On the Graph Building Problem in Camera Networks

In this paper, the problem of building a model of a sensor (camera) network from observations is considered. By model, we mean a graph where the nodes represent states that are observable and distinguishable by the sensor network and edges are the feasible transitions among these states: the edges are also weighted by the probability of transition from one state to another. Remarkably, since merely static observations are not sufficient to discern all states in the networked system, the dynamics of transition is also considered. In this respect, the proposed graph model appears falling into the class of hidden Markov models, where the discovery of hidden states is made possible by exploiting the temporal evolution of the transitions and the implementation of a splitting procedure of previously identified graph nodes.