Semi-regular mesh representation of 3D dynamic objects based on correspondence matching

Abstract

In this paper, we propose a novel method to represent 3D dynamic objects, which are captured with a laser scanner in successive time instances. The objective is to convert an input mesh sequence into a semi-regular mesh sequence with time-invariant topology information, since it enables the easier manipulation of 3D data using many signal processing techniques. We achieve this objective by performing global motion estimation, local deformation estimation, and correspondence optimization subsequently. Simulation results show that the proposed algorithm reconstructs the original geometry faithfully. Furthermore, the resulting semi-regular mesh sequences can be effectively compressed by adopting wavelet coding schemes.