Accelerated Multi-Sensor Control for Selective Multi-Object Tracking

Abstract

This paper focuses on selective multi-object tracking applications where objects with particular labels are of interest, and multiple sensors need to be controlled to achieve the optimum tracking performance for those objects. We formulate a novel solutions for a centralized multi-sensor multi-object system with labeled multi-Bernoulli filters in each sensor node. Our solution includes optimizing a closed-form objective function that can be directly calculated after prediction step in the central node of the sensor network. Importantly, our proposed cost function does not need to be computed after pseudo-update operations, hence a large amount of computation is saved. Simulation results indicate how the proposed methods can lead to significant improvements in terms of tracking accuracy of objects of interest, compared to using the generic (non-selective) sensor control methods. Numerical experiments involving a challenging multisensor target tracking application demonstrate that while our proposed method significantly outperforms the common (non-selective) sensor control methods, it performs similar to the state of art method for selective sensor control (selective-PEECS) in terms of the mean-square-error of tracking of the targets of interest. Despite similar performance in terms of tracking error, our method is significantly faster than the state of art (eight times faster in our experiments).