基于标签随机有限集的多量测多目标跟踪算法

针对一个扫描周期内单个目标可能产生多个量测的问题, 提出一种基于标签随机有限集的扩展算法. 结合脉冲扩展标签多伯努利(-GLMB) 滤波器和多量测模型, 推导出新的更新方程; 采用假设分解策略对关联过程进行降维, 避免了量测分组过程. 实验分析表明: 所提出算法能对目标数进行无偏估计, 在低探测概率条件下跟踪性能明显优于多量测概率假设密度(MD-PHD) 算法; 计算开销在量测较少时高于MD-PHD, 量测个数增加时增幅低于MD-PHD.

Multi-detection multi-target tracking with labeled random finite sets

For the case that one target may generate multiple detections per scan, an extended method within the labeled random finite sets framework is proposed. A new update equation is derived based on the delta-generalized labeled mult- Bernoulli(delta-GLMB) filter and multi-detection model. The hypotheses decomposition strategy is employed to reduce the dimensions of association process, so that partitioning of the detections set is avoided. Experiments indicate that the proposed method can estimate the target number without bias, and significantly outperforms the multi-detection probability hypothesis density(MD-PHD) filter in low probability of detection situation. The computational cost of the proposed method is higher than MD-PHD with a few detections, and grows slowly than MD-PHD when the amount of detections increases.