This thesis is concerned with problems in the field of learning-based template tracking as well as in the field of 3D data processing. In the field of template tracking, efficient and robust learning approaches are presented that allow learning at run-time and high-speed tracking at more than 1000 fps. In the field of 3D data processing, a highly efficient normal estimation method as well as a learning-based 3D interest point detection approach are presented. The latter mimics existing methods at much higher speed as well as allows for optimizing a detector for specific characteristics.
«
This thesis is concerned with problems in the field of learning-based template tracking as well as in the field of 3D data processing. In the field of template tracking, efficient and robust learning approaches are presented that allow learning at run-time and high-speed tracking at more than 1000 fps. In the field of 3D data processing, a highly efficient normal estimation method as well as a learning-based 3D interest point detection approach are presented. The latter mimics existing methods a...
»