Efficient characterization of labeling uncertainty in closely-spaced targets tracking

In this paper we propose a novel solution to the labeled multi-target tracking problem. The method presented is specially effective in scenarios where the targets have once moved in close proximity. When this is the case, disregarding the labeling uncertainty present in a solution (after the targets split) may lead to a wrong decision by the end user. We take a closer look at the main cause of the labeling problem. By modeling the possible crosses between the targets, we define some relevant labeled point estimates. We extend the concept of crossing objects, which is obvious in one dimension, to scenarios where the objects move in multiple dimensions. Moreover, we provide a measure of uncertainty associated to the proposed solution to tackle the labeling problem. We develop a novel, scalable and modular framework in line with it. The proposed method is applied and analyzed on the basis of one-dimensional objects and two-dimensional objects simulation experiments.