Event detection using sensor networks: A case for a hybrid detector

This paper investigates the problem of event detection using a Wireless Sensor Network (WSN). Specifically, it investigates three detectors. Firstly, the Mean Detector (MD) where the test statistic is the sample mean of each sensor node by itself. Secondly, the Covariance Detector (CD) that evaluates the sample covariance between pairs of neighboring sensor nodes. If the estimated sample covariance is above a threshold, then the CD reports that an event is present. Finally, the Hybrid Detector (HD) where each sensor decides independently between the MD and the CD based on the distance from its closest neighbor. Extensive simulation results are also presented that compare the performance of the detectors. The main contribution of this paper is to show that when the sensor nodes are located close to each other the CD can exploit possible correlation between their measurements to achieve significantly better detection. In other situations, when measurements do not exhibit spatial correlation or when a sensor node is isolated from its neighbors the MD is the best choice. The idea of the HD is to exploit the advantages of the two algorithms.