A faster data assignment algorithm for maximum likelihood-based multitarget motion tracking with bearings-only measurements (Q5944024)

This paper examines the problem multi-target motion tracking, restricted to the case of no cluttering and no missing measurements. The maximum likelihood-based relaxation approach is used to obtain a solution to the multi-target tracking problem, where a non-convex conditional likelihood function of bearings-only measurements is maximized with respect to both data associations computed and target initial states. This is equivalent to minimizing the magnitude of the average square errors (ASEs) for the minimization equation of the associated problem. A global minimum can be attained in principle by repeating the following two steps until the solution converges. (1) The first step consists in minimizing the ASE with respect to initial target state vector for fixed data assignment matrices, this phase of the algorithm using the Gauss-Newton downhill type algorithm when resolving local minima trappings. (2) For the second step in obtaining the global minima, the authors introduced a new rotational sorting-based algorithm that replaces the Hungarian type assignment algorithms. The result is a novel general algorithm that is capable of reducing the computational complexity of the Hungarian type algorithms in the second step from \(O(n^3)\) to \(O(n^2)\).