Abstract:

In the Dempster-Shafer theory based target recognition fusion system, the sensor reliability evaluation and the evidence discount are the key problems for the combination of multisensor evidence using the Dempster’s rule. A method for sensor dynamic reliability evaluation is proposed according to the conflict between the current sensor’s output evidence and all sensors’ averaged evidence. Then, using the idea of Dubois & Prade rule in conflict processing for reference, an evidence discounting rule is proposed, which distributes the discounting belief mass to the union set of conflicting focal elements according to the size of their local conflict respectively. Simulation and real data experiment  results demonstrate the validity of the proposed scheme.