划分交互式箱粒子概率假设密度滤波法

Partitioned Interacting Multiple Model Box-PHD Filter

针对现有的多机动目标追踪问题,将交互式多模型(interacting multiple model,IMM)思想与箱粒子概率假设密度滤波器(box probability hypothesis density filter,Box-PHD)相结合,并针对箱粒子在区间密集杂波等复杂环境下箱体偏大,所导致的箱粒子冗余和目标跟踪位置估计不精确等问题,引入箱粒子划分技术,提出一种划分交互式概率假设密度滤波(partitioned interacting multiple model probability hypothesis density filter,PIMM-Box-PHD)算法,来处理椭圆形多机动目标的跟踪问题。该算法首先在预测阶段针对多目标的机动问题引入IMM预测,利用多模型交互方法来解决目标运动时模型失配问题;其次,利用箱划分技术将预测得到的箱粒子划分为大小和权值相同的多个子箱,以提高目标位置估计精度;最后,利用Box-PHD滤波对划分后的小箱粒子集进行区间量测更新。利用实验验证了PIMM-Box-PHD算法在多机动目标跟踪方面的良好性能,以及相较于IMM-Box-PHD算法在目标位置估计方面的优势。

Aiming at the existing problem of multi-maneuvering target tracking,the idea of interacting multiple model(IMM)is combined with box probability hypothesis density filter(Box-PHD).And for the problems such as box particle redundancy and inaccurate target tracking position estimation caused by large cabinets in the complex environment with dense interval clutter,the box particle partitioning technology is introduced and a partitioned interactive probability assumption density filter(PIMM-Box-PHD)is proposed to deal with the tracking of elliptical multi-maneuvering targets.The algorithm first introduces IMM prediction for the multi-target maneuvering problem in the prediction phase,and uses the multi-model interaction method to solve the model mismatch problem when the target moves.Secondly,the box division technology is used to divide the predicted box particles into the same size and weight multiple sub-boxes to improve the accuracy of target position estimation.Finally,Box-PHD filtering is used to perform interval measurement and update on the divided small-box particle sets.The experiment shows that the PIMM-Box-PHD algorithm proposed is efficient in tracking multiple maneuvering targets and superior to the IMM-Box-PHD algorithm in target position estimation.