退化环境中基于空间几何特征的激光SLAM方法
Laser SLAM method based on spatial geometric features in degraded environment
何登科 1曾天乐 1晏非凡 1何云艳 1杨天娇1
作者信息
- 1. 中国矿业大学(北京) 地球科学与测绘工程学院,北京 100083;煤炭精细勘探与智能开发全国重点实验室,北京 100083
- 折叠
摘要
针对同步定位与地图构建(SLAM)在无全球定位系统(GPS)信号、缺乏环境特征纹理的退化环境中出现定位失败、建图重叠漂移和无法实时运行的问题,提出了一种基于空间几何特征的激光SLAM方法.算法前端设计了一种基于空间线面几何特征的特征点提取方式,依据点线面约束构建点云配准残差函数,采用高斯-牛顿法优化残差以实现点云配准.算法后端基于关键帧构建子图,通过子图间的map-to-map匹配来获得精确位姿,采用插值融合前后端位姿,实现了全局定位优化.仿真与实际退化环境中的实验结果表明:所提算法在 20 m的里程计测试中位姿估算误差小于 5%;退化环境中的建图效果优于Hector、Gmapping和Cartographer算法,地图更新平均速度提高近 4倍.
Abstract
Aiming at the problems of localization failure,mapping overlap drift and non-real-time operation of simultaneous localization and mapping(SLAM)in the degraded environment without global positioning system(GPS)signal and lack of environmental feature texture,a laser SLAM method based on spatial geometric features is proposed.At the front end of the algorithm proposed,a feature point extraction method based on the spatial geometric features of lines and surfaces is designed,and the residual function of point cloud registration is constructed according to the constraints of points,lines and surfaces,and the residual is optimized by Gauss-Newton method for point cloud registration.At the back end of the algorithm,subgraphs are constructed based on keyframes,and precise poses are obtained by map-to-map matching among subgraphs.Interpolation is used to fuse the poses of the front and back ends to achieve global positioning optimization.Extensive experiments conducted in both simulated and real-world degraded environment demonstrate that the pose estimation error of the proposed algorithm remains below 5%within a 20-meter odometer test.The mapping effect under the degraded environment is better than that of Hector,Gmapping and Cartographer algorithms,and the average speed about map updating has increased by nearly 4 times.
关键词
激光同步定位与建图/特征提取/环境感知/退化环境Key words
laser simultaneous localization and mapping/feature extraction/environment perception/degraded environment引用本文复制引用
基金项目
国家自然科学基金(42274194)
高等学校学科创新引智计划(B18052)
中国矿业大学(北京)越崎学者(2019JCA01)
出版年
2024
NETL
NSTL
万方数据