Multi-target tracking using proximity sensors

We consider the problem of tracking multiple moving targets in a continuous field using proximity sensors, which are binary sensors that can sense target presence by performing local energy detection subject to noise. Compared with more sophisticated sensors, proximity sensors have the advantage of having lower costs and lower energy consumption, but also the disadvantage of being less accurate. In this paper, we propose a hybrid tracking scheme where a coarse-scale tracking is first performed by proximity sensors to narrow down the areas of interest, and then a fine-scale tracking is performed by high-end sensors to estimate the exact target locations, with our focus on the former. In contrast to classic multi-target tracking which assumes 1-1 association between measurements and targets, we show that proximity measurements do not have such association and thus require a different objective. Formulating the coarse-scale tracking as a problem of tracking the histograms of targets in a cell-partitioned field, we develop both an optimal and two approximate solutions via Bayesian Filtering (BF). In particular, one of our approximate solutions decouples the tracking of different targets and thus reduces the dimensionality of BF by relaxing the likelihood function, and the other further reduces the problem into discrete space by quantizing the target mobility model and the relaxed likelihood function. Together with the optimal solution, they provide flexible tradeoffs between accuracy and complexity. Simulations show that the proposed solutions can effectively track targets to the accuracy of a cell and thus reduce uncertainty for the fine-scale tracking.