Distributed Variational Filtering for Simultaneous Sensor Localization and Target Tracking in Wireless Sensor Networks

Abstract

The tracking of a moving target in a wireless sensor network (WSN) requires exact knowledge of sensor positions. However, precise information about sensor locations is not always available. Given the observation that a series of measurements are generated in the sensors when the target moves through the network field, we propose an algorithm that exploits these measurements to simultaneously localize the detecting sensors and track the target (SLAT). The main difficulties that are encountered in this problem are the ambiguity of sensor locations, the unrestricted target moving manner, and the extremely constrained resources in WSNs. Therefore, a general state evolution model is employed to describe the dynamical system with neither prior knowledge of the target moving manner nor precise location information of the sensors. The joint posterior distribution of the parameters of interest is updated online by incorporating the incomplete and inaccurate measurements between the target and each of the sensors into a Bayesian filtering framework. A variational approach is adopted in the framework to approximate the filtering distribution, thus minimizing the intercluster communication and the error propagation. By executing the algorithm on a fully distributed cluster scheme, energy and bandwidth consumption in the network are dramatically reduced, compared with a centralized approach. Experiments on an SLAT problem validate the effectiveness of the proposed algorithm in terms of tracking accuracy, localization precision, energy consumption, and execution time.