Toward non-rigid dynamic cage capture

Abstract

This chapter explores the problem of non-rigid alignment using cage-based parametrization to obtain consistent dynamic meshes without any assumptions on temporal matching. Then, we describe a reconstrucstion framework to retrieve 3D geometric information from multi-view silhouette cues at each time step. To achieve this goal, we establish robust temporal correspondences relying only on geometric features. This investigation toward cage-based registration from multi-view silhouettes is a first insight in the context of dynamic mesh capture with controllable, reusable and low-dimensional outputs designed for commercial animation engines. We finally provide a series of results using our iterative cage-based registration technique in order to obtain a consistent dynamic mesh structure.