Reconstruction of object surfaces by using occlusion information from active stereo vision

Abstract

With robots moving in an unknown 3D environment, it is necessary to work out a method for reconstruction of surface structures (that is, where a surface is present and where there is a free space). In this paper, a method is proposed to recover surfaces by means of active moving of a stereo camera. First, the surface structure of the environment under consideration is represented by a set of virtual 3D segments (arcs) obtained by connecting edge points that lie on the same epipolar line; then the 3D position information and occlusion information acquired by active moving of the stereo camera are used to decide if a physical surface exists between the arcs. Using surface structures that have been so far recovered, camera location that offers the best results in surface reconstruction is predicted, and the camera is moved to this new location to obtain more precise results. The proposed method offers a way to select viewpoints that ensure efficient surface reconstruction, which has not been discussed in any previous study. Experiments showed that the proposed method offers correct reconstruction of object surfaces with a small number of measurements. © 1997 Scripta Technica, Inc. Syst Comp Jpn, 28(9): 86–97, 1997