Optimizing Uncertainty in Dempster-Shafer Detectors Fusing Multi-Sensor Data

摘要

Fusion algorithms based on the Dempster-Shafer (DS) Theory of Evidence lack a universally standard method for automatically assigning probability mass to the 'don't know' hypothesis for a particular input. For example, when fusing automatic target detection (ATD) algorithm outputs from multiple sensors, one must associate a measure of uncertainty with the output from the ATD algorithm of each sensor. We describe such a fusion algorithm, developed using the DS formalism, and present a method for automatically determining the required assignment of uncertainty. We also evaluate the entire procedure using simulated data and receiver operating characteristic curves.