The task of sensory data fusion may involve the aggregation of sensory
measurements that may be from different phenomenological domains and
that, in many cases, could embrace some conflicting information cues.
It is rather a challenge to find suitable strategies by which measurem
ents obtained by the different sensors of the system can be aggregated
so that a consistent interpretation of these measurements is achieved
. In this article, we present a novel approach to achieve this goal. A
recursive group utility function that is capable of bringing the grou
p of sensors into consensus is used. After each sensor in the group ga
thers information relevant to the sensory task, the group engages in w
hat we call the uncertainty estimation stage. This is an information t
heory-based process that allows each sensor to assess its self-uncerta
inty and the conditional uncertainties of other sensors. This process
facilitates the computation of a weighting scheme that operates recurs
ively on sensor observations until the group reaches a consensus. When
ever new observations are made, the uncertainty estimates of sensors a
re updated and used to compute a new weighting scheme. To demonstrate
the efficacy and to show how the methodology works, the article discus
ses how the method can be used to tackle the multi-sensor object ident
ification problem. (C) 1993 John Wiley and Sons, Inc.