The Fusion of Decisions for Distributed Recognition

Abstract

Data fusion is now a widely accepted approach for increasing the performance of data analysis when the data sources are distributed (see [1] or [2] for example). The combination of the opinions from multiple experts (or sensors) is an attractive and conceptually simple problem in multi-sensor data fusion. The task arises in a diversity of situations, from non co-operative target recognition [3] to medical diagnosis [4] and machine condition monitoring [5]. In each case the underlying problem is similar — how should decisions from two or more separate classifiers be combined to provide a fused decision of higher quality? At this stage we intentionally leave the measure of quality undefined, since accuracy, timeliness or robustness (amongst others) will each play a part depending on the specific application.