改进的自适应新生目标强度PHD滤波

Improved adaptive target birth intensity for PHD filter

自适应新生目标强度(probability hypothesis density,PHD)滤波是一种新颖的量测驱动的多目标跟踪算法。然而,该算法存在归一化失衡问题,且在航迹生成方面存在一定的滞后现象。针对以上问题,提出一种改进算法。首先,在分析归一化失衡问题的基础上,提出一种归一化因子修正方法,有效解决该问题。其次,在高斯混合框架下对算法进行实现,并引入一种新的航迹回溯机制,通过对每个高斯分量进行标记,然后对存在概率超过确认门限的分量进行回溯,从而得到每个目标的完整航迹。实验结果表明,改进算法在新生目标搜索和多目标航迹生成方面均优于传统算法,具有良好的工程应用前景。

The adaptive target birth intensity probability hypothesis density （PHD） filter is a novel measurement-driven algorithm for multi-target tracking. However, there is a normalized unbalance problem and some lags of the extracted tracks in the filter. To solve these problems, an improved algorithm is proposed. Firstly, a modified normalized factor is proposed based on the analysis of the normalized unbalance problem. Secondly, a Gaussian mixture implementation is proposed, and then a recalling procedure for track maintenance is developed, which labels each Gaussian component and recalls the previous tracks for the components with existence probabilities larger than the confirm threshold. The simulation results show that the improved algorithm has the advantages over the ordinary one in the aspects of newborn target searching and multi-target track extracting, implying good application prospect.