Active Fusion using Dempster-Shafer Theory of Evidence

Image understanding applications are often tainted with a high degree of complexity, uncertainty, and imprecision. The large amount of data makes it necessary to select the most useful information. The active fusion system proposed in this paper is able to eeectively select information sources, to control the acquisition process, to select processing strategies, to integrate results, and to draw a decision. It is implemented in the framework of Dempster-Shafer evidence theory, which is able to cope with uncertain data. A sample application is given for the classiication of crop categories in agricultural areas. This experiment shows a signiicant reduction of required information sources. Finally, we discuss advantages and disadvantages of Dempster-Shafer evidence theory for active fusion and compare evidence theory with Bayesian and fuzzy modeling.