月面特征稀疏环境下的视觉惯性SLAM方法

月球车在执行科学探测任务过程中，其自身的高精度定位是一项亟需解决的关键问题。针对在特征稀疏的月面环境下的定位问题，提出一种视觉惯性融合的SLAM方法，将视觉测量与惯性传感器的信息利用位姿图优化方法融合，实现高精度的联合定位。针对特征稀疏环境下的前端视觉数据关联误差较大的问题，提出了一种基于四元树的光流跟踪算法，能够有效地跟踪鲁棒的特征点，提升了关键帧之间相对位姿估计的准确性。并且针对月面环境特有的恒星无穷远点干扰问题，提出一种高效的恒星点剔除算法，能够有效改善无穷远点导致的定位精度下降的问题。搭建了一套模拟月面环境的计算机仿真系统，并构建了多个月面环境视觉惯性SLAM仿真数据集，在不同的模拟月面场景下进行定位性能仿真验证，仿真测试结果表明本文算法的鲁棒性更强，具有更高的定位准确度。

Visual-inertial SLAM in featureless environments on lunar surface

During lunar surface scientific explorations, high-accuracy self-localization of lunar rovers is a key problem to be solved. Aiming at accurate localization in the featureless environments on the lunar surface, we propose a new visual-inertial Simultaneous Localization and Mapping (SLAM) algorithm, which fuses the measurements of vision and the inertial sensor by pose-graph optimization to achieve high-precision self-localization. An optical flow tracking algorithm based on the quadtree method is proposed to address the unbounded front-end visual measurements correlation error in featureless environments. This algorithm can effectively track robust feature points, thereby improving the accuracy of pose estimation between adjacent frames. Moreover, an effective star point removal algorithm is proposed to effectively remove the star points at infinity, which is beneficial to solve localization accuracy decrease caused by unstable landmarks at infinity. A computer simulation system of the lunar surface environment as well as a set of various lunar visual inertial SLAM simulation datasets is further built, and several localization tests in different lunar simulation environments are conducted. Simulation results verify that our algorithm is more robust with better localization accuracy.