Australian Institute for Machine Learning
A central task in computer vision involves fitting parametric models to digitised images. In particular, given a collection of image feature locations, one seeks a set of parameters that describe a specific relationship between the image locations. Often the data-parameter connection takes the form of a system of equations, and the underlying estimation task falls into the category of non-linear heteroscedastic errors-in-variables regression problems. I will present an approximate maximum likelihood cost function and concomitant numerical scheme which encapsulates a broad class of problems including conic fitting, optical flow, homography and fundamental matrix estimation; trifocal tensor estimation; camera calibration and three-dimensional rigid motion estimation. Thanks to Julia’s Unicode support and multiple dispatch capability I will be able to demonstrate a clear and concise implementation which embodies the various natural levels of mathematical abstraction that lead to generic code.
Zygmunt L. Szpak received his Ph.D. degree in Computer Science from the University of Adelaide, Australia, in 2013, and his M.Sc. degree in Computer Science from the University of KwaZulu-Natal, South Africa, in 2009. He is a senior research fellow at the Australian Institute for Machine Learning—an institute at the University of Adelaide. He works on numerous industry inspired computer vision problems. In the last couple of years, his work has focused on the application of machine learning and multiple-view geometry techniques for the development of smart medical devices.
