When surveillance sensors disagree

An important application of data fusion is the integration of inputs from a suite of surveillance sensors. An accurate coherent picture of the real world situation must be built up from partial and sometimes conflicting interpretations of the real-world situation. The more complex the situation, the more difficult it is to piece the true picture together. We outline a formalism for maintaining representations of multiple versions of situations which are consistent with the information derived from multiple sensors. A key feature of our approach is its treatment of indeterminate data values. Our work recognises that an operator´s interest in a situation depends on what aspects of information are important to them. We define the information contained in a sensor input as a function of higher level interpretations and associations, rather than raw signal information in the sense of Shannon. We indicate how this view of information can be used to develop flexible notification strategies to reduce operator workload. Our framework is extensible to the even richer definitions of “interest” situations which will be required as sensor inputs become increasingly integrated into operational networks passing information up and down the hierarchy