Covariance control joint integrated probabilistic data association filter for multi‐target tracking

Abstract

The joint integrated probabilistic data association (JIPDA) filter is effective for automatic multi-target tracking in cluttered environment. At each time step, the posterior probability density function (PDF) is approximated by a Gaussian PDF to estimate track states. However, JIPDA suffers from the track coalescence problem. When targets are closely spaced, their tracking gates may overlap, so that the track state estimates become biased and tend to coalesce. The authors note that the covariance can influence both the size of the tracking gate and the accuracy of the Gaussian approximation. Therefore, in this study, the posterior PDF is optimised by controlling the covariance to improve the accuracy of the track state estimates. A cost function measuring the track estimate error covariance is developed as the optimisation criterion, and an iterative strategy is adopted for minimising it. Finally, the posterior PDF with the minimal cost function is derived as the desired one. The theoretical analysis and example show that the proposed approach can reduce the extent of tracking gate overlap and improve the accuracy of the Gaussian approximation. Simulation results show that the new approach can handle track coalescence and performs better than the traditional methods.