Database Reference
In-Depth Information
Another event-based technique based on threshold is Approximate
Caching [31] whereby nodes only report readings if they satisfy a condi-
tion. A more recent paper [32] suggests a mixture of hardware and soft-
ware as a solution for detecting rare and random events. The event types
they consider are tracking and detecting events using the eXtreme Scale
Platform (XSM) motes equipped with infrared, magnetic and acoustic
sensors. Central to their architecture is the concept of passive vigi-
lance, which is inspired from sleep states of humans where the slightest
noise can wake us up when we are asleep. This is implemented with
Duty Cycling and recoverable retask. A similar approach [33] proposes
a sleep-scheduling algorithm that minimizes the surveillance delay (event
detection delay) while it maximizes energy conservation. Sleep schedul-
ing is coordinated locally in a fair manner, so all nodes get their fair
share of sleep. A minimal subset that ensures coverage of the sensing
field is always awake in order to be able to capture rare events.
The earliest work that addresses the need for complex event detection
is the one by Girod et al [34]. It suggests a system that would treat a
sequence of samples (a signal segment) as a basic data type and would
offer a language (WaveScript) to express signal processing programs as
declarative queries over streams of data. The language would be able
to execute both on PCs and distributed sensors. The data stream man-
agement system (called WaveScope) combines event-stream and data
management operations.
REED [35] is an approach that falls under both the Event-Based and
the Query-Based subcategories. REED is an improvement on TinyDB
[36]. Basically it extends TinyDB with the ability to support joins be-
tween sensor data and building static tables outside the network. The
tables outside the network describe events in terms of complex predi-
cates. These external tables are joined with the sensor readings table,
and returned tuples that satisfy the predicates indicate readings of in-
terest, for example, where an event has occurred.
Abstract Regions [37, 38] is a somewhat different method that sup-
ports geographic grouping of sensor nodes. Abstract Regions is essen-
tially a family of spatial operators for TinyOS that allows nodes to
form groups with the objective of data sharing and reduction within
the groups by applying aggregate operators such as min, max, sum, and
others. The work by [39] extends the types of aggregates supported
by introducing (approximate) quantiles such as the median, the consen-
sus, the histogram and range queries. Support for spatial aggregation is
also suggested by [40] where sensor nodes would be grouped and aggre-
gates would be computed using Voronoi diagrams. Another approach
[41] models the sensor network as a distributed deductive declarative
