基于分层核粒子滤波的快速运动目标跟踪

针对目标发生快速、运动不规则及遮挡等情况下的跟踪问题,提出了一种分层核采样策略。首先通过先验转移和后验转移分别预测2组粒子来建立联合分布,利用聚类算法近似联合分布粒子集的混合高斯分布;然后对每个聚类进行采样;最后采用均值漂移算法将粒子移动到后验密度的局部极值处。实验结果表明:算法在目标发生快速机动情况时,跟踪性能优于传统粒子滤波、核粒子滤波及分层粒子滤波,且对遮挡具有较好的鲁棒性。

Tracking of Fast-moving Targets Based on Stratum Kernel Particle Filter

A stratum kernel sampling method is presented to resolve tracking problems such as rapid generation of target, irregular movement, under occlusion, etc. Stratum kernel sampling uses a random sampling method to sample the unit particles distribution set, and particles are moved to the posterior density at the local mode by mean shift. Experimental results show that the proposed algorithm is superior to the traditional particle filter, kernel particle filter and stratum particle filter while target is undergoing speedily irregular movement; and it＇ s also robust to occlusion.