Multiple visual feature identification and integration for 3D recovery of rotating objects

Abstract

The purpose of this study was to investigate ways to separate and extract various visual information from the continuous images of a rotating object and to recover 3D shapes by integrating the results. In this study, the pattern, corners, (contours) and highlights in an image are separated and extracted through clear qualitative analysis of the visual features. The extracted visual features are successively applied to the “shape from X,” and the results obtained are integrated. Initially, the features that can be acquired with stability are applied to the shape from X, and the shape is recovered. Then, based on the information obtained, ambiguous visual information is determined. By iterating this procedure, ambiguous information is converted to clear explicit information, making it possible to derive a more exact 3D shape. © 1998 Scripta Technica, Syst Comp Jpn, 29(13): 56–65, 1998