Constrained clustering for flocking-based tracking in maneuvering target environment

Abstract

Self-organizing ability is one of the most important requirements of modern sensor networks; particularly for tracking maneuvering targets. Flocking-based approaches are biologically inspired methods that have recently gained significant attention to address the control and coordination problem in self-organizing sensor networks. These approaches are exemplified by the two well-known algorithms, namely, the Flocking and the Semi-Flocking algorithms. Although these two algorithms have demonstrated promising performance in tracking linear target(s), they have deficiencies in tracking maneuvering targets.This paper introduces a constrained clustering approach that uses a novel extension of K-means algorithm to provide better coverage over maneuvering targets. This extension clusters the sensors based on certain background knowledge, then uses the information about the clusters to improve coverage. The performances of flocking-based algorithms, both with and without the proposed approach, are examined in tracking both linear and maneuvering targets. Experimental results demonstrate how constrained clustering yields better tracking of maneuvering targets, and how applying constraints on the clustering process improves the quality of clustering and increases the speed of convergence.