Researches On Conflicting Evidences In Data Fusion

The study of effective combination rules in DS theory when evidences are in conflict is always an interesting topic. This research is concentrated on analyzing the degree of conflict among evidences in Multi-Sensor Information Fusion System and proposing the combination rules based on the TBM model. The major works are as follows:(1) Discussed what really constitutes a conflict among two beliefs, and proposed the new judgment of the conflict by using both the mass of the combined belief assigned to the empty set before normalization and the distance between betting commitments of beliefs. After analyzing the different fusion conditions defined by the two values, a set of rules guiding whether Dempster's rule can be used or cannot be applied are proposed.(2) Proposed the method of computing the general basic belief assignment used in the object detection in the open world. First, the triangle fuzzy number and the weight of all the properties are constructed based on the original data, then the general bbas of sensor reports which can be used in the information fusion of the open world are computed according to the triangle fuzzy numbers and their weights.(3) Proposed the fusion model and the decision rules of the open world based on the TBM model. According to the belief assigned to the empty set of a bba and the value of the conflict among evidences, the general bbas obtained from sensor reports are abandoned or combined or updated on the fusion center of the credal level, and then translated into probability of each object by using the pignistic transformation when needed. The decisions are made using the pignistic probabilities derived from the bba by the pignistic transformation.