Despite years of research, the reliable shape reconstruction of highly specular objects is still a largely unsolved problem, especially for complex objects or worse-than-ideal observation conditions. In this report, we elaborate on a novel multi-view specular reconstruction method based on the consistency of normal vector maps (NVMs). In particular, this algorithm is applicable to complex moving objects, where most "standard" techniques fail. We start by demonstrating how NVMs represent the specular reflection data, then re-formulate the reconstruction problem in terms of an energy functional to be optimized. Finally, we suggest an efficient solution of the problem as a modification of the probabilistic voxel carving approach.
展开▼