Estimation du mouvement apparent par des fonctions splines vectorielles : application à l'océanographie et à la météorologie

Abstract : The document describes the issue of apparent motion estimation from satellite image data. Major aplications concern oceanography and meteorology. The observed displacements of these geophysical fluids on image sequences are mostly due to surface currents on ocean, and winds in the atmosphere. Classical image processing methods are not well adapted to these phenomena, as they are mostly dedicated to rigid motion. They do not take into account deformation, compressibility, turbulence, which are occuring when studying fluid motion. This PhD-thesis details various motion estimation methods, dedicated to fluid observations and based on the formalism of vector splines. With this approach, motion estimation relies on a part of the images, for which the transport equation is verified, and uses regularity properties on the remaining data. This appears to be more suitable than methods which search a local compromise, at each pixel, between conservation of grey level value and regularity. The first contribution, that is described in the document, extend an existing method, which has been designed for incompressible fluid flows and relies on conservation of grey level values and second-order div-curl regularity properties. An improved model is proposed for processing compressible fluid images, thanks to the mass conservation equation and a given set of control points. The next contributions perform a multi-scale estimation, in order to take turbulence into account in the estimation process. The second model proposes a multi-scale description of control points, that are defined at full resolution. This description is obtained by generating sub-sets of control points, with a decimation algorithm, that uses geometrical properties. Sub-sets have a decreasing density of control points and each one corresponds to a specific scale. At a given scale, motion is described by a vector spline, that satisfies the brightness conservation and minimize the second-order div-curl regularity constraint. The third model uses a pyramidal representation of image data, in order to perform the multi-scale motion estimation. A scale is defined at each level of the pyramid. Motion is first estimated at low resolution, and increments are then computed from one level to the next one. Estimation relies on a vector spline, that satisfies brightness or mass conservation equation and minimizes the second-order div-curl regularity. Result is looked for in the set of functions of the studied scale. A qualitative and quantitative analysis of results obtained with these models is provided at the end of the document. That confirms that these methods are more suitable than classical ones for estimating fluid flow motion. Future applications concern the use of the resulting motion fields as initial conditions of geophysical models or as observations in a data assimilation process.
Keywords : .
Document type :
Theses
Complete list of metadatas

https://hal.inria.fr/tel-01301974
Contributor : Nathalie Gaudechoux <>
Submitted on : Wednesday, April 13, 2016 - 1:40:29 PM
Last modification on : Friday, May 25, 2018 - 12:02:03 PM
Long-term archiving on: Tuesday, November 15, 2016 - 2:35:43 AM

Identifiers

  • HAL Id : tel-01301974, version 1

Collections

Citation

Till Isambert. Estimation du mouvement apparent par des fonctions splines vectorielles : application à l'océanographie et à la météorologie. Modélisation et simulation. Université Paris Descartes, 2007. Français. ⟨tel-01301974⟩

Share

Metrics

Record views

1000

Files downloads

268