Detecção de eventos em redes de sensores sem fio

Abstract

Event detection is a fundamental problem in wireless sensor networks. Typical applications related to event detection include, among others, medical and military developments, environment monitoring and presence detection. The goal of event detection is to identify when a collected data represents the occurrence of an event of interest. The lack of uniformity on treating this problem (such as the variations in modeling of data collection made by the sensor and of event, and the set of methods for its detection) does not allow the comparison among different proposals. This work presents a framework to detect events for scenarios with fault occurrence, measuring imprecision of the monitored characteristic performed by a non-ideal sensor device at an environment with background noise. For this purpose, we adapted the framework Diffuse to the case of event detection, and evaluated it with a new proposed method based on Control Charts Theory. The framework Diffuse was initially proposed to detect sensor faults in a scenario of constant data streaming over a wireless sensor network using traffic metrics. The adapted version of the Diffuse framework consists of four elements: (i) Measurement Processor, which is a module to filter the collected data and detect faults; (ii) Event Processor, which is used to estimate the occurrence of a given event from a set of monitored characteristics; (iii) State Estimator, which infers the occurrence of events given the estimations of occurrence of neighborhood, and (iv) Decision Maker, which chooses sending an event occurrence notification to the sink (in case it has been detected). Simulation results performed with the Sinalgo simulator showed that: (i) for 10% of sensor faults (i.e., when a sensor device measures a wrong data from the environment) our framework implementation has approximately 96% of event detection hit rate; and (ii) for 30% of sensor faults, our framework implementation has 75% of event detection hit rate. The contribution of this work is to provide an adapted version of the Diffuse framework that covers event detection methods based on estimation and inference theories using inaccurate collected data providing a way to compare event detection implementations. Furthermore, we also provided a set of default models to compare event detection methods and two new methods: one based on Control Charts Theory and another one based on Confidence Interval.