A model for evaluating data fusion systems

Data fusion systems are being proposed for a wide variety of applications around the world to obtain the maximum benefit from expensive sensors and information systems. Examining surveys of data fusion work, it is apparent that while much activity has been devoted to implementing Level 1 systems with rigour derived from existing knowledge, implementation of Level 2 and 3 systems is often ad-hoc. Various testbeds are being developed to evaluate methods at all Levels. Methods of testing them are needed. This paper attempts to make a contribution to the general problem of performance measures for data fusion systems by presenting a general model. The model describes the data structures, processing stages and measures from generalised information theory. This theory includes measures of imprecision and ambiguity that are not represented by probability theory. Some intuitive results are shown to be qualitatively predicted.<>