Non-Rigid Articulated Point Set Registration with Local Structure Preservation

Abstract

We propose a new Gaussian mixture model (GMM)-based probabilistic point set registration method, called Local Structure Preservation (LSP), which is aimed at complex non-rigid and articulated deformations. LSP integrates two complementary shape descriptors to preserve the local structure. The first one is the Local Linear Embedding (LLE)-based topology constraint to retain the local neighborhood relationship, and the other is the Laplacian Coordinate (LC)-based energy to encode the local neighborhood scale. The registration is formulated as a density estimation problem where the LLE and LC terms are embedded in the GMM-based Coherent Point Drift (CPD) framework. A closed form solution is solved by an Expectation Maximization (EM) algorithm where the two local terms are jointly optimized along with the CPD coherence constraint. The experimental results on a challenging 3D human dataset show the accuracy and efficiency of our proposed approach to handle non-rigid highly articulated deformations.