Uncertain Reasoning Using Time-Dynamic Markov Random Field for Sensor-Network Applications

Sensor networks are often used in environment monitoring. We consider uncertain reasoning in sensor network-based monitoring, in particular, in detecting and tracking plumes under heavy noise. We extend Markov random field to a new time-dynamic Markov random field (TD-MRF) and use it to model the environment. We provide an algorithm for inferring, based on TD-MRF, the plume situation in the environment given noisy sensor output. Our experiments showed that TD-MRF and our inference framework can lead to better detection and tracking results even when there is a high level of