Feature-aided localization of ground vehicles using passive acoustic sensor arrays

Localization of moving ground targets using acoustic signals obtained by a passive sensor network, made up of acoustic sensor arrays on the ground, is a difficult problem as the signals are contaminated by wind noise and hampered by road conditions and multipath, etc., and are generally not deterministic. It becomes even more challenging when some of the vehicles are wheeled (e.g., cars) and some are tracked (e.g., tanks), and are closely spaced. In such cases the stronger acoustic signals from the tracked vehicles can mask those from the wheeled vehicles, leading to poor detection of such targets. A novel detection scheme is presented, according to which the direction of arrival (DoA) angle estimates of emitters are obtained by each sensor array using real data. The full position estimates of targets, obtained following the association of the DoA angle estimates of the same target from at least three sensor arrays, are used for target tracking. However, because of the particular challenges encountered in multiple ground vehicle scenarios, this association is not always reliable and thus, target tracking using kinematic (DoA angle) measurements only is difficult and it can lead to lost tracks. In this paper we propose a new feature-aided multidimensional assignment algorithm, to augment the existing assignment algorithms which use only kinematic measurements, to improve the association performance, especially in the case of wheeled vehicles. We present a novel frequency domain feature extraction technique by implementing a statistical characterization of the features, in order to enhance the accuracy of data association. The feature and DoA angle measurements are used simultaneously, via a joint likelihood function, in a multidimensional assignment (MDA) to localize targets.