Levi Valgaerts
Levi Valgaerts did his master's thesis at the Mathematical Image Analysis Group under supervision of Prof. Weickert at Saarland University. He graduated from CSE/BGCE in January, 2007, and is now PhD student at the Saarland University (see his homepage).
Master's Thesis:
Combining Variational and Feature-based Methods for
Motion Estimation
Variational methods currently belong to the most accurate techniques for the recovery of the optic flow of an image sequence. Major contributions have been made in the last years to both their accuracy and their real-time performance. Most of the efforts to improve the accuracy of these methods have focussed on the design of new data terms and smoothness terms that enter the energy functional. So far no research has been done on combining variational optic flow with featurebased methods such as SIFT. The goal of this thesis is to investigate how and to what extend optic flow methods can benefit from the results of a feature extractor. We will introduce a new method for optic flow estimation that is based on a unified model for tensor field interpolation. This method uses the sparse displacement field, that can be established by matching features from two images, to estimate the dense optic flow. Because of their similar formulation, the variational method and the feature-based method can easily be merged into one combined method. The accuracy of the three methods is finally compared in a series of experiments.
Estimation of the dense optic flow from a sparse displacement field
that has been established by a feature matching algorithm
(download high-resolution versin as PDF).
