D-S证据理论的改进算法在时-空信息融合中的应用

Improved Temporal-Spatial Fusion Based on D-S Evidence Theory

研究基于D-S证据理论的时-空信息融合,即多传感器多测量周期的信息融合,给出了三种时-空信息融合的方法:集中式、分布式无反馈和有反馈的融合算法。采用这些传统的时-空信息融合方法时,最终的融合结果会产生概率分配过分集中的现象,而且证据冲突时还会产生有悖常理的结果。本文在传统证据理论的合成公式的基础上给出了一个有效的合成规则,并由此提出了改进的融合方法。证据冲突的概率按照各个命题的平均支持程度加权进行分配,从而提高了融合结果的可靠性与合理性。

Based on temporal-spatial information fusion of Dempster-Shafter (D-S) theory, i.e. multi-sensors with different measurement periods, this paper developes three fusion approaches including the centralized fusion approach, distributed-without-feedback and distributed-with-feedback algorithms. In the traditional methods, the final probability assignment is excessively convergent. Furthermore, when evidences conflicting, there is a contrary result. In order to solve the problems, this paper presents an efficient combination rule. Depended on this rule, the improved temporal-spatial information fusion methods are provided. In these methods, the conflicting probability of evidences is distributed to every proposition according to its average supported degree, thus improving the reliability and rationality of the result.