Tracking and Characterization of Highly Deformable Cloud Structures

Abstract

Tracking and characterizing convective clouds from meteorological satellite images enable to evaluate the potential occurring of strong precipitation. We propose an original two-step tracking method based on the Level Set approach which can efficiently cope with frequent splitting or merging phases undergone by such highly deformable structures. The first step exploits a 2D motion field, and acts as a prediction step. The second step can produce, by comparing local and global photometric information, appropriate expansion or contraction forces on the evolving contours to accurately locate the cloud cells of interest. The characterization of the tracked clouds relies on both 2D local motion divergence information and temporal variations of temperature. It is formulated as a contextual statistical labeling problem involving three classes “growing activity”, “declining activity” and “inactivity”.