Increasingly correct message passing algorithms for heat source detection in sensor networks

Solving complex source inference and estimation problems in the distributed environment of sensor networks is a difficult task. Data is distributed, as is computational power, and energy is limited. We consider the problem of detecting and locating a heat source appearing in a region monitored by a sensor network. This task requires complex computations that must be shared by all sensors. One significant difficulty we overcome is the problem of local minima. By using a two step procedure, where in the first step the nodes estimate their distances to the source, and in the second step we localize it and avoid the problem of having erroneous location estimates of the local minima. Furthermore, to organize the computations involved, we draw on ideas from graphical models. We develop an algorithm that has a property which we call "increasing correctness" that at any time the algorithm can be stopped and it nevertheless provides the correct answer for the problem defined by the information that has been fed into the algorithm up to that time.