Conjugate priors for Bayesian object tracking

Sammanfattning: Object tracking refers to the problem of using noisy sensor measurements to determine the location and characteristics of objects of interest in clutter. Nowadays, object tracking has found applications in numerous research venues as well as application areas, including air traffic control, maritime navigation, remote sensing, intelligent video surveillance, and more recently environmental perception, which is a key enabling technology in autonomous vehicles. This thesis studies conjugate priors for Bayesian object tracking with focus on multi-object tracking (MOT) based on sets of trajectories. Finite Set Statistics provides an elegant Bayesian formulation of MOT in terms of the theory of random finite sets (RFSs). Conjugate priors are also of great interest as they provide families of distributions that are suitable to work with when seeking accurate approximations to the true posterior distributions. Many RFS-based MOT approaches are only concerned with multi-object filtering without attempting to estimate object trajectories. An appealing approach to building tracks is by computing the multi-object densities on sets of trajectories. This leads to the development of trajectory filters, e.g., filters based on Poisson multi-Bernoulli mixture (PMBM) conjugate priors. In this thesis, [Paper A] and [Paper B] consider the problem of point object tracking where an object generates at most one measurement per scan. In [Paper A], it is shown that the trajectory MBM filter is the solution to the MOT problem for standard point object models with multi-Bernoulli birth. In addition, the multi-scan implementations of trajectory PMBM and MBM filters are presented. In [Paper B], a solution for recovering full trajectory information, via the calculation of the posterior of the set of trajectories from a sequence of multi-object filtering densities and the multi-object dynamic model, is presented. [Paper C] and [Paper D] consider the problem of ex- tended object tracking where an object may generate multiple measurements per scan. In [Paper C], the extended object PMBM filter for sets of objects is generalized to sets of trajectories. In [Paper D], a learning-based extended ob- ject tracking algorithm using a hierarchical truncated Gaussian measurement model tailored for automotive radar measurements is presented.