相似环境下激光结合视觉的全局定位方法

A Global Localization Method of Laser Combined with Vision in Repetitive Environment

在室内结构相似或含有多几何结构重复区域的环境中,基于激光的全局定位方法很难实现可靠且高效的结果。针对该问题,本文提出一种基于优化的视觉辅助激光定位的方法,仅利用单帧图像信息和激光(LiDAR)信息进行重定位。首先,利用视觉位置识别和BA(Bundle Adjustment)优化来估计机器人的初始姿态。然后,利用视觉初始位姿缩小搜索空间,提高约束计算速度,并将初始估计和激光约束结合验证。最后通过稀疏姿态优化方法构建目标函数,求解全局位姿。实验结果表明,提出的全局定位方法相比单一激光全局定位具有更高的成功率,并且全局定位时间在室内小场景环境和长走廊环境分别减少了56.81%和89.43%,满足了实时性和鲁棒性要求。

In indoor environments with similar indoor structures or repetitive geometric regions,achieving reliable and efficient results using laser-based global localization methods can be challenging.To address this issue,we propose an optimization-based visual-assisted laser localization method that relies solely on single-frame image information and LiDAR data for global localization.Firstly,we employ visual pose estimation and Bundle Adjustment(BA)optimization to estimate the robot's initial pose.Subsequently,we utilize the initial visual pose to narrow down the search space,improving constraint computation speed.We then integrate the initial estimation with laser constraints to validate the pose.Finally,we construct an objective function using sparse pose optimization methods to solve for the global pose.Experimental results demonstrate that the proposed global localization method outperforms single-laser global localization in terms of success rate.Moreover,the global localization time is reduced by 56.81% in small indoor scenes and 89.43% in long corridor environments,meeting real-time and robustness requirements.