Event Location: NSH 1507
Bio: Ijaz Akhter is a PhD candidate at the LUMS School of Science and
Engineering co-advised by Sohaib Khan and Yaser Sheikh. His area of
research is nonrigid structure from motion, particularly in modeling
dynamic objects. He spent the last few months interning with Iain
Matthews at Disney Research, Pittsburgh.

Abstract: Many objects that abound around us are time-varying 3D structures, like
faces, bodies, and vehicles. The analysis of their motion is a key
problem in computer vision. In this talk, we propose a linear space-time
representation for the analysis of dynamic 3D structures. The proposed
representation assumes that dynamic structures lie in a low-dimensional
shape space and exhibits smoothness across trajectories. By exploiting
both the shape compactness and time smoothness, we show that a linear
joint space-time model can be built, which has more compactness than
individual linear shape and trajectory models. We demonstrate the use of
the spacetime model for the tracking of unlabeled motion capture data.