基于核Fisher判别的群结构更新模型及群目标跟踪算法

传统的群结构模型(如群演化网络模型)通过比较两个目标间的马氏距离与根据先验知识所设阈值的大小来对群的分裂合并进行判断, 跟踪效果依赖于设定的阈值, 难以应对群目标跟踪中的各种复杂情况。本文将分群的问题看作一个二分类问题, 提出了一种基于核Fisher判别分析的群结构更新模型, 通过离线训练得到符合群分裂和群合并特性的群结构更新模型, 将其直接用于群结构更新。结合箱粒子概率假设密度滤波算法的群目标跟踪仿真实验表明, 对比群演化网络模型, 本文提出的群结构更新模型对群结构的估计更加准确, 其在数目估计方面更稳定, 对群目标的跟踪效果更好。

Group structure update model and group target tracking algorithm based on kernel Fisher discriminant

The traditional group structure model, such as the group evolution network model, can judge the group splitting and merging by comparing the Mahalanobis distance between two targets and by the size of the threshold set according to the priori knowledge. As the tracking effect depends on the threshold, it is difficult to deal with various complex situations in the group target tracking. In this paper, the problem of grouping is regarded as a binary classification problem, a group structure update model based on kernel Fisher discriminant analysis (KFDA) is proposed. The group structure update model is obtained by off-line training, which meets the characteristics of group splitting and group merging, and is directly used for group structure updating. The simulation experiments of group target tracking combined with the box particle probability hypothesis density (BP-PHD) algorithm show that, compared with the group evolution network model, the proposed group structure update model is more accurate in the estimation of group structure, more stable in the estimation of number, and has better performance in the group target tracking.