Self-stabilizing Distributed Data Fusion

The Theory of Belief Functions is a formal framework for reasoning with uncertainty that is well suited for representing unreliable information and weak states of knowledge. In information fusion applications, it is mainly used in a centralized way, by gathering the data on a single node before computation. In this paper, a distributed algorithm is proposed to compute the neighborhood confidence of each node, by combining all the data of its neighbors using an adaptation of the well known Dempster’s rule. Moreover, a distributed algorithm is proposed to compute the distributed confidence of each node, by combining all the data of the network using an adaptation of the cautious operator. Then, it is shown that when adding a discounting to the cautious operator, it becomes an r-operator and the distributed algorithm becomes self-stabilizing. This means that it converges in finite time despite transient faults. Using this approach, uncertain and imprecise distributed data can be processed over a network without gathering them on a central node, even on a network subject to failures, saving important computing and networking resources. Moreover, our algorithms converge in finite time whatever is the initialization of the system and for any unknown topology. This contribution leads to new interesting distributed applications dealing with uncertain and imprecise data. This is illustrated in the paper: an application for sensors networks is detailed all along the paper to ease the understanding of the formal approach and to show its interest.