Best PhD Thesis in Computer Vision by CEA 23-24 (Sep’25)

Dr. Mercedes Garcia-Salguero has received the prestigious ‘Best PhD Thesis in Computer Vision‘ award from the Comité Español de Automática (CEA) for the 2023-2024 period. Her award-winning thesis, titled “Certifiable algorithms for 3D computer vision problems,” was supervised by Dr. Javier Gonzalez-Jimenez and successfully defended at the University of Malaga in April 2024.

Thesis Summary: Many problems encountered in computer vision can have multiple local solutions. However, algorithms traditionally used to solve these problems are iterative, meaning they can return any of these solutions without the possibility of verifying whether the solution is globally optimal or not. A priori, there is no way to know how far the local minima are from the global one, and this difference in solution quality can affect other blocks that use them, and even the final result of the system. Therefore, having an algorithm or system in general that allows us to know if the solution is optimal becomes especially important in these situations. In this thesis, we propose a set of algorithms that are capable of certifying whether the solution is the global optimum or not for two highly relevant problems: the calculation of relative pose between cameras and the triangulation of points in space.