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

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VINSFusion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。

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-meterlong 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.