Multi-sensor data fusion method based on divergence measure and probability transformation belief factor

Abstract

Dempster–Shafer evidence theory is widely used in multi-sensor data fusion. However, how to manage the counterintuitive result generated by the highly conflicting evidence remains an open question. To solve the problem, a novel multi-sensor data fusion method is proposed, which analyses the credibility of evidence from both the discrepancy between evidences and the factors of evidence itself. Firstly, a new Belief Kullback–Leibler divergence is put forward, which evaluates the credibility of evidence from the discrepancy between evidences. Secondly, another credibility measure called the Probability Transformation Belief Factor is defined, which assesses the credibility of evidence from the evidence itself. These two credibilities are combined as the comprehensive credibility of evidence. Furthermore, considering the uncertainty of evidence, a new belief entropy based on the cross-information within the evidence is presented, which is applied to quantify the information volume of evidence and to adjust the comprehensive credibility of evidence. The adjusted comprehensive credibility is regarded as the final weight to modify the body of evidence. Finally, the Dempster’s combination rule is applied for fusion. Experiment and applications show that the proposed method is effective and superior.