Adaptive sensor fusion architecture through ontology modeling and automatic reasoning

This paper presents a solution for implementing context-based self-adaptive sensor fusion systems. The adaptation process works over an ontology-based description of the problem space that includes sensors and other information sources, a repository of algorithms, and data types managed by the fusion system. An automatic reasoning module integrates this description with contextual information of the system, and determines how to combine available solution elements, to produce a fused output that best satisfies the goals of the system. Our proposal keeps the system working in the best conditions under events that include (a) intermittent sensor availability, (b) changing fusion requirements and (c) uneven information quality. Compared with existing proposals, our solution provides a generic mechanism to integrate arbitrary external factors in the adaptation process, such as context-related events, constraints and specific knowledge about the algorithms. We present an example on ground vehicle navigation, which combines on-board sensors with those available in a smart-phone.