Weather satellites play a key role in modern meteorology and give an undoubted contribution to a correct weather forecast. Image sequences transmitted by geostationary satellites like METEOSAT permit to track the temporal evolution of typical clouds structures (vortices, fronts) associated to depression phenomena. A fully automated system for clouds tracking is an important challenge to image processing applied to meteorology. This paper proposes an enhancement of modal matching techniques as a more suitable alternative to optical flow-based or parametrical methods. Modal matching permits a robust point-feature association among strongly deformed shapes, as well as a highly detailed description of complex shapes. Modal expansions of virtual elastic bodies associated to the shapes provide coarse-to-fine descriptions of them: pairing homologous low frequency modes and discarding noise and sampling error affected high frequency ones gives a robust basis for vibration-based point correspondence. The classical technique requires dense sampling for strong modes similarity: this paper tackles the problem of compensating the degenerative effects of low sampling by efficiently controlling modes pairing and by forcing the locality constraint through balls tracking the contours: since mismatch probability is higher in particular localized areas, balls actually work as a shield, letting modal match perform reliably inside them.

Date Published: 16 September 1999
PDF: 12 pages
Proc. SPIE 3821, Environmental Sensing and Applications, (16 September 1999); doi: 10.1117/12.364197

Published in SPIE Proceedings Vol. 3821:
Environmental Sensing and Applications
Michel R. Carleer; Moira Hilton; Torsten Lamp; Rainer Reuter; George M. Russwurm; Klaus Schaefer; Konradin Weber; Klaus C. H. Weitkamp; Jean-Pierre Wolf; Ljuba Woppowa, Editor(s)