Glint噪声环境下的PHD滤波方法研究

发布时间：2018-06-16 19:21

本文选题：目标跟踪 + PHD滤波　；参考：《哈尔滨工业大学》2015年硕士论文

【摘要】：随着科学研究和技术的发展,目标跟踪方法在军事和民用领域均得到了普遍的应用,而目标跟踪领域中近几年的研究热点之一就是概率假设密度(Probability Hypothesis Density,PHD)滤波。PHD滤波是基于有限集统计(Finite Set Statistics,FISST)的滤波方法。本文主要研究随机有限集理论框架下glint噪声环境中目标跟踪的PHD滤波方法。在本文研究中,第二章首先研究了glint噪声的特征。glint噪声是在雷达跟踪环境下出现的测量噪声,根据其显著的非高斯分布及长拖尾特性考虑采用t分布的建模方法。接着基于RFS理论,研究了PHD滤波算法,并考虑到仿真实验中的非线性观测方程,采用了基于扩展卡尔曼滤波(Extended Kalman Filter,EKF)的高斯混合PHD(GM-PHD)滤波算法。第三章通过增广目标状态和噪声参数来扩展PHD滤波,为了得到扩展PHD的封闭解,对噪声参数应用先验伽马分布使得预测和更新强度能由高斯-伽马项混合表示。因为目标状态和噪声参数在似然函数中是耦合的,所以应用变分贝叶斯方法得到近似分布使得更新强度的表达形式和预测强度一样,并且生成的变分贝叶斯PHD(VB-PHD)滤波算法是递归的。最后一章研究了扩展目标的跟踪情况。扩展目标跟踪的重点在于测量集的分割,原则上应该是将来源于同一个目标的测量都分到一起,但是本文为了研究方便,采用了比较简单的马氏距离分割法。仿真实验表明了所提出的VB-PHD滤波的跟踪效果要优于GM-PHD滤波。
[Abstract]:With the development of scientific research and technology, target tracking method has been widely used in both military and civil fields. In recent years, one of the research hotspots in the field of target tracking is the probability hypothesis density hypothesis probability density filter. PhD filter is a filtering method based on finite set Statistics set (FISST). In this paper, the PhD filtering method for target tracking in glint noise environment is studied in the framework of stochastic finite set theory. In the second chapter, we first study the feature of glint noise. Glint noise is the measurement noise in radar tracking environment. According to its significant non-Gao Si distribution and long tail characteristics, we consider a t distribution modeling method. Then, based on the RFS theory, the PhD filtering algorithm is studied, and considering the nonlinear observation equation in the simulation experiment, the Gao Si hybrid PHD GM-PHD filter algorithm based on extended Kalman filter (EKF) is adopted. In chapter 3, the extended PhD filter is extended by extending the target state and noise parameters. In order to obtain the closed solution of extended PhD, a prior gamma distribution is applied to the noise parameters so that the prediction and update intensity can be represented by the Gauss-Gamma term mixture. Because the target state and the noise parameters are coupled in the likelihood function, the variational Bayesian method is used to obtain the approximate distribution so that the expression of the updated strength is the same as the predicted intensity. And the generated variational Bayesian PHD VB-PHD filtering algorithm is recursive. In the last chapter, the tracking of extended targets is studied. The emphasis of extended target tracking is on the segmentation of measurement sets. In principle, all measurements from the same target should be divided together. However, in order to facilitate the research, a simple Markov distance segmentation method is used in this paper. Simulation results show that the proposed VB-PHD filter has better tracking performance than GM-PHD filter.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN713

【参考文献】