重尾非高斯定位噪声下鲁棒协同目标跟踪

针对定位数据的统计特性未知且易受异常值干扰而影响协同目标跟踪性能的问题，提出一种重尾非高斯定位噪声下的鲁棒协同目标跟踪方法. 该方法假设定位噪声服从多元学生t-分布，建立联合估计目标状态与定位噪声参数的贝叶斯模型. 针对目标状态与噪声分布参数相互耦合而难以计算联合后验分布的问题，应用变分贝叶斯推断原理和平均场理论对后验分布进行解耦，将目标状态与定位噪声参数的联合后验分布估计问题转化为最优化问题，以交替优化的方式实现系统参数的在线递推估计. 对提出的协同目标跟踪方法进行测试. 仿真结果表明，当定位数据中存在未知的野值噪声时，提出的协同跟踪算法具有较好的鲁棒性.

Robust cooperative target tracking under heavy-tailed non-Gaussian localization noise

The statistical properties of the localization data are unknown and the localization data are susceptible to outlier interference, which affects the cooperative target tracking performance. Aiming at the problem, a robust cooperative target tracking method for heavy-tailed non-Gaussian localization noise was proposed. It was assumed that the localization noise followed multivariate Student’s t-distribution and a joint Bayesian estimation model of target state and localization noise parameters was constructed. To overcome the difficulty of computing joint posterior distribution owing to the coupling of target state and noise distribution parameters, the variational Bayesian inference and the mean-field theory were applied to decouple the joint posterior distribution and convert the problem of joint estimation of target state and localization noise parameters into a optimization problem. Alternative optimization was implemented to achieve recursive estimation of system parameters. The proposed cooperative target tracking method was evaluated experimentally. Simulation results showed that when unknown outliers exist in localization data, the proposed algorithm has better tracking robustness.