RNDr. Zuzana Kúkelová, Ph.D.

Camera geometry estimation is a core task in many 3D computer vision systems, including Structure-from-Motion and visual localization, with applications in augmented reality and autonomous vehicles and drones, among others. Camera geometry estimation algorithms rely on solving system of algebraic equations. Traditionally, these systems have been solved with classical methods such as Groebner bases. Recently, there have been a few approaches based solely on machine learning. The goal of this project is to combine traditional algebraic and modern learning-based approaches to solve hard open problems in camera geometry estimation.

Camera geometry estimation is a core task in many 3D computer vision systems, including Structure-from-Motion and visual localization, with applications in augmented reality and autonomous vehicles and drones, among others. Camera geometry estimation algorithms rely on solving systems of algebraic equations. Traditionally, these systems have been solved with classical methods such as Groebner bases. Recently, there have been a few approaches based solely on machine learning. The goal of this project is to combine traditional algebraic and modern learning-based approaches to solve hard open problems in camera geometry estimation.