航空学报 > 2019, Vol. 40 Issue (6): 322650-322650   doi: 10.7527/S1000-6893.2018.22650

基于高斯混合模型的航迹抗差关联算法

李保珠, 董云龙, 丁昊, 关键   

  1. 海军航空大学 信息融合研究所, 烟台 264001
  • 收稿日期:2018-09-05 修回日期:2018-10-15 出版日期:2019-06-15 发布日期:2018-11-26
  • 通讯作者: 关键 E-mail:guanjian96@tsinghua.org.cn
  • 基金资助:
    国家自然科学基金(61871392,61531020,61871391,61471382,U1633122);航空科学基金(20150184003,20162084005,20162084006)

Anti-bias track association algorithm based on Gaussian mixture model

LI Baozhu, DONG Yunlong, DING Hao, GUAN Jian   

  1. Institute of Information Fusion, Naval Aviation University, Yantai 264001, China
  • Received:2018-09-05 Revised:2018-10-15 Online:2019-06-15 Published:2018-11-26
  • Supported by:
    National Natural Science Foundation of China (61871392, 61531020, 61871391, 61471382, U1633122); Aeronautical Science Foundation of China (20150184003,20162084005,20162084006)

摘要: 针对雷达系统误差时变、上报目标不完全一致等复杂场景下目标航迹关联问题,采用高斯混合模型(GMM)与航迹间拓扑信息相结合的方法实现航迹抗差关联。将航迹关联问题转化为图像匹配中的非刚性点集匹配问题,建立对非同源航迹具有鲁棒性的高斯混合模型,根据航迹间的邻域拓扑信息决定高斯混合模型中各高斯组成部分的权重,利用期望最大值(EM)算法求解高斯混合模型的最优闭合解,在期望步(E-step)阶段求解航迹的对应关系,在最大化步(M-step)阶段求解非同源航迹比例,最后进行航迹关联判决以获得关联结果。仿真结果表明,该算法在不同系统误差、目标分布密度、探测概率等环境下具有较好有效性和鲁棒性。

关键词: 航迹关联, 时变系统误差, 邻域拓扑信息, 高斯混合模型, 期望最大值(EM)

Abstract: To address the track-to-track association problem in the presence of time-varied sensor biases and different targets reported by different sensors, Gaussian Mixture Model (GMM) and neighborhood topology information are used in this paper. The robust track-to-track association problem is turned into a non-rigid point matching problem. The Gaussian mixture model is established with better robustness to ‘unpaired’ tracks. The weight of each Gaussian component is decided by the neighborhood topology information between tracks. The optimal closed solution of the Gaussian mixture model is solved by Expectation Maximization (EM) algorithm. In Expectation-step of the EM algorithm the correspondence of tracks is solved, and in Maximization-step the ‘unpaired’ tracks ratio are calculated. Finally, the track-to-track association is obtained by judgment. Monte carlo simulation demonstrates the effectiveness of the proposed approaches under different sensor biases, targets densities and detection probabilities.

Key words: track association, time-varied sensor biases, neighbor topology information, Gaussian mixture model, Expectation Maximization (EM)

中图分类号: