GM-PHD雷达密集多目标跟踪应用研究

Radar Based Space-close Multi-target Tracking Using Gaussian Mixture Probability Hypothesis Density

针对雷达密集多目标跟踪数据关联的难题,深入研究了可以避免数据关联的多目标跟踪方法-高斯混合概率假设密度算法(GM-PHD)。首先,将多目标的运动和多目标量测建模为随机有限集的形式,并给出了相应的最优多目标贝叶斯滤波器;然后,在线性高斯假设条件下,详细给出了GM-PHD均值、方差和权值的递归形式,降低了计算复杂度,满足跟踪实时性要求;最后,开展了仿真实验和实测数据实验,实验结果显示GM-PHD在不需要数据关联的情况下,能够有效抑制大量杂波,稳定地跟踪密集多目标。

Considering the challenge of data association for space-close multi-target tracking,the Gaussian mixture probability hypothesis density (GM-PHD) algorithm is studied in this paper,which does not require data association.Firstly,the motion and the measurement model of multi-target are modeled as random finite sets,and the corresponding optimal multi-target Bayes filter is given. Then,under the condition of linear and Gaussian assumption,the mean,covariance,and the weights of GM-PHD are derived to meet the real-time requirement.At last,simulation and real-data experiments are conducted and the results show that GM-PHD can effectively suppress the clutters and stably track the space-close multi-target without data association.