Estimation of 3D model and object color using multiple color images

Abstract

AbstractIn this study, the generation of realistic 3D CG images is considered, and in particular, a method is proposed to estimate an object's 3D shape and color from image series taken at fixed positions of the camera and light source while rotating the object. An object placed on a turntable is treated as a set of surface elements each having seven degrees of freedom (DOF). Epipolar plane images (EPI) are analyzed using input image series, and all DOF for each surface element are calculated in order to estimate the 3D shape and color. First, loci drawn by surface elements on the EPI are tracked. Correct tracking is ensured by preventing the influence of fluctuations in object surface reflectivity, and self‐occlusions caused by uneven shape. For highly reliable loci, the 3D positions of the surface elements are calculated, while the other surface elements are located by interpolation. In addition, the RGB values of surface elements can be expressed as linear sums of diffuse reflection components and specular reflection components, so that vectors in RGB color space can be separated into those related to the object, and those related to the light source. Color estimation for all surface elements is performed by using a two‐color reflection model. In experiments, all surface element DOF were calculated for various objects, and 3D CG images, including those with light source data added, were synthesized successfully, thus demonstrating the effectiveness of the proposed method. © 2001 Scripta Technica, Syst Comp Jpn, 32(11): 49–60, 2001