A scale adaptive generative target tracking method based on modified particle filter

This paper proposes an advanced particle filter (PF) algorithm based on the quantum particle swarm optimization method (QPSO) and adaptive genetic algorithm (QAPF). After resampling of the PF, the position updating equation of the QPSO is applied to improve the particle distribution. Then replace the individuals with lower fitness with those with higher fitness. The genetic operation from the adaptive genetic algorithm (AGA) is then applied to increase the accuracy and sample diversity. An frame size adaptive adjustment model is proposed to reduce the number of useless features and improve the accuracy of target positioning. Multiple simulations of the nonlinear target tracking model are carried out, and the results demonstrate that the numerical stability, efficiency and accuracy of our QAPF algorithm are significantly better than those of other similar algorithms. QAPF is also compared with similar tracking algorithms via a set of tracking experiments. Our experiments on the OTB-100 dataset prove that the QAPF algorithm is much better than the PF, PF improved by particle swarm optimization (PSO-PF) and PF advanced by genetic algorithm (GAPF) tracking algorithms and other typical generative trackers in terms of the tracking precision, success rate, efficiency and robustness.

Automated summary

This paper proposes an advanced particle filter (PF) algorithm based on the quantum particle swarm optimization method (QPSO) and adaptive genetic algorithm (AGA). The PF is improved by applying the position updating equation of QPSO and replacing individuals with lower fitness. The accuracy and sample diversity are increased by applying the genetic operation from AGA. Furthermore, a frame size adaptive adjustment model is proposed to improve target positioning accuracy.