Object recognition on the sea-bottom using possibility theory and sonar data fusion

摘要

The purpose of the proposed study is the recognition of objects laying on the sea-bottom where the images are acquired with a sonar device. The adopted conceptual framework consists in having different views to cooperate through the recognition rpocess. Informaiton gathered by different knowledge sources from the same scene are often uncertain, imprecise or incomplete. In this paper, we described a data fusion scheme base don the possibility theory. Such an approach prooved to be relevant to deal with the problems of managing uncertainty and combining measurements. However,due to the severe behaviour of the underwater environment and the nature of the sonar device itself, conflicts may arise among pieces of evidence. To cope with the lack of consistency among the sources, a revision of the possibility distribution was achieved. This belief revision involved risk factors computation to take a wrong decision from the data fusion results, and a hypothetical reasoning resting onthe assumption of each class presence on the sea-bottom. The experimental results displayed the sahpe distorsion robustness of the method and its capacity to successfully recognize objects even when one view is unexploitable.