Multitarget tracking is an important topic in visual surveillance system. Considering imperfections of the cardinalized probability hypothesis density (CPHD) filter and the target maneuvers, we propose an adaptive genetic multiple-model… Click to show full abstract
Multitarget tracking is an important topic in visual surveillance system. Considering imperfections of the cardinalized probability hypothesis density (CPHD) filter and the target maneuvers, we propose an adaptive genetic multiple-model CPHD filter in this paper. First, we discuss the filtering process and combined the standard CPHD filter with the multiple-model-based framework. Afterward, the sequential Monte Carlo implementation of the proposed filter for the nonlinear and non-Gaussian state estimates is presented in detail. To enhance the tracking performance as target start to maneuver, the adaptive genetic algorithm is used to improve the target state estimation accuracy at the time of state switching with the excellent particles. On the other hand, the undetected component of the measurement-updated weight of survival particle is compensated by the excess weight of newborn particle to correct the number estimates of targets. The simulation results are provided to illustrate the reliability and efficiency of the proposed filter.
               
Click one of the above tabs to view related content.