Fusing light and temperature data for fire detection

Wireless Sensor Networks (WSN) are often designed to detect events based on measurements provided by sensor nodes. In this work, we propose and evaluate two algorithms, based on information fusion methods, for fire detection in rainforests when we have only light and temperature sensors. The idea is to have acceptable detection methods even when we do not have (or cannot afford) other sensor units that would lead to more accurate inferences, such as humidity sensors, smoke detectors or video cameras. In this sense, we collect and analyze real environmental data in situations that are similar to those found in real scenarios. Then, we create some controlled fire (events) of small proportion, close to sensor nodes, to force event occurrences. Experimental results indicate that forest fire can be detected at night by using just light measurements. In addition, light data can be used to distinguish heat resulting from sunbeam and fire. The first method is based on the moving average filter and thresholds (Threshold method), while the second uses the moving average filter and the Dempster-Shafer theory. Both methods present similar results regarding the detection efficacy considering the generated instances. However, the Threshold method is more stable, due to the use of state machine to represent the environment conditions.