基于高效回环检测的大场景下的SLAM算法
SLAM Algorithm in Large Scenes Based on Efficient Loop Detection
杨娜 1程磊1
作者信息
- 1. 沈阳理工大学 信息科学与工程学院,沈阳 110159
- 折叠
摘要
在室外大场景情况下,激光里程计累计误差会随着时间的增加而逐渐增加.针对现有激光同步定位与地图构建(SLAM)算法精度低、鲁棒性不足等问题,提出一种基于精确高效回环检测的大场景下的激光SLAM算法.该算法以激光里程计框架LeGO-LOAM为基础,改进回环检测部分,使用关键特征(Ring-Key)描述子构建KD-tree查找相似帧加速搜索,回环帧间匹配使用激光雷达虹膜描述子计算汉明距离判断相似度,避免暴力匹配,实现平移和旋转不变性.在KITTI数据集上进行仿真实验,结果表明:与原LeGO-LOAM算法相比,改进算法绝对位姿误差标准差平均降低了51.54%,相对位姿误差标准差平均降低了14.42%.
Abstract
In the case of outdoor large scenes,the cumulative error of laser odometry will gradually increase with the increase of time.In view of the problems of low accuracy and lack of robustness of existing lidar simultaneous localization and mapping(SLAM)algorithms,a lidar SLAM algo-rithm based on accurate and efficient loop detection in large scenes is proposed.Based on the LeGO-LOAM lidar odometry framework,the loop detection part is improved,and the Ring-Key de-scriptor is used to construct KD-tree to find similar frames to accelerate the search.The lidar iris de-scriptor is used to calculate the Hamming distance to judge the similarity between loop frames,so as to avoid violent matching and realize the translation and rotation invariance.Simulation experiments on KITTI dataset show that compared with LeGO-LOAM algorithms,the absolute pose accuracy and relative pose accuracy of this algorithm are improved by 51.54%and 19.39%on average respec-tively.
关键词
激光雷达/回环检测/LeGO-LOAM/同步定位与地图构建Key words
lidar/loop detection/LeGO-LOAM/simultaneous localization and mapping引用本文复制引用
基金项目
国家重点研发计划项目(2022YFC3302500)
出版年
2024
NETL
NSTL
万方数据