Detail of the student project

Topic:Relative pose from 4 points in 3 views with the help of CNNs
Department:Katedra kybernetiky
Supervisor:RNDr. Zuzana Kúkelová, Ph.D.
Announce as:Diplomová práce, Bakalářská práce, Semestrální projekt
Description:Relative pose estimation solvers are core components of 3D reconstruction pipelines. They are used for autonomous driving, augmented reality, 3D surveying, and many other applications. Among relative pose estimation problems, the 4 points in 3 views problem is the problem of estimating the relative poses between 3 images from 4 shared observations. This problems leads to complex systems of polynomial equations for which efficient solvers are missing. On the other hand, the 5 point in 2 views problem (the classical essential matrix estimation problem) has fast solvers. The goal of this project is to combine deep neural networks with the efficient 5 point solver to solve the unsolved 4 points in 3 views problem. For more details, please contact the supervisor.
Responsible person: Petr Pošík