视觉/惯性组合导航中的SWF与MSCKF对比研究

A comparative research of SWF and MSCKF on visual/inertial integrated navigation

利用单目相机获取的地物特征辅助惯性导航系统解算,可有效减小单一惯导系统工作时引起的定位误差发散.结合定位精度和时效性对最优匹配滤波算法的要求,提出滑动窗口滤波(SWF)和多状态约束卡尔曼滤波(MSCKF)两种融合算法,并对视觉/惯性组合系统进行误差补偿,分析特征密度、特征跟踪长度和窗口大小对定位精度的影响,采用相同测试数据集分别对两种滤波方法进行对比实验.结果表明:1)在可见特征较多条件下,SWF和MSCKF算法定位结果精度均优于纯惯性系统(INS)解算精度;2)SWF算法解算精度高于MSCKF算法解算精度;3)相比于SWF算法,MSCKF算法对参数变化更敏感;特征数增多时,MSCKF算法定位精度提高、计算复杂度更低;4)适当改变特征跟踪长度和窗口取值,不同滤波融合方案解算结果精度可小幅提高.

Using the features acquired by the monocular camera to assist the inertial navigation solution,the positioning error divergence caused by the single inertial navigation system can be effectively reduced.In this paper,two fusion algorithms,the sliding window filter(SWF)and multi-state constrained kalman filter(MSCKF),are proposed to compensate the errors of the visual/inertial system combined with the positioning accuracy and timeliness of the optimal matching filter algorithm requirements.The effects of feature density,feature track length and window size on the positioning accuracy are analyzed,and the same test data sets are adopted to conduct the contrast experiment of the two filters.The results showed that under the condition of more visible features,the positioning accuracies of two proposed filter algorithms were better than that of the pure inertial navigation system(INS)solution.The positioning accuracy of SWF is obviously better than that of MSCKF.However,MSCKF is more sensitive to parameter changes compared with SWF.And the positioning accuracy of MSCKF is improved and computational complexity is lower with the increase of the number of features.Moreover,the positioning accuracies of different filter fusion methods are slightly improved after changing the feature track length and window size.