顾及视觉地图点协方差的VIO/UWB融合室内定位算法  

Indoor localization algorithm of VIO/UWB fusion considering visual map point covariance

作　　者：高旺 何少鹏 王澄非[1] 潘树国[1] 徐锦乐 朱道华 GAO Wang;HE Shaopeng;WANG Chengfei;PAN Shuguo;XU Jinle;ZHU Daohua(School of Instrument Science and Engineering,Southeast University,Nanjing 210096,China;State Grid Jiangsu Electric Power Co.,Ltd.Research Institute,Nanjing 211100,China)

机构地区：[1]东南大学仪器科学与工程学院,南京210096 [2]国网江苏省电力有限公司电力科学研究院,南京211100

出　　处：《中国惯性技术学报》2025年第3期239-248,共10页Journal of Chinese Inertial Technology

基　　金：国家电网有限公司科技项目资助(5700-202318596A-3-2-ZN)。

摘　　要：针对室内复杂环境中弱纹理、光线变化和动态干扰所导致的里程计长航时定位精度下降、易发散等问题,提出了一种以惯导系统(INS)为主线的顾及视觉地图点协方差的视觉惯性里程计(VIO)/超宽带(UWB)融合室内定位算法。首先,对三维地图点协方差进行数学建模,估计视觉里程计中地图点在观测帧中的不确定性并计算对应的协方差。然后,根据协方差剔除动态不稳定的视觉地图点,求取剩余地图点的置信度,并作为优化权重加入到因子图中。最后,采用INS信息建立高频位姿主线,以其为先验对相机和UWB定位结果进行预测与抗差,将所有传感器数据以因子图优化的方法进行融合。在EuRoc数据集和真实环境室内数据集上进行验证,实验结果表明:所提VIO算法较VINS-MONO平均定位精度提升约37%以上,加入UWB的融合定位算法精度较VINS-MONO平均提升49%,能够在复杂室内环境中实现较高精度且连续稳定的定位。Aiming at the problems of decreased long-term positioning accuracy and easy dispersion of the odometer caused by weak texture,significant variations in illumination and dynamic interference in complex indoor environment,an inertial navigation system(INS)-centric visual-inertial odometry(VIO)/ultra-wideband(UWB) fusion indoor positioning algorithm considering visual map point covariance is proposed.Firstly,the covariance of 3D map points is modeled,the uncertainty of each map point in the visual odometry is estimated,and the corresponding covariance of the map points is calculated.Then,according to the obtained covariance,the unstable dynamic visual map points are eliminated,and the confidence of the remaining map points is obtained,which is added to the factor map as an optimization weight.Finally,INS information is used to establish the high-frequency pose main line,which is used as a priori to predict and robust the positioning results of the camera and UWB,and all sensor data are fused by factor graph optimization.Validation is performed on EuRoc dataset and real-world indoor dataset.The experimental results indicate that the average positioning accuracy of the proposed VIO algorithm is improved by more than 37% compared with VINS-MONO,and the accuracy of the fusion positioning algorithm with UWB is on average 49% higher than that of the VINS-MONO,which can achieve relatively high precision and continuous and stable positioning in complex indoor environment.

关 键 词：视觉惯性里程计 超宽带 多源融合定位 室内定位 

分 类 号：U666.1[交通运输工程—船舶及航道工程]