摘要： 针对高斯混合概率假设密度(GM-PHD)滤波器未检查一对一假设以及难以跟踪跨越目标的问题,在其基础上提出一种约束权重的改进多目标跟踪方法。通过构建权重矩阵,从所有生成的目标中寻找权重最大的目标。根据权重关系,重新归一化除最大权重外所有行的目标,并使归一化和权重约束迭代进行。在GM-PHD滤波器的更新步骤中生成目标的相应权重,完成滤波操作。通过蒙特卡罗仿真对该方法进行评估,检测有杂波、不同目标速度和不同帧率情况下的滤波器性能,分别对穿越和密集的目标进行仿真。实验结果表明,与使用GM-PHD滤波器和基于序贯蒙特卡洛概率假设密度(SMC-PHD)滤波器的方法相比,该方法整体跟踪性能较优。

关键词: 多目标跟踪, 一对一假设, 高斯混合概率假设密度滤波器, 权重约束, 归一化

Abstract: Concerning that the Gaussian Mixture Probability Hypothesis Density(GM-PHD) filter does not check one-to-one assumption and it is difficult to track crossing targets,an improved multi-target tracking method with weight constraint is proposed based on GM-PHD filter.Firstly,the weight matrix is constructed and the target with the maximum weight is searched from the generated targets.Then,according to weight relationship,targets in all rows are re-normalized except that with the maximum weight.Normalization and weight restriction are operated iteratively.Finally,the target weight is generated in the update steps of GM-PHD filter for complete filtering operation.The method is evaluated through Monte Carlo simulation,and the performance of filters are tested on the conditions of clutter,different target speeds and different frame rates,with simulations of crossing and dense targets.Experimental results shows that,compared with the methods using GM-PHD filter and Sequential Monte Carlo PHD (SMC-PHD) filter,the proposed method has better overall performance.

Key words: multi-target tracking, one-to-one assumption, Gaussian Mixture Probability Hypothesis Density(GM-PHD)filter, weight constraint, normalization

中图分类号: 

• TP391