3D Modeling from Captured Images Using Bi-directional Ray Traversal Method

Abstract

In general, modeling techniques are classified into the creative approach and the reproductive approach. We discuss the latter approach to reconstruct existing objects. Lately, computers and networks are significantly improved, so that application algorithms are also expected to be improved as well. For instance, capturing and modeling existing objects without special equipment is really desirable. We propose a method to reconstruct whole appearances of existing objects with voxel space from 2D images that are captured using ordinary video cameras. Our approach does not involve extracting and tracking feature points of target objects unlike some conventional methods. Our method identifies the voxel properties so as to keep the appearance from frame to frame consistent. We call the way of expression screened voxel expressions. The voxel properties are represented by state variables that represent shape and color, and they are calculated by backward and forward ray shooting. During the calculation each voxel has several candidate colors and the correct one is only determined at the end. In this paper, we will discuss the feature value assessment using some experiments. We will also explain the basic methodology of our approach.