Association likelihoods for directional estimation

In multitarget tracking, using the association of tracks to measurements that maximizes the association likelihood is a well-established strategy in Euclidean spaces. We explain how this strategy can be adopted for circular domains. Formulae are provided for the association likelihood for three density representations used by important filters for periodic domains-von Mises densities, density approximations based on trigonometric polynomials, and particle-based representations. The presented closed-form formulae allow for efficiently determining the most likely association. In the evaluation, the approaches based on particles and trigonometric polynomials outperform an approach based on a Kalman filter that was adapted to the periodic domain.