Abstract

The talk will be structured in two parts. I will first talk about my research on marker-less motion capture, in particular presenting my recent ECCV work on hands tracking. Capturing the motion of hands interacting with each other and with an object is a challenging task due to the large number of degrees of freedom, self-occlusions, and similarity between the fingers. I will show how we addressed this problem by proposing a generative approach modeling all the elements of the scene as a unique articulated deformable object, and exploiting some discriminatively learnt features. In the second part, I will talk about my research on interactive free-viewpoint video. I will show how it is possible to navigate a collection of videos representing an event casually captured by people in the audience. Instead of recovering a perfect 3D representation of the recorded scene, which in general is very challenging, I will propose several simplifying assumptions allowing video based rendering in such a scenario. I will also show a live real-time demo of the proposed interactive navigation tool on different video collections. URLs: http://cvg.ethz.ch/research/ih-mocap/ http://cvg.ethz.ch/research/unstructured-vbr/ http://www.inf.ethz.ch/personal/lballan/