网格驱动PHD/CPHD滤波多目标跟踪算法

Grid-driven PHD/CPHD Algorithm for Multitarget Tracking

针对传统PHD/CPHD (probability hypothesis density/cardinalized probability hypothesis density)滤波在对未知运动参数(如机动参数,过程噪声参数等)及未知新生目标等复杂环境下跟踪多目标时性能低的问题,提出了一种自适应网格驱动PHD/CPHD滤波算法.首先,对跟踪区域进行网格初始划分;然后,根据量测信息进行新生目标识别,并更新网格权重,根据权重进行网格区域收缩以提取目标状态;最后,提出依据目标速度进行网格扩张的方法,以自适应重新划分网格,达到对复杂环境下数目未知的多目标的自适应跟踪.实验结果表明,当新生目标与存活目标在理想距离范围内,所提算法能够实现对未知运动参数及未知新生目标的变数目多目标自适应跟踪,在平均OSPA (optimal sub-pattern assignment)距离和平均目标数估计方面,比传统粒子PHD滤波方法具有更好的跟踪性能.

To address the low tracking performance of the traditional probability hypothesis density/cardinalized probability hypothesis density( PHD/CPHD) filter in complex scenarios,such as unknown dynamic parameters( e. g.,maneuver parameters and process noise) and unknown newborn targets,we propose an adaptive grid-driven( GD) PHD/CPHD filter algorithm. First,we divide the tracking area into an initial grid. Then,we identify newborn targets based on the measurement information,update the weights of the grids,and then construct grid regions based on these weights to extract the target states. Finally,we propose a method for expanding the grid area based on the target speed by adaptively re-dividing the expanded grid area based on the grid resolution to achieve multitarget tracking in complex scenarios. The experimental results show that when the newborn and surviving targets are within the ideal distance,the proposed algorithm performs better than the traditional particle-filter-based PHD method in terms of the average optimal sub-pattern assignment distance and the estimation of the average target number for tracking multitargets with unknown dynamic parameters and unknown newborn targets.