Multiple-hypothesis and graph-based tracking for kinematic and identity fusion

A key challenge in disparate-sensor multi-target tracking is how to exploit sparse but highly relevant information from some sensors, in combination with voluminous data from other sensors. Effective exploitation is problematic for both MHT-based approaches, which cannot contend with the required hypothesis-tree depths, and for graph-based approaches, which rely on a Markovian assumption (yielding a pairwise-cost structure) that does not apply to all data in this context. Our work extends both MHT-based and graph-based approaches to the problem, with particularly promising results for the latter
¹²
.