Skip to Main Content
In this paper, the problem of distributed dynamic event region detection is studied by using a wireless sensor network. The spatiotemporal relationship of the evolving regions is assumed and modeled by dynamic Markov random fields. Conventional quickest change detection methods such as the CUSUM test do not take into account the spatiotemporal correlation information and therefore suffer a performance loss. By utilizing the system dynamics to predict the field evolution, a distributed event region detection algorithm using mean field approximation is proposed and examined. To further improve the detection performance especially for events that rarely occur, a hybrid detection scheme is introduced which incorporates the CUSUM test at the initial stage of the detection algorithm. The performance of the proposed algorithm is evaluated and compared using synthetic data from a crack simulation. The results demonstrate the effectiveness of the proposed algorithm and its superiority when exploiting the system dynamic model information.