Context-Aware Sensors

Abstract

Wireless sensor networks typically consist of a large number of sensor nodes embedded in a physical space. Such sensors are low-power devices that are primarily used for monitoring several physical phenomena, potentially in remote harsh environments. Spatial and temporal dependencies between the readings at these nodes highly exist in such scenarios. Statistical contextual information encodes these spatio-temporal dependencies. It enables the sensors to locally predict their current readings based on their own past readings and the current readings of their neighbors. In this paper, we introduce context-aware sensors. Specifically, we propose a technique for modeling and learning statistical contextual information in sensor networks. Our approach is based on Bayesian classifiers; we map the problem of learning and utilizing contextual information to the problem of learning the parameters of a Bayes classifier, and then making inferences, respectively. We propose a scalable and energy-efficient procedure for online learning of these parameters in-network, in a distributed fashion. We discuss applications of our approach in discovering outliers and detection of faulty sensors, approximation of missing values, and in-network sampling. We experimentally analyze our approach in two applications, tracking and monitoring.