Modeling and Information Processing

TAF - Advanced Fusion Techniques

Objective

To study innovatory theoretical concepts that will extend the area of formalization of the problems posed by data fusion.

Onera's contribution

The upstream work in data fusion carried out recently at the DTIM mainly concerns the development on new fusion rules in the framework of the DSmT (Dezert-Smarandache Theory of plausible and paradoxical reasoning). Several rules, called PCR rules (Proportional Conflict Redistribution), based on the proportional transfer of the credibility weightings of the conflicting data sets have been proposed as alternatives to the hybrid rule of direct transfer (DSm hybrid rule) in order to improve the accuracy and pertinence of the data combination results, and allows us to work beyond the application limits of the fusion rules proposed in the framework of the theory of evidence proposed in 1976 by Glenn Shafer. The advantages of the new rules have been demonstrated on multi-target tracking scenarios using kinematic and attribute type data at the same time. The DSmT is both a new paradigm for data fusion and a natural extension of the DST (Dempster-Shafer Theory) which allows for the numerical fusion of belief functions associated with data sources that are uncertain, imprecise and may be highly conflictual. The work currently in progress concerns the fusion of data whose belief functions are now expressed in qualitative terms (Granular computing with words) rather than numerical, in order to take human expertise expressed in natural language into account more easily in the architectures of complex fusion systems.

You can find the latest advances in DSmT at: http://www.gallup.unm.edu/~smarandache/DSmT.htm

Illustration

General principle of fusion by DSmT

Last Update: - © ONERA 2009 - Terms of use