Stable Anchors for Matching Unlabelled Point Clouds

Ubaldo Ruiz, Stéphane Marchand-Maillet, Edgar Chávez

Abstract: We define an anchor as a small sample representing a point cloud. Weighted or labeled point clouds are amenable for a stable anchor extraction, a sampling method ensuring a consistent selection of points across realizations of the same point cloud. In this work, we present a heuristic to extract a stable anchor from unlabeled point clouds when the points have no weights and are indistinguishable. This problem arises when we need to query an extensive collection of point clouds and want to avoid a sequential comparison with all members of the collection. Our method consists in assigning as weight a centrality measure. We show that our approach preserves several times the bare minimum required to identify point clouds under similarity transformations.