Abstract
Perception of our own bodies and people around us is a crucial part of everybody’s daily life. People’s analysis is not limited to parsing the current scene, including the posture and shape of humans in it. We can also imagine the appearance of the scene few seconds later, or how each individual would look like when performing a completely different motion or when gaining a few kilos. The main goal of our research is to provide these reasoning capabilities to a computer. To achieve this, we believe it is essential to have a computational model of human visual appearance and human motion. In this talk, I’ll give an overview of three challenges related to this topic that our group has explored in recent years. The first one is to develop compact models of the human geometry and motion that can extrapolate to new poses and new shapes. Second, the use of these models for fitting different types of data like 3D scans, RGBD video, optical flow or video will be discussed. Finally, I will describe the use of these models as an introspection tool for studying our own perception of human bodies.
Biography
Javier Romero received his M.Sc. from the Technical University of Madrid (2007) and his Ph.D. from the Royal Institute of Technology in Stockholm (2011) under the supervision of Danica Kragic and Hedvig Kjellström. After post-doctoral research in the Perceiving Systems department of the Max Planck Institute for Intelligent Systems in Tübingen, he became a Research Scientist in the same department in 2013. His research interest lies at the intersection between computer vision and graphics, where realistic geometrical models are computed from real data and used for understanding such data. He focuses on the perception of human bodies due to their interesting shape properties and the broad impact of their related applications.