Managing context data for smart spaces

We describe our on-going efforts to construct a service infrastructure to s upport smart environments. We characterize "fusion services," which extract and infer useful context information from sensor data, using evidential re asoning techniques. We specify sensing services as Bayesian networks and us e information theoretic algorithms to optimize the resources consumed by th e rendering of a service. We define a "Quality-of-Information" metric to ch aracterize sensing service performance. We have implemented an infrastructu re for supporting a dynamic set of sensors and services in a smart space. U sing this infrastructure and an IEEE 802.11 network, we implemented a proba bilistic indoor location system that optimizes the number of sensors consul ted when determining the location of a user while maintaining a high degree of accuracy.