一种基于变分推断的雷达多目标跟踪JPDA算法

A JPDA Algorithm for Radar Multi-target Tracking Based on Variational Inference

针对雷达邻近多目标跟踪问题,提出了一种基于变分推断的联合概率数据关联算法(Joint Probability Data Association,JPDA)。通过建立关于目标状态和两个关联指示的概率图模型,并根据不同变量之间的信息传递构造对应的自由能目标函数,迭代该目标函数求解出目标和当前检测量测之间的最佳边缘关联概率。将所提算法与经典JPDA和k近邻联合概率数据关联(k Nearest Neighbor-Joint Probability Data Association,kNN-JPDA)算法进行对比,结果表明新算法具备更高的跟踪位置精度,并且能够有效地避免因邻近目标数量增多而引起的计算上的组合爆炸问题。

A Joint Probabilistic Data Association(JPDA)algorithm based on variational inference is proposed for the problem of radar adjacent multi-target tracking.A probabilistic graphical model about the target states and the two associated indicators is established.And the corresponding free energy objective function is constructed according to the information between different variables.By iterating the objective function,the marginal association probabilities between the target and measurements will be calculated.Compared with the classic JPDA and k Nearest Neighbor-Joint Probability Data Association(kNN-JPDA)algorithm,the experiment results show that the proposed algorithm has obvious advantages in tracking accuracy and tracking position accuracy,and the problem of combinatorial explosion caused by the increase of adjacent targets can be effectively solved.