融合全局位姿信息的视觉惯性SLAM算法  

Visual inertial SLAM algorithm integrating global pose information

作　　者：彭滨 蔡成林[1,2] 刘元成 PENG Bin;CAI Chenglin;LIU Yuancheng(School of Information and Communication,Guilin University of Electronic Technology,Guilin 541004,China;School of Information Engineering,Xiangtan University,Xiangtan 411105,China)

机构地区：[1]桂林电子科技大学信息与通信学院,广西桂林541004 [2]湘潭大学信息工程学院,湖南湘潭411105

出　　处：《桂林电子科技大学学报》2021年第2期113-117,共5页Journal of Guilin University of Electronic Technology

基　　金：国家自然科学基金(61771150);湖南省科技创新计划(2018GK2014,2018RS3089);广西科技计划(AB17129028)。

摘　　要：为了解决纯视觉SLAM算法无法在弱纹理、特征点缺失等场景中维持良好的鲁棒性,提出一种视觉传感器与惯性器件紧耦合的策略。即在视觉信息缺失的情况下,可通过惯性器件提供位置和姿态信息,避免误差的大量累积。在SLAM算法后端优化中,将全局位姿信息与视觉重投影误差、惯性元件残差和边缘化误差一起进行非线性优化,增加后端的约束条件。在公开数据集EuRoC上的实验结果表明,相较于其他方法,该算法的定位精度得到了提升,且具有良好的稳定性和实用性。In order to solve the problem that the pure visual SLAM algorithm cannot maintain good robustness in scenes such as weak texture and missing feature points,this paper proposes a strategy of tight coupling between visual sensors and inertial devices.That is,in the absence of visual information,inertial devices can be used to provide position and posture information to avoid a large amount of error accumulation.In the back-end optimization of SLAM algorithm,this paper proposes to perform nonlinear optimization of global pose information together with visual reprojection error,inertial element residual error and marginalization error to increase the back-end constraints.Finally,it is tested experimentally on the public data set EuRoC.The results show that compared with other methods,the positioning accuracy of this algorithm is improved,and it has good stability and practicability.

关 键 词：VSLAM 紧耦合 全局位姿信息 视觉重投影误差 惯性元件残差 

分 类 号：TN391.90[电子电信—物理电子学]