default Context-based reasoning for maritime situation awareness


Context-based reasoning for maritime situation awareness. Jousselme, Anne-Laure. CMRE-FR-2015-022. December 2015.

The explicit consideration of context in information fusion systems offers the necessary flexibility and adaptability to generalise processes while at the same time improving the interpretation of their outputs. It does however, raise the challenges of adequate context definition and formalisation so that context provides a useful and appropriate contribution to the underlying processing framework. In this report, we propose a formalisation of context-based reasoning from an information fusion perspective, tying together the themes of source quality, uncertainty representation and measurement space versus decision space, all around the central notion of context. We first illustrate the fact that context is a relative notion on the sub-problem of Maritime Situation Awareness. The various levels of processing, the embedded problems, the different granularity levels, the required dynamic of the processing, as well as the place and role of the human are highlighted. The use of context in the literature from different domains (e.g. artificial intelligence, information fusion, natural language processing) is mapped onto the modelling and processing steps. Information and source quality are placed at the core of the reasoning process for a sound consideration of the different quality dimensions and an appropriate representation and processing of uncertainty. The proposed scheme in this report is represented as a Direct Acyclic Graph where context is a central variable influencing the other variables of source quality, of measurement, of decision, of measurement and decision spaces, and of information gathering policy. The proposed scheme is simple enough to provide an appropriate level of abstraction with only 6 compound variables. It is general enough to be applied to different application domains, in particular from information processing to information gathering and source tasking. The proposed framework offers the required generalisation and flexibility of a problem-solving method which can be tuned and adapted to dynamical contextual information. An example of implementation using a Bayesian network is given, and its extension to other uncertainty representations is explored. A Maritime Anomaly Detection problem is illustrated through the proposed approach. We illustrate how the context can then be used to adapt information processing on the fly by changing granularity of the problem or correcting the information based on sources’ performance assessment.