igl | Interactive Geometry Lab

abstract

Realistic modeling of the human body in 3D has many applications ranging from fashion to the production of movies and video games. However, leveraging data coming from state-of-the-art 3D acquisition systems poses a set of problems. A 3D scan of a person contains holes and thousands of unordered points. In addition, a 3D scan is a single snapshot of the human body in time, while the shape of the human body changes with motion, breathing, aging, etc. In this talk, I will present two projects related to modeling the human body in 3D: data registration and human shape representation. First, I will describe a method for finding sparse pairwise correspondences between 3D triangular meshes of articulated objects, such as humans, in various shapes and poses. Our search framework effectively explores the space of possible correspondences and is more robust to local optima than previous work. Central to this framework are features based on mesh surface paths that are invariant to shape, pose, and resolution. Second, I will talk about an approach for capturing and modeling the non-rigid intrinsic shape variation of the human body during breathing. In this work, we learn a detailed model of body shape deformations due to breathing for different breathing types and provide simple animation controls to render lifelike breathing regardless of body shape. We also develop a novel interface for breathing animation using a spirometer, which measures the breathing volume of a "breath actor". Our approach generates fine-scale body shape deformations due to breathing with greater ease and realism than previously achieved.

bio

Aggeliki Tsoli is a postdoctoral researcher in the Computer Vision and Robotics Laboratory (CVRL) at the Foundation for Research and Technology-Hellas (FORTH). She earned her B.Sc. in Electrical and Computer Engineering from the University of Thessaly in 2005 and her M.Sc. in Computer Science from Brown University in 2008. She obtained her Ph.D. in 2014 from Brown University and the Max Planck Institute for Intelligent Systems in Tuebingen, Germany, under the supervision of Professor Michael Black in 2014. She is interested in modelling articulated objects in 3D with special interest in the human body and hands. So far, her work has focused on data-driven approaches for modeling 3D surface geometry. Applications range from 3D shape matching to animating subtle non-rigid deformations of the human body e.g. due to breathing.

Homepage: http://www.ics.forth.gr/~aggeliki/