Probability hypothesis density filter with adaptive estimation of target birth intensity

Target birth intensity, which plays a role similar to track initialisation, is an important part of the probability hypothesis density (PHD) filter. Inmost papers, the intensity is always known as a priori, but it is too restrictive for real application. Besides, existing algorithms for the birth intensity estimation only consider the measured component of the target state (e.g. position), but the unmeasured component (e.g. velocity) is viewed as a priori or modelled by a simple distribution. As a result, they are not efficient enough to represent the initial states of newborn targets. To overcome that an adaptive estimation method combining with the single-point and two-point difference track initialisation algorithms is proposed. The target birth intensity is approximated by a Gaussian-mixture (GM) form whose means are equal to the locations and velocities of the candidate newborn targets, so that the estimated intensity can be closer to the truth. On the basis of that both the GM and sequential Monte Carlo implementations of the extended PHD filter are presented in this study. Experiment results show that the proposed method can effectively estimate the target birth intensity and has a better performance in the simulated scenarios.