Forschung | Prof. Dr.-Ing. Niclas Zeller

Map-based Localization

Thu, 25 Mar 2021 00:00:00 +0000

Map-based localization (relocalization) focuses on the task of estimating the global position of a vehicle or robot within a pre-built map by using the vehicle's/robot's onboard sensors. Especially for camera based relocalization a major challenge lies in the design of algorithms that are robust to appearance changes caused by weather, lighting and structural changes. To enable robust map-based localization there is an increasing demand for multi-weather datasets suitable for the development and evaluation of state-of-the-art algorithms.

Dense 3D Reconstruction

Fri, 01 Jan 2021 00:00:00 +0000

Dense 3D reconstruction aims at recovering a complete and detailed 3D model and understanding of the environment using only one or a few cameras. Our research combines classical multi-view geometry and visual odometry with deep neural networks for depth estimation and scene representation.

Light Field Processing

Thu, 25 Oct 2018 00:00:00 +0000

In contrast to standard cameras, light field (or plenoptic) cameras capture the light field of a scene in four dimensions (i.e. spatial and directional information). Therefore, among others, one is able to obtain metric 3D information from a single recording of a light field camera. Our research focuses on calibration, depth estimation, SLAM, etc. based on light field recordings.