中文摘要：

针对传统的高斯混合概率假设密度（GM—PHD）滤波器在噪声先验特性未知或不准确时跟踪性能会下降，提出了一种基于噪声方差估计的高斯混合概率假设密度（NCE—GM—PHD）滤波算法．该算法可以同时在线估计时变的目标个数、多目标状态以及噪声方差．首先，通过引入遗忘因子和采取有偏估计的方法改进了传统的Sage-Husa自适应滤波器．基于改进的自适应滤波器，推导了带噪声方差估计的GM-PHD滤波算法．仿真结果表明，在非时变或时变量测噪声方差未知的情况下，NCE-GM—PHD算法的跟踪性能优于传统的GM—PHD算法，对噪声变化的适应能力更强．

英文摘要：

When the prior noise statistics was unknown or inaccurate, the conventional Gaussian mixture probability hypothesis density （GM-PHD） filter declined in tracking performance. To solve this problem, a noise covariance estimation based GM-PHD （NCE-GM-PHD） filter was proposed for jointly estimating the time-varying number of targets, multi-target states and noise covariance. First, the adaptive filter of Sage and Husa was modified by introducing a forgetting factor and using the biased estimation. Based on the modified adaptive filter, the noise estimation method was adopted to derive a closed-form solution for NCE-GM-PHD filter. Simulation results demonstrate that the NCE-GM-PHD filter has a favorable per- formance and adaptability of noise changes compared to the classical GM-PHD filter with unknown time-in- variant or time-varying measurement noise statistics.