An Possibility Fusion Method Based On The Uncertainty Of Multi-source Information

The multi-source information fusion system usually has the uncertainty, caused by thecomplexity of battlefield environments, the limitations of the sensor performance andsubjective cognition, as well as the imperfection of the way of information acquisition andprocessing. The uncertainty exists in many aspects of information fusion, such asmeasurement, fusion model and parameters etc., which seriously affects the precision andaccuracy of the fusion results.In the traditional handing of uncertainty, probability theory can only describe and processrandom uncertainty, also it is difficult to obtain the priori and conditional probability inBayesian methods. The complexity of the focal elements’structure in belief functionincreases when the reasoning steps increase in D-S evidence theory. Possibility theory cancharacterize and quantify the uncertainty properly with the lower amount of information andcomputational complexity, so it provides a new approach for information fusion.In this paper, a new possibility fusion method on multi-source information is presented bymean of possibility theory and information fusion theory, with the background of groundtargets. On the basis of analysing the uncertainty sources and essence, a possibilitydistribution construction method of small samples of measurement information of the sensorsis put forward. The possibility measurement system of multi-source information uncertainty isset up. The fuzzy information uncertainty can be expressed by possibility distribution, andfuzzy sets consisting of multi-source information can be measured by possibility measure andnecessity measure, as well as the uncertainty can be transformed into the measures of setsfrom the distributions in point process. Then the track association method and targetidentification recognition method are presented respectively based on possibility theory, andthe distributed asynchronous information fusion method is proposed, for measurementinformation from different detection time and sensor types, in order to obtain the precise stateestimation and the accurate identification recognition. In the medium-density environment, the typical simulation scene is designed fordemonstrating the effectiveness of fusion method. The simulation results show that theaccuracy of the possibility track association can improve by 7% compared with classicsequential method, and 13% compared with K-neighbor domain method. The RMSE ofasynchronous information fusion method reduces by 30% ~ 50%, compared with thearithmetic mean of predicted data, significantly improving the tracking precision. When thestandard deviation of measurement error is 3%, the average rate of target identificationrecognition based on possibility theory can reach above 82%, and improves by 8% and 15%compared with D-S evidence theory and Bayes method respectively.