| 摘要: |
| 目标跟踪的精度取决于滤波器的性能，但在许多实际的毫米波雷达测量场景中，模型及相关参数的不确定会导致传统滤波器的性能下降。为了应对这种情况，基于贝叶斯准则设计了一种鲁棒卡尔曼滤波器。利用Metropolis Hastings算法，从不确定噪声参数的似然函数中采样并以样本分布来近似后验噪声分布，然后计算其统计平均值，并借助后验噪声统计量扩展了经典卡尔曼滤波器。另外，针对采样时提议分布（Proposal Distribution）难以确定的问题，提出了一种提议分布自适应方法，即利用离散Fréchet距离来评价候选分布与先验分布变化趋势的相似性，然后选取相似性最大的候选分布为提议分布。通过处理毫米波雷达的测量数据证明了该滤波器在应对观测噪声不确定的场景时性能优越。 |
| 关键词: 毫米波雷达 卡尔曼滤波 贝叶斯鲁棒性 不确定观测噪声 提议分布 离散Fréchet距离 |
| DOI： |
|
| 基金项目:国家自然科学基金资助项目（U20A20163）；北京市教委科研计划（KZ202111232049,KM202011232021）；“勤信人才” 培育计划（QXTCP A201902） |
|
| A Bayesian robust Kalman filter based on posterior noise |
| ZHANG Chuyuan,CAO Lin,ZHAO Zongmin,WANG Dongfeng |
| (1a.Key Laboratory of the Ministry of Education for Optoelectronic Measurement Technology and Instrument;1b.School of Information and Communication Engineering,Beijing Information Science and Technology University,Beijing 100101,China;Beijing TransMicrowave Technology Company,Beijing 100101,China) |
| Abstract: |
| The accuracy of target tracking depends on the performance of the filter,but in many practical millimeter wave radar measurement scenarios,uncertain model and parameters will lead to the performance degradation of the traditional filter.To deal with this situation,a robust Kalman filter based on Bayesian criterion is designed.By using the Metropolis Hastings algorithm,samples are taken from the likelihood function of uncertain noise parameter,and the posterior noise distribution is approximated by the sample distribution,then the statistical average is calculated,and the classic Kalman filter is extended by the posterior noise statistic.Besides,for the problem that it is difficult to determine the proposal distribution during sampling,an adaptive method of proposal distribution is proposed,which uses discrete Fréchet distance to evaluate the similarity of the changing trend between candidate distribution and prior distribution,and then selects the candidate distribution with the largest similarity as the proposal distribution.By processing the measurements of millimeter wave radar,it is proved that the filter has superior performance in dealing with the scene with uncertain observation noise. |
| Key words: millimeter wave radar Kalman filter Bayesian robustness uncertain observation noise proposal distribution discrete Fréchet distance |
《电讯技术》编辑部 