Evidence combination based on credibility and separability

D-S evidence theory has been widely used in various fields of information fusion due to its efficiency in dealing with uncertain information. Unfortunately, combination of conflicting evidences with the classical Dempster's rule may produce the counter-intuitive results. In this paper, Definitions of correlation coefficient and credibility are first presented, followed by the introduction of evidence separability. A modified weighted factor determined by credibility and separability is then presented, which is then applied to average the basic probability assignment (BPA). Finally, the modified BPA is combined following the D-S rule. It is demonstrated the more reasonable combination results can be obtained by this method. Moreover, better convergence performance can be obtained.