Estimation et analyse de champs denses de vitesses d'écoulements fluides. (Estimation and analysis of dense fluid flows)

This thesis is concerned with uid motion analysis in image sequences and involves two main issues.In a rst part, we address the problem of estimating a dense motion eld. Due to the great deal ofspatial and temporal distortions that luminance patterns exhibit in uid imagery, standard techniquesfrom Computer Vision, originally designed for quasi-rigid motions with stable salient features, are notwell adapted. In that prospect, we investigate a dedicated energy-based motion estimator. The consi-dered functional includes an original data model relying on the continuity equation of uid mechanics.This new data model, which is speci cally designed to be embedded in a multiresolution framework,is associated to an original \div-curl" type regularization. The performances of the resulting uid owestimator are demonstrated both on synthetic and real meteorological image sequences. The methodis furthermore validated on an experimental ow representing a free turbulent shear layer.The second step of this study is concerned with the analysis of a dense motion eld preliminaryestimated between two images of a uid video. We propose a method to extract the vortices andsources/sinks following the Rankine model. This problem is for instance essential in meteorology (toidentify and track depressions or convective clouds in satellite images) or to characterize experimentaluid ows. The knowledge of such structures allows in addition a compact representation of the owwhich is very useful in both experimental and theoretical uid mechanics. The method we propose hereis based on an analytic representation of the ow and has the advantage to extract others pertinentinformations of the motion (such as potential functions, the Helmholtz decomposition, singular points,...). This approach is experiment on various displacements elds showing di erent physicals events.