Kiran Varanasi was born in India and has obtained his Bachelors and Masters degrees in Computer Science from the International Institute of Information Technology (IIIT) at Hyderabad, India. He has worked for his Masters thesis "Geometric grouping of planar patterns under perspective skew" under the direction of Dr. P.J. Narayanan, which he defended in Nov 2006. As part of his masters project, he built an interactive UI for correcting perspective distortion in images of architectural models. This got him interested into the wider field of image based modeling and rendering (IBMR). He has later joined the PERCEPTION group at INRIA Grenoble, France to pursue his doctoral studies. Since then, he has been working with Dr. Edmond Boyer on the modeling of 3D dynamic scenes from a multi-camera environment. Over the course of his Ph.D, he has convinced himself that the richness of the real world can only be modeled dynamically.

AG 1, AG 3, AG 5, SWS, AG 4, RG1, MMCI   Info

AG Audience

English

60 Minutes

Saarbrücken

Recent technological advances in various 3D sensing technologies have provided us with the means to capture the 3D shape of a scene at a fast frame-rate. Various methods exist to produce 3D video, i.e, 3D snapshots of a dynamic scene as independent visual reconstructions. For example, multi-view silhouettes from a synchronized multi-camera setup can produce a sequence of visual hull meshes, each of which can be readily mapped with texture information from the taken images. However, these 3D reconstructions are not temporally consistent and suffer from severe geometric and topological artifacts. Such data are often called space-time models or 4D models.

I will discuss a few methods that we have developed during my thesis to build a coherent spatio-temporal model from such reconstructions, from the bottoms-up. I will present three problems (a) temporally coherent segmentation of a sequence (b) feature detection and matching between meshes and (c) tracking the surface of an unknown and possibly varying topology. We make no assumptions about a template skeleton, mesh model or the topology of the shape being observed. Thus, the solutions we have developed are applicable to a dynamic scene in all its generality - composed of multiple actors, who are dressed in loose clothing, and who are interacting with each other in an arbitrary fashion.

Christian Theobalt

--email hidden

passcode not visible

logged in users only

Computer Graphics, Computer Vision