesh [7], etc., have been proposed. The quality of the dynamic 3-D model depends not only on the accuracy of its shape, but also the quality of texture mapping. When multiple cameras