Research Of Information Fusion Algorithm Based On Evidence Theory In Open Frame Of Discernment

With the appearance of various of intelligence device,information fusion applied to intelligence device become more and more important,the more complex the environment under which sensors work is,the lower the degree of accuracy and the reliability of data retrieved from sensors are.So it is a difficult problem to obtain useful information by dealing with these inaccurate data of which degree of reliability we do not know.Evidence theory is a powerful tool for uncertainty reasoning,but counter-intuitive result is usually obtained while dealing with highly conflicting evidences.Facing the problem,it can be know by analysis that both of exiting unknown state and reliability of data retrieved from sensors are probably the reasons.For acquire a better result by handling these two reasons,combining information theory with evidence theory,an approach that is based on information entropy,conflicting degree and priori-knowledge to modify the frame of discernment in advance and a method that is based on open frame of discernment to fuse evidences are presented.The main contents of this paper include the following three aspects:(1)An approach is proposed based on information entropy,priori-knowledge and conflicting degree between evidences.Considering the problem with frame of discernment,firstly,that there might be a unknown state in the frame of discernment could be inferred according to Smets' open world assumption.Secondly,a equation set is presented based on the relationship between information entropy,conflicting degree and priori-knowledge.Finally a new basic belief function is presented by solving the equation set and the experiments with comparison among two sides show that this approach is reasonable.(2)A new combination rule is presented in open frame of discernment.Considering the problems with sensors,firstly,define the reliability between evidences,the concrete steps that is to calculate similarity degree using supporting degree based on Jousselme distance and normalize supporting degree to acquire it.Secondly define certainty of evidence,then define synthesis reliability synthesizing reliability between evidences with certainty as weight to acquire weighted evidences.Finally use D-S combination rule fuse weighted evidences n-1 times.After the over steps,not only does method of this paper inherit the D-S rule complete mathematical properties but also it comprehensively takes account of that two reasons.(3)The combination consequence demonstrate that the method presented in this paper is effective and robust based on simulation,which using data about car recognition does cross comparison between methods in closed world assumption and open world assumption respectively.