Distributed tracking systems and their optimal inference topology

Abstract

This paper addresses the "distributed tracking" problem, or the problem of local integration of estimation communicated by different sensors through a fixed multicast communication topology. The recursive nature of this shared information which can be delayed by communication links suggests a careful integration because of the cross correlation in the estimation errors. The work of Durrant-Whyte and Grime (1992) has proposed channel filters for a multicast tree topology. Other authors have proposed the utilisation of Covariance Intersection to solve the distributed tracking problem for any topology. In this paper one explore the possibility of extending the channel filtering principle to any topology by constraining the number of shortest path.