利用高斯混合概率假设密度滤波器对扩展目标量测集进行划分
A Measurement Set Partitioning for Extended Target Tracking Using a Gaussian Mixture Extended？？Target Gaussian Mixture Probability Hypothesis Density Filter

针对杂波环境下多扩展目标高斯混合概率假设密度(ET？？GMPHD)量测集划分难、计算量大的问题,提出了一种新的基于网格密度分布和谱聚类的扩展目标量测集划分方法。利用动态网格生成技术来获得量测集的网格密度分布;在获得网格划分后,将全部量测数据映射到网格单元中并统计网格单元的密度,且采用双密度阀值法来滤除量测集中的杂波;在谱聚类算法中利用密度敏感距离测度对去除杂波后的量测集构造相似矩阵,继而变换得到拉普拉斯矩阵;利用k？簿？值聚类算法对拉普拉斯矩阵的特征向量进行聚类划分。采用网格密度划分法滤除量测集中的杂波,使划分子集尽可能多地包含真实量测,增加划分子集与量测集合的近似度,从而在减少计算量的同时保证算法的跟踪性能损失不大。仿真实验表明,与典型的量测集划分算法相比,所提方法在跟踪误差损失约5％的前提下,计算效率提高了38％以上,具有更好的性能。
A new measurement set partitioning based on grid density and spectral clustering is proposed to overcome the problem that it is impossible to implement all the possible partitioning of a measurement set by the filters with extended？？target Gaussian mixture probability hypothesis density. Firstly, the dynamic grid generation technique is used to acquire the grid density of measurement set, then the double？？density threshold is adopted to remove the clutters of measurements set. Lastly, the spectral clustering based on the sensitive distance is applied in partitioning the measurement set from which the clutters have been removed. Simulation results show that, compared with the typical partition algorithm of measurement set, though the tracking performance of the proposed algorithm loses 5％, the computational efficiency is increased by 38％