Evidential network for multi-sensor fusion in an uncertain environment

Abstract

Interpreting and quantifying the confidence granted to signals transmitted and received in a sensor network is likely to be called into question by various factors. On an architectural plan, first of all, the nature of the networks or the distance between sensors can induce risk of false alarm or non-detection by misinterpretation of the analyzed signals. External factors related to stresses induced by the environment are also potential sources of measurement errors. Finally, despite the maturity of techniques, internal influence factors related to the accuracy or reliability sensors may also, at a more basic level, impact the confidence placed in the test or the performed diagnosis. A system-embedded intelligence is then necessary to compare the information received for the purpose of decision aiding based on margin of errors converted in confidence intervals. In this paper, we present three complementary approaches to quantify the interpretation of signals exchanged in a network of sensors in the presence of uncertainty.