Use this URL to cite or link to this record in EThOS:
Title: Motion tracking in meteorological satellite imagery for atmospheric motion vector derivation
Author: Hernández-Carrascal, María Angeles
ISNI:       0000 0004 2721 3321
Awarding Body: University of Reading
Current Institution: University of Reading
Date of Award: 2010
Availability of Full Text:
Access from EThOS:
The focus of this thesis is the tracking of apparent motion from sequences of meteorological satellite images, in the context of Atmospheric Motion Vector (AMV) derivation. AMVs are estimates of atmospheric wind, they are routinely produced from visible, infrared window and water vapour (WV) imagery, and they represent a major contribution to the observation of the Earth's atmosphere. This research tackles a number of issues related to the derivation of AMVs from WV imagery, using region-matching methods for tracking motion. WV imagery is particularly challenging, as images typically have a soft appearance, with no edges, no background, and large areas of low contrast. The datasets used in the experiments are real sequences of images in the WV 6.2 urn band from the geostationary satellite Meteosat-9. The underlying approach throughout the thesis is Gaussian multi-scale representation, a sound mathematical framework, developed by the computer vision community, that allows to analyse images at differente scales and to handle partial derivatives in a way deeply connected to the scale. The main contributions of this thesis are: It shows how Gaussian multi-scale representation can be used with WV meteorological satellite imagery, and in particular its value to handle scales and spatial derivatives. It proposes a new method to detect locations of interest, based on a difference of Gaussians, and shows that it performs better than other detectors, including those commonly used in operational AMV derivation schemes, in the experiments carried out with Meteosat-9 imagery. It proposes a group of distances, based on the Sobolev norm HI, which includes a term to evaluate partial derivatives, to quantify the similarity between neighbourhoods, and it shows that these distances produce better results than the widely used £2 norm, especially when the weight given to the derivatives term is relatively large.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available