IPIU workshop on Vision-based localization
13th February 2019
Martin Humenberger will give a talk: “Modern approaches of visual localization”.
Abstract: Visual localization refers to estimating the 6DOF camera pose within a given 3D scene model using images. In this talk, I will first present a quick overview of different methods, and second, I will detail more recent approaches which utilise deep learning techniques for visual localization. Traditional methods use descriptor matching between the query image and the 3D scene model as the basis for pose computation, while modern learning-based approaches try to directly regress the pose or intermediate results of the localization pipeline. Furthermore, I will introduce a new synthetic dataset which explicitly targets challenging situations such as lighting changes and occlusions.