Indoor multi-sensor fusion positioning based on federated filtering
Indoor multi-sensor fusion positioning based on federated filtering
复制标题
基于联合滤波的室内多传感器融合定位
DOI：
10.1016/j.measurement.2020.107506
复制
发表时间：
2020-03-15
期刊：
影响因子：
5.6
通讯作者：
Yan, XiaoYi
中科院分区：
文献类型：
--
作者：
Li, HuiXia;Ao, LongHui;Yan, XiaoYi
为了满足高精度室内导航的要求，提出了一种基于尺度不变特征变换（Voxel-SIFT）的点云快速配准方法，提高了激光雷达（LiDAR）的配准效率。由于信息分布因素的影响，传统的联邦滤波器无法实现动态自适应系统。在这种情况下，设计了一种基于加权最小二乘法的混合联邦滤波器。提出混合联邦滤波器的主滤波器采用最小方差准则，子滤波器的最优估计采用最优系数加权算法进行融合。信息分布因子可以实时动态更新，得到全局最优估计。最后，实验结果表明，该方法可以显著提高移动机器人的室内定位精度。平均定位误差为0.22 m，比目测里程表法或单独使用LiDAR里程表法定位精度更高。(C) 2020 Elsevier Ltd.版权所有。
In order to meet the requirements of high-precision indoor navigation, a fast point cloud registration method based on Voxel-SIFT (Scale-Invariant Feature Transform) is proposed to improve the registration efficiency of LiDAR (Light Detection and Ranging). Due to the influence of information distribution factors, the dynamic adaptive system cannot be realized in the traditional federated filter. In this case, a hybrid federated filter based on weighted least square method is designed. It is proposed that the main filter in the hybrid federated filter adopts the minimum variance criterion and the optimal estimate of the subfilter are fused according to the optimal coefficient weighted algorithm. The information distribution factor can be dynamically updated in real time to obtain a global optimal estimate. Finally, the experimental results show that the method can significantly improve the indoor positioning accuracy of mobile robots. The average positioning error is 0.22 m, which is more accurate than the visual odometer method or the LiDAR odometer method alone. (C) 2020 Elsevier Ltd. All rights reserved.