Multi-view Normal Field Integration for 3D Reconstruction of Mirroring Objects

摘要: In this paper, we present a novel, robust multi-view normal field integration technique for reconstructing the full 3D shape of mirroring objects. We employ turntable-based setup with several cameras and displays. These are used to display illumination patterns which reflected by object surface. The pattern information observed in enables calculation individual volumetric fields each combination camera, turntable angle. As might be blurred depending on surface curvature or due non-perfect characteristics, locally adapt decoding finest still resolvable resolution. complex real-world scenarios, contain regions without observations occlusions outliers interreflections noise. Therefore, reconstruction using only is challenging. Via non-parametric clustering hypotheses derived point scene, obtain both most likely local consistency estimate. This utilized an iterative min-cut based variational approach reconstruct geometry.