The Effects of Pre-Fusion Probability Calibration

Quite often the state of information provided by disparate sources, or experts, is not conducive to fusion without additional pre-fusion processing. The focus of this paper involves testing the effects of two pre-processing steps to help illustrate the contribution of each to performance, and also testing two different fusion methods to illustrate how robust each are against lacking/imperfect pre-processing steps. The pre-processing steps, which are the focus of this study, are two calibration steps. The first calibrates overconfident expert information, and the second calibrates for track-to-track association ambiguity. The two fusion methods tested areWeighted Conflict Redistribution (WCR) Rule number 0-or WCR0 for short, and is also known as Dempster's rule-and WCR2. WCR0 continues to produce the optimal fusion solution when both pre-processing calibration steps are performed perfectly, and yet, it is shown to be significantly less robust given significant miscalibration in the source probabilities.