Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Content-based Long-term Visual Localization for Aerial Navigation
This project proposes to combine both a visual localization pipeline and style transfer techniques to diminish the effects of appearance changes that occur over long periods of time (e.g. seasons, illumination).
In order to autonomously navigate in a workspace, a robot needs to know where it is. As such, Visual localization attempts to estimate a 6 Degree-of-Freedom (DoF) camera pose from which an image was taken relative to a known scene. Unmanned Aerial Vehicles (UAVs) can visit the place from very different viewpoints, making image matching for visual localization very challenging. On top of this, over long periods of time, seasonal and illumination changes (e.g. dawn, day, sunset, night) can drastically affect the appearance of the location, posing major challenges for localization. To diminish this problem and aid place recognition, this project aims to apply style transfer neural networks to separate the content of an image from its style (texture), making it easier to tackle localization at different appearances.
In order to autonomously navigate in a workspace, a robot needs to know where it is. As such, Visual localization attempts to estimate a 6 Degree-of-Freedom (DoF) camera pose from which an image was taken relative to a known scene. Unmanned Aerial Vehicles (UAVs) can visit the place from very different viewpoints, making image matching for visual localization very challenging. On top of this, over long periods of time, seasonal and illumination changes (e.g. dawn, day, sunset, night) can drastically affect the appearance of the location, posing major challenges for localization. To diminish this problem and aid place recognition, this project aims to apply style transfer neural networks to separate the content of an image from its style (texture), making it easier to tackle localization at different appearances.
WP1: Literature Review in Visual Localization and Style Transfer.
WP2: Development of a style transfer approach to be integrated in an existing visual localization pipeline.
WP3: Experimentation of the method against state-of-the-art approaches in Place Recognition.
Wp4: Report Writing.
WP1: Literature Review in Visual Localization and Style Transfer. WP2: Development of a style transfer approach to be integrated in an existing visual localization pipeline. WP3: Experimentation of the method against state-of-the-art approaches in Place Recognition. Wp4: Report Writing.