Project Details
Projekt
Restoration and Post Processing of Optical Flows
Applicants
Privatdozent Dr. Christoph Garbe; Professor Dr. Tobias Preusser; Professor Dr. Martin Rumpf
Subject Area
Mathematics
Term
from 2005 to 2010
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 12569739
Image sequences recorded from motion processes such as water waves, traffic streams or growth processes are usually characterized by image artifacts. Very often, a complete sup pression of artifacts in the experimental set-up is impossible to achieve. Examples are reflec tive zones an water waves due to inhomogeneous lightning conditions, reflections in thermal imagery, occlusions due to moving objects in 3D scenes and rain and dirt on windscreens. In dependent on whether they rely on classification by local estimation or on global variational methods, optical flow computations face serious difficulties with these artifacts. We are aiming at new methods which combine variational techniques for optical flow and image in-painting with local flow estimators, and a reliability control based on quantized confidence estimates and power laws from Fourier and wavelet analysis. We plan to integrate in this project knowledge from different research groups including the LOCOMOTOR project and the group of Prof. Maaß.Furthermore, we are aiming at post processing of the resulting flow data with image processing methodology based on anisotropic diffusion and at clustering of the fields using algebraic multi grid methods, developed in a previous SPP-project.
DFG Programme
Priority Programmes
