Abstract:

Extended Target Gaussian Mixture Probability Hypothesis Density（ET-GM-PHD） filter has demonstrated as a promising approach in extended target tracking. However, when multiple targets are closely spaced to each other, state estimation accuracy of the algorithm decreases, which is due to the inability of the Distance Partitioning-Kmeans+（DP-Kmeans++） measurement set partitioning algorithm to output the correct results. To solve this problem, an improved DP-Kmeans++ measurement set partitioning algorithm is proposed, which uses Target Prediction（TP） information to set partition measurement, thus the partitioning accuracy will be improved. Simulation results show that, the proposed partitioning algorithm leads to lower OSPA values of the ET-GM-PHD filter, when targets are closely spaced.

Key words: target tracking, extended target, measurement set partitioning, density analysis, Probability Hypothesis Density（PHD）

摘要：

扩展目标高斯混合概率假设密度（Extended Target Gaussian Mixture Probability Hypothesis Density，ET-GM-PHD）跟踪算法是扩展目标跟踪领域内最为重要的跟踪算法之一。然而当多个目标邻近时，该算法的状态估计精度降低，这是由于距离-Kmeans++（Distance Partitioning-Kmeans++，DP-Kmeans++）量测集划分算法无法输出正确的结果所导致。为解决该问题，提出了改进的DP-Kmeans++量测集划分算法，利用目标预测信息来分割量测集，从而提高了划分精度。仿真结果表明，当目标邻近时，使用提出划分算法使ET-GM-PHD跟踪算法的OSPA误差距离减小。

关键词: 目标跟踪, 扩展目标, 量测集划分, 密度分析, 概率假设密度（PHD）