Evidence processing with empirical belief functions

A data-driven method for combining evidence from multiple sensors is presented. Empirical functions are used to compute a set of belief values for each sensor. These functions contain information about the degree of belief in the presence of an object as well as the uncertainty about the belief. The belief values are then combined using Dempster´s rule of combination. The empirical belief functions can be designed to take into account signal-to-noise characteristics and detection limits. Hard sensors that produce a yes/no output can also be modeled. Some advantages of this approach over sequential logic or pattern recognition are greater robustness with respect to faulty or inoperative sensors and more modularity