基于权重约束GM-PHD滤波的多目标跟踪方法

发布时间：2018-01-09 11:07

本文关键词：基于权重约束GM-PHD滤波的多目标跟踪方法　出处：《计算机工程》2017年03期 　论文类型：期刊论文

【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。
[Abstract]:For Gao Si hybrid probability assumption density density (GM-PHD) filter, the one-to-one hypothesis is not checked and it is difficult to track cross targets. On the basis of this method, an improved multi-target tracking method with constrained weights is proposed. By constructing the weight matrix, we can find the target with the largest weight from all the generated targets, according to the weight relationship. The target of all rows except the maximum weight is normalized and normalized and weighted constraint iterations are carried out. The corresponding weights of the target are generated in the update step of the GM-PHD filter. The filtering operation is completed. The method is evaluated by Monte Carlo simulation to detect the filter performance with clutter, different target speed and different frame rate. The simulation results of traversing and dense targets show that it is compared with the methods using GM-PHD filter and Sequential Monte Carlo probability assumption density density (SMC-PHD) filter. The overall tracking performance of this method is better.
【作者单位】：哈尔滨理工大学计算机科学与技术学院;
【分类号】：TN713
【正文快照】： 进行了研究,表征目标的形式也有很多种,如颜色直方图[5]、直方图方向梯度[6]和协方差描述符[7]等。多目标跟踪[1]是指在嘈杂观测、杂波以及不确根据这些特性可分为2种方案:整体方法[8和基于定检测的情况下,对目标的数量以及它们的状态进子空间的方法[1°"]。行联合估计,其

【参考文献】