A method for modelling and tracking convective clouds within radar images is described. Clouds are non-rigid heterogeneous objects; a good representation of the weather scene is performed by extracting skeletal lines of the 2-dimensional grayscale pictures. Skeletons are reduced to sets of segments within a graph structure and tracked among successive pictures by means of relaxation labelling processes. We present preliminary results of our method on airborne weather radar data. The conventional morphological skeleton constraint -i.e., to be centered in relation to the object boundaries- is here replaced by attachment to the gray levels local maxima. The resulting skeletal line is forced to go through each intense convective cell, and links neighboring ones into a global structure. Moreover, each feature point can be labelled and enriched by meteorological information. Such pattern is particularly suitable for weather nowcasting, since the user (for example, an aircraft's crew) is specially interested in the convective cells evolution.
展开▼