Spatial Statistical Data Fusion on Java-enabled Machines in Ubiquitous Sensor Networks

Abstract

Wireless Sensor Networks (WSN) consist of small, cheap devices that have a combination of sensing, computing and communication capabilities. They must be able to communicate and process data efficiently using minimum amount of energy and cover an area of interest with the minimum number of sensors. This thesis proposes the use of techniques that were designed for Geostatistics and applies them to WSN field. Kriging and Cokriging interpolation that can be considered as Information Fusion algorithms were tested to prove the feasibility of the methods to increase coverage. To reduce energy consumption, a compression method that models correlations based on variograms was developed. A second challenge is to establish the communication to the external networks and to react to unexpected events. A demonstrator that uses commercial Java-enabled devices was implemented. It is able to perform remote monitoring, send SMS alarms and deploy remote updates.