A Graphical Model Approach to Source Localization in Wireless Sensor Networks

Collaborative localization and discrimination of multiple acoustic sources is an important problem in Wireless Sensor Networks (WSNs). Localization approaches can be categorized as signal-based and feature-based methods. The signal-based methods are not suitable for collaborative localization in WSNs because they require transmission of raw acoustic data. In feature-based methods, signal features are extracted at each sensor and the localization is done by multisensor fusion of the extracted features. Such methods are suitable for WSNs due to their lower bandwidth requirements. In this paper, we present a feature-based localization and discrimination approach for multiple harmonic acoustic sources in WSNs. The approach uses acoustic beamform and Power Spectral Density (PSD) data from each sensor as the features for multisensor fusion, localization, and discrimination. We use a graphical model to formulate the problem, and employ maximum likelihood and Bayesian estimation for estimating the position of the sources as well as their fundamental and dominant harmonic frequencies. We present simulation and experimental results for source localization and discrimination, to demonstrate our approach. In our simulations, we also relax the source assumptions, specifically the harmonic and omnidirectional source assumptions, and evaluate the effect on localization accuracy. The experimental results are obtained using motes equipped with microphone arrays and an onboard FPGA for computing the beamform and the PSD.