Seminars/Colloquia/Invited Talks

Seminars

Fredrik Kahl

Multiple View Geometry and L-infinity Optimization

PLACE:	Clark 314
EVENT:	CIS Seminar Series
DATE:	April 12, 2005
TIME:	1:00 - 2:00

Abstract

In this talk, a framework for solving geometric reconstruction problems in computer vision will be presented based on the L-infinity norm. Instead of using the common sum-of-squares cost-function, that is, the L-2 norm, the model-fitting errors are measured using the L-infinity norm. Unlike traditional methods based on L-2, this framework allows for efficient computation of global estimates. It will be shown that a class of geometric structure and motion problems, for example, triangulation, camera resectioning and homography estimation can be recast as a quasi-convex optimization problem within the framework. These problems can be efficiently solved using Second Order Cone Programming (SOCP) and Bisection which are standard techniques in convex optimization. The methods have been validated on real data in different settings with small and large dimensions and with excellent performance.

Brief biography

Fredrik Kahl received his MSc degree in computer science and technology in 1995 and his PhD in mathematics in 2001. His thesis was awarded the Best Nordic Thesis Award in pattern recognition and image analysis 2001-2002 at the Scandinavian Conference on Image Analysis 2003. He has been a postdoctoral research fellow at the Australian National University (ANU) and is currently a visiting Research Fellow 2004-2005 at the University of California, San Diego (UCSD). His main research area is computer vision, in particular, geometric reconstruction problems, photometric stereo, geometry of curves & surfaces and machine learning.