GPI Seminar Series: Gerard Pons-Moll
 |
Gerard Pons-Moll, Max Planck Institute for Intelligent Systems (Tübingen) |
Modeling Humans in Motion: Capture and Animation
Abstract:
In the first part of the talk I will describe novel techniques to estimate human pose from video and inertial sensors both in semi-controlled and uncontrolled scenarios such as monocular images. In such settings, the human motion is modeled as an articulated kinematic chain to make inference tractable. Although articulated structures are a good approximation of human motion, they are a crude representation for many applications such as animation in movie production. Indeed, to look human, digital avatars need to have soft-tissue deformations. In the second part of the talk I will present a realistic model of humans in motion. We address this with a novel 4D capture system and a method to register a template mesh to sequences of 3D scans. This allows us to learn a model that relates articulated motion to non-rigid soft-tissue deformations. We demonstrate the generalization of our model to unseen motions and subject shapes.
Gerard's short bio:
Postdoctoral researcher at Perceiving Systems department at the Max Planck Institute for Intelligent Systems, directed by Michael Black. Gerard obtained his PhD from the TNT lab at the Leibniz University of Hannover, while working with people from MPI-Informatik, UofT and Microsoft Research, Cambridge.
Gerard is interested in computer vision and machine learning, with a special emphasis on the annalysis and estimation of human shape and pose from images.