Multitarget Localization on Road Networks with Hidden Markov Rao-Blackwellized Particle Filters.

This paper considers the problem of tracking multiple moving targets on a road network with sparse, highly localized, unattended ground sensor data that are subject to clutter and missed detections. Hidden Markov models for single-target localization with unattended ground sensor data are first derived for road networks, under the assumption of perfect data association. These hidden Markov models are then used to solve the data association problem in the presence of clutter and missed detections for multitarget tracking using a Rao-Blackwellized particle filter. The proposed hidden Markov model tracking approach permits easy generation of accurate probabilistic models from a priori road network structure information, and it naturally enables sparse computationally efficient handling of multimodal target state uncertainties using both positive and negative unattended ground sensor information. The Rao-Blackwellized particle filter provides a fully Bayesian solution to the data association problem, enabling exploration of the association hypotheses space that leverages the computational advantages of exact hidden Markov model inference for multimodal state estimation. Numerical simulations demonstrate the effectiveness of the hidden Markov model/Rao-Blackwellized particle filter on challenging multitarget tracking scenarios with high false-alarm and missed detection rates.