Semi-dynamic approaches to node clustering for sensor networks

This work studies the problem of node clustering for wireless sensor networks. In this context, clustering is intended as the process of electing leader nodes and partitioning the remaining nodes among the leaders. Clustering is needed in any tasks and applications for sensor networks that require some form of locally centralized processing. This work proposes a graphical model that represents node clustering as a graph cutting problem. This model is considered in different application contexts. In order to deal with a reduced number vertices and edges, a semi-dynamic clustering strategy is proposed: i.e. new clusters are formed over an a priori defined sensor partitions. Existing clusters are split into subclusters A cluster or a portion of it is just viewed as a macronode. Finally an heuristic implementation of the above ideas is applied to a multiple target tracking scenario in particular to the tasks of multiple target counting and localization.