Multiple Target Tracking With Constrained Motion Using Particle Filtering Methods

Particle filtering has been successfully used in target tracking applications when the state and measurement models are nonlinear and the associated noise is non-Gaussian. Among the most difficult scenarios of target tracking, are multiple target tracking (MTT) problems that require the application of realistic and often multiple kinematic models. Recently algorithms such as the independent partitions (IP) and coupled partitions (CP), that make use of the particle filter, have proven their ability to handle MTT problems effectively, demonstrating the possibilities offered by the particle filter. The particle filter can also use different kinds of additional information that may be available, with the use of likelihood functions and sampling distributions. Such information may arise from targets having constraints in their motion. The ability to incorporate such kinematic behavior into the tracking algorithm can improve tracking performance. The IP and CP assume independence in target motion, therefore ignore information about kinematic constraints, when available. We introduce the constrained motion proposal (COMP) algorithm that uses multitarget proposal densities and motion models that incorporate kinematic constraint information into a particle filter. More specifically, it uses methods of sampling and likelihood functions that take into account motion constraint information. It also introduces variability in partition proposal order, that reduces the effects of erroneously imposed constraints. We compare error performance results of the COMP with those of other proposal methods. We show that the methods used by the COMP make effective use of motion constraint information and, thus, improve tracking performance.