Impact on Accuracy of Deployment Trade-Offs in Localized Sensor Network Event Detection

Event detection in sensor networks is a very desirable capability, insofar as it can be seen as a building block for higher-level functionalities. Of particular interest in this paper is the ability to characterize the geometry of physical phenomena (e.g., a weather front, or a forest fire) as they take place in the sensing scope of a deployed network. If such geometries can be accurately characterized in a cost-effective manner, then we can construe them as induced geometrical extents. Taken together with asserted geometrical extents (e.g., those of civil structures such as buildings and roads and bridges, or of geographical features, such as forests, rivers, lakes, mountains), physical phenomena can then be targetted by high-level, declarative spatial and spatio-temporal queries. For this vision to be realized, one must understand better the conditions under which best-ofbreed event detection algorithms for sensor networks perform well on realistic geometries given realistic assumptions about deployment circumstances and operating conditions. In this paper, we contribute a study of two event detection algorithms, viz., FEBD and T-Fit. The study revealed shortcomings in both algorithms, which we used to develop improved versions of both. We have quantified the trade-offs between accuracy, areal coverage and sensor density that they incur. Our overarching goal has, therefore, been a practical one, viz., to inform future deployments as to what accuracy can be expected, given a desired areal coverage and such sensor density as can be afforded, by the use of the algorithms under evaluation.