Privacy-preserving event detection in pervasive spaces

In this paper, we consider privacy challenges in event-driven pervasive spaces where multimedia streams captured by sensors embedded in the infrastructure are used to detect a variety of application-specific media events. In particular, we develop techniques to detect events without disclosing any identifying information unless necessary. We characterize the nature of inference channels that arise and model privacy preserving event detection as an optimization problem that attempts to balance disclosure with performance. We design and test efficient communication protocols that realize this tradeoff.