The ability to handle large amounts of outliers is essential for many
applications in, for example, multiple view geometry and feature-based image
registration. In this talk, one of the most successful approaches for this
purpose, namely RANSAC will be reviewed. The general goal of RANSAC is to find
the solution which has the maximum number of inliers. However, the method has
no guarantee of finding the optimal solution. I will show how one can modify
the standard scheme and obtain such guarantees, resulting in an algorithm with
polynomial-time complexity. Other robust loss functions, such as the truncated
L2-norm, will also be considered.

Prof. Fredrik Kahl is with
Centre for Mathematical Sciences, Lund University
as well as with
Signals and Systems, Chalmers University of Technology

The seminar takes place in Dejvice Campus, room T2:A4-7.