Sparse analysis for mesoscale convective systems tracking

In this paper, we study the tracking of deformable shapes in sequences of images. Our target application is the tracking of clouds in satellite images. We propose to use a recent state-of-the-art method for off-the-grid sparse analysis to describe clouds in image as mixtures of 2D atoms. Then, we introduce a method to handle the tracking with its specificities: apparition or disappearance of objects, merging, and splitting. Numerically, this method corroborates the magnitude of the results provided by recent state-of-the-art alternatives. Unlike its counterparts, the choice or regularization and correlation parameters allows additional flexibility regarding the interpretation of clouds’ life cycles. Finally, it also provides additional information on the cloud temperature during its life cycle, which seem in accordance with the underlying physical processes.