A sliding window approach for dynamic event-region detection in sensor networks

Event region detection is an important problem in wireless sensor networks. However, a fundamental assumption required for the event-region detection schemes proposed in the literature is that the event-region under detection is quasi-static, i.e., the event-regions do not change during the detection period. This assumption could fail in detecting fast varying events, such as forest fire events and mudslides. Therefore, the detection task considered in the present study is to cope with the continuously varying event regions. Specifically, we adopt a sliding window approach, in which only the sensor measurements from the present to a fixed length time ago are utilized for making decision at any particular time slot. Our simulation results demonstrate the advantages of the proposed detection scheme.