Learning Sensor Network Topology through Monte Carlo Expectation Maximization

Abstract

Abstract-We consider the problem of inferring sensor positions and a topological (i.e. qualitative) map of an environment given a set of cameras with non-overlapping fields of view. In this way, without prior knowledge of the environment nor the exact position of sensors within the environment, one can infer the topology of the environment, and common traffic patterns within it. In particular, we consider sensors stationed at the junctions of the hallways of a large building. We infer the sensor connectivity graph and the travel times between sensors (and hence the hallway topology) from the sequence of events caused by unlabeled agents (i.e. people) passing within view of the different sensors. We do this based on a firstorder semi-Markov model of the agent's behavior. The paper describes a problem formulation and proposes a stochastic algorithm for its solution. The result of the algorithm is a probabilistic model of the sensor network connectivity graph and the underlying traffic patterns. We conclude with results from numerical simulations

Assigning cameras to subjects in video surveillance systems

“…Interestingly, the inverse problem has also been attempted. Marinakis et al [17] infer the relative positions of a set of sensors by observing the motions of objects between their non-overlapping fields of view.…”

Unsupervised camera localization in crowded spaces

“…Pioneering works [19], [27], [24], [4] have shown encouraging results towards our goal of locating cameras in constrained settings. They assume that the correspondence between an observed pedestrian from one view is known across a different view, e.g., a single person moves around the scene.…”

Topology Inference for a Vision-Based Sensor Network

“…The technique assumes knowledge of the number of agents in the environment and attempts to augment the given observations with an additional data association that links each observation to an individual agent. (See Marinakis, Dudek, and Fleet [7]. )…”

“…In previous work [7], we presented and verified, through numerical simulations, a network topology inference method based on constructing plausible agent trajectories. The technique employed a stochastic Expectation Maximization (EM) algorithm, an established statistical method for parameter estimation of incomplete data models [2] [15] that has been applied to many fields including multi-target tracking [9] and mapping in robotics [1,12].…”