Pairwise Surface Registration Using Local Voxelizer

Abstract

Surface registration is the process that brings scans into a common coordinate system by aligning their overlapping components, which can be achieved by finding a few pairs of matched points on each scan pair using shape descriptors and employing the matches to compute an alignment transformation. This paper proposes a local voxelizer descriptor, and the key idea is to define a unique local reference frame (LRF) using the local shape around a basis point, perform voxlization for the local shape within a cubical volume aligned with the LRF, and concatenate local features extracted from each voxel to construct the descriptor. A pairwise registration algorithm is developed by choosing a single pair of matched points using the local voxelizer descriptor, and computing a rigid transformation based on aligning the corresponding LRFs. Quantitative experiments show that our algorithm can register scan pairs with small overlap, while maintaining acceptable registration accuracy.