3D registration of partially overlapping surfaces using a volumetric approach

摘要

A novel volumetric approach for the 3D registration of a generic crown and a complete tooth of the same tooth type is presented. Unlike most existing 3D partial surface registration methods, our algorithm does not require the two surfaces to be instances of the same object, only that they are of the same class (i.e., a group of similar objects). The Euclidean distance fields of the two surfaces are computed, and a set of distance vectors within these distance fields are used to generate a series of candidate transformations. By verifying the set of candidate parameters, the one resulting in the smallest registration error is chosen as the optimal solution. An additional advantage of our algorithm is that initial pose estimation is not required. Its effectiveness is confirmed by a series of experiments.