基于图优化的GNSS/双目视觉/惯性SLAM系统开发及应用

The developing and application of graph optimization-based GNSS/stereo visual/inertial SLAM system

夏琳琳 ¹宋梓维 ¹方亮 ¹孙伍虹志²

扫码查看

作者信息

1. 东北电力大学自动化工程学院,吉林 132012
2. 吉林化工学院信息与控制工程学院,吉林 132022
折叠

摘要

为提高机器人室外长航时定位精度,提出一种基于图优化的全球导航卫星系统(GNSS)/双目视觉/惯性同时定位与建图(SLAM)系统开发及应用.将空间中的线特征作为几何约束的补充,集成至前端的特征提取及后端的位姿优化线程,提升位姿解算精度.同时,以因子图构建联合优化的图结构,并推导出全局观测误差模型.近 200 m的BullDog-CX机器人巡检结果表明,所提算法相比于VINS-Fusion 和 PL-VINS分别取得约 12.6%及 3.4%的定位精度提升,为室外机器人长航时导航提供了一种可行方案.

Abstract

In order to improve the outdoor long-endurance positioning accuracy of robots,the developing and application of graph optimization-based global navigation satellite system(GNSS)/stereo visual/inertial simultaneous localization and mapping(SLAM)system is proposed.As the extra geometrical constraints,the spatial line features are integrated into the threads of the front-end feature extraction and back-end pose optimization to enhance the pose estimates.At the same time,the graph structure for joint optimization is constructed via factor graphs and the global observation error model is further derived.The nearly 200-meter-long BullDog-CX robot substation experiment shows that compared with VINS-Fusion and PL-VINS,the proposed algorithm has achieved about 12.6%and 3.4%improvement in positioning accuracy,which provides a feasible scheme for long-endurance navigation of outdoor robots.

关键词

GNSS/双目视觉/惯性SLAM系统/图优化/线特征约束/全局观测/多传感器融合

Key words

GNSS/stereo visual/inertial SLAM system/graph optimization/line feature constraint/global observation/multi-sensor fusion

引用本文复制引用

基金项目

吉林省科技厅自然科学研究项目(20220101240JC)

出版年

2024

中国惯性技术学报

中国惯性技术学会

中国惯性技术学报

CSTPCD北大核心

影响因子：0.792

ISSN：1005-6734

段落导航