Topology-Aware Non-Rigid Point Cloud Registration

In this paper, we introduce a non-rigid registration pipeline for pairs of unorganized point clouds that may be topologically different. Standard warp field estimation algorithms, even under robust, discontinuity-preserving regularization, tend to produce erratic motion estimates on boundaries associated with `close-to-open' topology changes. We overcome limitation by exploiting backward motion: in the opposite direction, event becomes `open-to-close', which is default handled correctly. At core our approach lies general, topology-agnostic algorithm, similar those employed recently introduced dynamic reconstruction systems from RGB-D input. improve changes an efficient post-processing phase. Based both forward and (inverted) hypotheses, explicitly detect regions deformed geometry undergo topological means local deformation criteria broadly classify them as `contacts' or `separations'. Subsequently, two hypotheses are seamlessly blended basis, according type proximity detected events. Our method achieves state-of-the-art accuracy MPI Sintel dataset. Experiments custom dataset annotations demonstrate effectiveness estimating boundaries, well promising performance explicit detection.