|
- Date: 08.01.2015
- Time: 11:15am
- Location: Room 02.09.023, FMI-Building, Garching, Boltzmannstr. 3
|
Learning-based Approaches for Template Tracking and Interest Point Detection
Abstract:
The first part of this thesis focuses on learning-based template tracking approaches, with the goal to follow the position of an object or region within an image sequence. Since the seminal work of Lukas and Kanade, a lot of advances have been made in this field. While learning-based approaches led to an increase in speed and robustness, they are still limited in learning speed, flexibility, and occlusion handling. These problems are addressed within this thesis.
The field of 3D data processing has recently obtained a boost due to the introduction of low cost depth sensors. Therefore, the second part of the thesis focuses on fast methods for 3D data processing. This includes an efficient normal estimation method as well as a learning-based approach for interest point detection in depth maps.
