Properties of Data Fusion

The properties of information fusion are listed: fusion of attributes, fusion of analysis and fusion of representations.

FUSION OF ATTRIBUTES

Assume the sources of information are aligned and associated. Fusion of attributes consists in merging the attributes of a same object, derived from two representations (X_S(1))^t and (X_S(2))^t at instant t obtained by means of the sources of information S(1) and S(2), in order to obtain new attributes in the space of sources S = S(1) U S(2).

FUSION OF ANALYSIS

Assume the sources of information are aligned and associated. Fusion of analysis consists in aggregating representations (X_S(1))^t and (X_S(2))^t, into a new representation (X_S)^t, then in generating an analysis or interpretation of the object for further use at instant (t+1), or at step i in an iterative process.

FUSION OF REPRESENTATIONS

Fusion of representations is defining and performing meta-operations applicable to representations (X_S(1))^t and (X_S(2))^t, to obtain a new representation (X_S)^t. Fusion of representations includes fusion of decisions. This fusion of representations may be performed at any moment, i.e. combined with other types of fusion.

This implies that fusion may operate at any of the three semantic levels: measurements (fusion of measurements), attributes (fusion of attributes) and rules (fusion of decision or rules), with possible crossings between levels. This property is not fully expressed in the literature using the JDL model, as already discussed. Like the two other properties, it impacts on the design of the architecture of a fusion system, on the selection of tools, suite of softwares and hardware (processing issues), communications (topological issues) and on the design of innovative procedures.

Reference: L. F. Pau. Sensor data fusion. Journal of Intelligent and Robotics Systems, vol. 1, pp. 103-116, 1988.

31-08-2001 - Copyright L. Wald, Armines / Ecole des Mines de Paris