ABSTRACT
It is a fascinating fact of nature that uncertainty in one or other form abounds in many aspects of the sensing, data communications, data analysis and data fusion (DF) processes. We should provide an explicit measure of this uncertainty to enable sensory information to be fused so that the data fusion process (DFP) becomes very efficient and the state of the plant/dynamic system or of any object in question or study (for the purpose of fusion) is predictable with enhanced accuracy (of prediction). Most such methods of representing uncertainty hitherto were based on the use of probabilistic models, because these models provide a powerful and consistent means of describing uncertainty in many situations (also these probabilistic models occurred at a much earlier time in the evolutionary/developmental frame of scientific search in the human mind). This concept naturally fits into the ideas of information fusion and decision making [1]. Of course, certain practical realities of a problem often might suggest the use of some alternative method of modelling uncertainty (and/or modelling different kind of uncertainties e.g. vagueness), for example leading to fuzzy logic (FL)-based concepts and their utilisation in filtering, control and decision fusion problems. Although human perception and experience of these kind of fuzzy events date much more into the past (having had occurred many centuries earlier than gambling and other probabilistic matters), the formal mathematics and analytical approaches had not been developed until five decades ago! Although probabilistic models are good for many situations, these models cannot capture all the information required to define and describe the operations of sensing and DF processes, for example, the heuristic knowledge of a human expert in a domain area/knowledge, for which the theory of FL and related possibility theory are very suitable. Yet, probabilistic modelling techniques have played and continue to play a good and very practical role in developing DFP and methods. Hence, it is necessary to understand the probabilistic modelling methods for DF applications. Also, when uncertainty is not random,
but it is deterministic and unknown (and has a bounded magnitude) some approaches based on the H-infinity norm can be considered for DF tasks, especially for basic state determination from the measurement data observed from the sensors.
