From unordered range images to 3D models: a fully automatic multiview correspondence algorithm

In this paper, we present a fully automatic multiview correspondence algorithm based on our tensor representation. Our algorithm can automatically establish correspondences between a set of unordered range images of an object and register them in a common coordinate basis. Existing techniques of correspondence are either manual or inefficient and require the ordering of range images according to their overlap. Our algorithm does not suffer from any such limitation. We also present a complete and fully automatic 3D modeling framework by integrating our multiview correspondence algorithm with a global registration, an integration and a surface reconstruction algorithm. Our results show that our algorithm is fully automatic, applicable to free-form objects, efficient and independent of the resolution and sampling of the range images