Dynamically clustering localization mechanism for wireless sensor networks

A new localization method is proposed for wireless sensor networks. The method uses iterative localization algorithm and distributed clustering scheme to decrease generated communication packets and increase the locating accuracy. The clustering scheme, named virtual cluster shift (VCS), makes self-organized sensor nodes of a wireless sensor network dynamically build one or more virtual clusters to localize and track mobile targets. When one or more sensors detect a target, a sensor node is selected to be the cluster head. The cluster head is in charge of organizing neighboring sensors to form a virtual cluster, fusing data and estimating the target's states such as location, velocity of mobile target. When the target is moving out of the monitor area of current virtual cluster, the cluster head gives place to one of his cluster member who is the nearest to the target. The new cluster head forms his virtual cluster by replacing some of the old cluster members. A mean shift based algorithm is used to iteratively estimating the target's position using data gathered inside virtual cluster. An optimal weight of the kernel function is given to minimize the variance of estimation. Analysis and simulation results show that the communication overhead and jitter of the proposed algorithm is less than 1/3 of the centralized algorithm.