A new feature-based method for robust and efficient rigid-body registration of overlapping point clouds

We propose a new feature-based registration method for rigid-body alignment of overlapping point clouds (PCs) efficiently under the influence of noise and outliers. The proposed registration method is independent of the initial position and orientation of PCs, and no assumption is necessary about their underlying geometry. In the process, we define a simple and efficient geometric descriptor, a novel k-NN search algorithm that outperforms most of the existing nearest neighbor search algorithms used for the same task, and a new algorithm to find corresponding points between PCs based on the invariance of Euclidian distance under rigid-body transformation.