The understanding and principled exploitation of contextual information in fusion systems is still very limited. Domain knowledge is generally acquired from an expert and applied to stove-piped solutions that can hardly scale or adapt to new conditions. However, “context” can play a vital role at any level of a modern information fusion system: from object recognition through physical context exploitation, to intention estimation through linguistic communication analysis. In this work, a few important elements that should be considered in designing a context-aware system are discussed including: context refinement using terrain information, context to promote fusion results to higher levels of abstraction, and context for resource management. We highlight concepts of context sifting, shifting, and adaptation for multi-level fusion.
