Random finite set-based Bayesian filters using magnitude-adaptive target birth intensity

Modelling new-born targets that spontaneously appear in the multi-target tracking scene is an indispensable yet challenging task for any multi-target tracker, which asks for a careful formulation of the target birth intensity (TBI) in random finite set based Bayesian filters. However, the TBI is widely assumed to hold for a constant magnitude that needs to be specified in advance, indicating a constant speculation for the number of new targets to be appeared at all scans. This is not always desirable and can be problematic as the TBI magnitude is generally unknown and varies in time. In this paper, a data-driven approach is proposed to determine the TBI magnitude in real time based on the information contained in the newest observations. Simulations of the sequential Monte Carlo implementation of the probability hypothesis density filter and the multi-Bernoulli filter have demonstrated the validity of our approach.