Distributed multi-target tracking over an asynchronous multi-sensor network

This paper addresses distributed multi-target tracking (DMTT) over an asynchronous multi-sensor network (AMSN). Within the AMSN, the sensor nodes are usually misaligned in time due to different sampling instants and/or rates. At the same time, time-offsets among nodes are always imprecise or even unknown. In such cases, time alignment (TA) needs to be carried out before fusion of information between different nodes. In the considered AMSN for DMTT, Probability Hypothesis Density (PHD) filters are run in each node for propagating in time a local first-order statistic, called intensity, of the target set, while arithmetic average (AA) fusion is used to combine intensities from different nodes. Recalling that AA intensity fusion admits an information-theoretic interpretation in terms of minimizer of the weighted average of Cauchy-Schwartz divergences (CSDs) with respect to the local intensities, the corresponding minimum weighted average CSD (MWCSD) is adopted as cost to be minimized for TA purposes. To ensure good convergence of the TA parameters, a convex combination of the instantaneous cost and the squared difference between current and previous estimates, is proposed. Furthermore, a sampling technique is adopted to solve the optimization problem. Finally, simulation experiments are provided to demonstrate the effectiveness of the proposed approach.