MCMC and MHT Approaches to Multi-INT surveillance

This paper proposes two track fusion methodologies for challenging multi-target tracking (MTT) settings where sensors have highly disparate characteristics and target density is high, leading to many competing tracking solutions. Though distributed multiple hypothesis tracking (MHT) is known to provide a viable solution paradigm, its applicability is limited to medium-size scenarios due to the need for deep hypothesis trees. For large-scale scenarios, a computationally efficient min-cost flow solution paradigm has been proposed that works well for kinematic sensor data, but is not applicable to multi-INT data that includes identity information that does not degrade over time. This paper introduces two approaches to the problem. The first is a natural extension to the MHT paradigm, and seeks to improve performance by considering out-of-sequence processing: the asynchronous MHT (A-MHT). The second adapts a recently proposed Markov Chain Monte Carlo (MCMC) approach to target tracking to multi-INT track fusion: the MCMC Data Fuser (MCMC-DF). A-MHT and MCMC-DF results are promising against an MHT baseline.