AP聚类和特征划分融合的群结构模型及跟踪算法  

Group structure model and tracking algorithm based on affinity propagation clustering and feature partitioning fusion

作　　者：王昊 宋骊平[1] WANG Hao;SONG Liping(School of Electronic Engineering,Xidian University,Xi’an 710071,China)

机构地区：[1]西安电子科技大学电子工程学院,西安710071

出　　处：《兵器装备工程学报》2025年第2期228-235,共8页Journal of Ordnance Equipment Engineering

摘　　要：针对群目标跟踪问题中发生群合并和分裂时,传统的演化网络模型通过将目标间的马氏距离与预设的阈值进行比较实现群组划分,导致其跟踪效果因依赖于阈值选择而在性能上受限的问题,提出了一种基于近邻传播聚类和特征划分融合的群结构模型,以避免上述问题并提升跟踪精度。新的群结构模型创新性地利用近邻传播聚类算法,依据目标点之间的距离和速度特征,在2个维度上对目标点进行有效划分,通过邻接矩阵表示聚类结果,并对两个邻接矩阵进行融合,构造出目标点的群组划分结构。结合高斯混合概率假设密度滤波进行群目标跟踪仿真对比实验,结果表明新的群结构模型在群组划分方面更接近群目标的真实划分,相较于传统的演化网络模型,新模型在群目标数目的估计及跟踪效果上有明显提升。所提出的群结构模型跟踪性能更好,模块化程度高并且具有更高的全局适应能力,为群目标跟踪提供了新的解决思路。Aiming at the problem that when group merging and splitting occurs in the group target tracking problem,the traditional evolving network model achieves group partitioning by comparing the Mahalanobis distances between the targets with the preset thresholds,which leads to the problem that tracking effect is limited in performance due to its dependence on the threshold selection.A group structure model based on affinity propagation clustering and feature partitioning fusion is proposed in order to avoid the above problems and enhance the tracking accuracy.The new group structure model innovatively makes use of the affinity propagation clustering algorithm to effectively divide the target points in two dimensions based on the distance and velocity characteristics between the target points,represents the clustering results through the adjacency matrix,and fuses the two adjacency matrices to construct the group division structure of the target points.Combined with Gaussian mixture probability hypothesis density filtering for group target tracking simulation and comparison experiments,the results show that the new group structure model is closer to the real division of group targets in terms of group division;and compared with the traditional evolving network model,the new model has significant improvement in the estimation of the number of group targets and tracking effect.The proposed group structure model has better tracking performance,a high degree of modularity and higher global adaptability,which provides a new solution idea for group target tracking.

关 键 词：群目标跟踪 近邻传播聚类 演化网络模型 概率假设密度滤波 邻接矩阵 

分 类 号：TN953[电子电信—信号与信息处理]