Optimal templates for nonrigid surface reconstruction

This paper addresses the problem of reconstructing a deforming surface from point observations in a monocular video sequence. Recent state-of-the-art approaches divide the surface into smaller patches to simplify the problem. Among these, one very promising approach reconstructs the patches individually using a quadratic deformation model. In this paper, we demonstrate limitations that affect its applicability to real-world data and propose an approach that overcomes these problems. In particular, we show how to eliminate the need for manually picking a template that is used to model the deformations. We evaluate our algorithm on both synthetic and real-world data sets and show that it systematically reduces the reconstruction error by a factor of up to ten.