<> "The repository administrator has not yet configured an RDF license."^^ . <> . . "Modular Optical Flow Estimation With Applications To Fluid Dynamics"^^ . "Optical flow is the apparent motion of intensities in an image sequence. Its estimation has been studied for almost three decades. The results can be used in a wealth of possible applications ranging from scientific applications like experimental fluid dynamics over medical imaging to mobile computer games. The development of a single solution for all optical flow problems seems to be a worthwhile goal. However, in this thesis, we argue that this goal is unlikely to be achieved. We thoroughly motivate this hypothesis with theoretical and practical considerations. Based on the results, we identify two major problems that significantly complicate the research and development of new optical flow algorithms: First, very few reference implementations are publicly available. Second, not all relevant properties of the proposed algorithms are described in literature. In the first part of this thesis, our contribution is to alleviate both problems. First, we discuss a number of algorithm properties which should be known by the user. Second, by decomposing existing optical flow methods into their individual algorithm building blocks, shortly called modules, we propose to individually analyze the properties of each module independently. A large number of existing techniques is composed of relatively few existing modules. By implementing these modules in a software library called Charon and adding tools for the evaluation of the results, we contribute to the accessibility of reference implementations and to the possibility of analyzing algorithms by experiments. In the second part of this thesis, we contribute two modules which are vital for the estimation of fluid flows. They are specifically tuned to the imagery obtained for particle tracking velocimetry (PTV). We call the first module estimatibility measure. It detects those particle locations where fluid motion can be estimated. It is based on the constant position of the center of gravity of the connected components generated by a large number of thresholded versions of the original image. The module only needs a few intuitive parameters. Experiments indicate its robustness with respect to noise with varying mean and variance. To analyze the properties of this module we also provide a framework for simulating the particle image generation. The second module is a motion model based on unsupervised learning via principal component analysis. Training data is provided through Computational Fluid Dynamic (CFD) simulations. The model describes local ensembles of trajectories which can be fitted to the image sequence by means of a similarity measure. Together with a standard similarity measure and a simple optimization scheme we derive a new PTV method. Compared to existing techniques, we obtained superior results with respect to accuracy on real and synthetic sequences with known ground truth. All source code developed during the thesis is available as Open Source following the GNU Lesser General Public License (LGPL)."^^ . "2009" . . . . . . . . "Daniel"^^ . "Kondermann"^^ . "Daniel Kondermann"^^ . . . . . . "Modular Optical Flow Estimation With Applications To Fluid Dynamics (PDF)"^^ . . . "thesis.pdf"^^ . . . "Modular Optical Flow Estimation With Applications To Fluid Dynamics (Other)"^^ . . . . . . "indexcodes.txt"^^ . . . "Modular Optical Flow Estimation With Applications To Fluid Dynamics (Other)"^^ . . . . . . "lightbox.jpg"^^ . . . "Modular Optical Flow Estimation With Applications To Fluid Dynamics (Other)"^^ . . . . . . "preview.jpg"^^ . . . "Modular Optical Flow Estimation With Applications To Fluid Dynamics (Other)"^^ . . . . . . "medium.jpg"^^ . . . "Modular Optical Flow Estimation With Applications To Fluid Dynamics (Other)"^^ . . . . . . "small.jpg"^^ . . "HTML Summary of #10184 \n\nModular Optical Flow Estimation With Applications To Fluid Dynamics\n\n" . "text/html" . . . "004 Informatik"@de . "004 Data processing Computer science"@en . .