面向点云退化的隧道环境的无人车激光SLAM方法

A laser SLAM method for unmanned vehicles in point cloud degenerated tunnel environments

基于激光同时定位与地图构建(simultaneous localization and mapping,SLAM)技术,不仅能够实现车辆在未知环境下的实时定位,还能高效地获取环境的三维地理空间信息,近年来受到了无人驾驶领域的广泛关注。在几何结构匮乏的隧道中,仅依赖几何信息无法配准点云,因此传统激光SLAM方法难以在隧道中应用。为解决这一问题,本文在LOAM的基础上,提出一种点云强度信息增强的改进激光SLAM技术。首先,改进特征提取方法,提出基于点云柱面投影图的自适应特征提取方法,从单帧点云中提取直线、平面、地面和反射强度特征;其次,针对点云的几何特征配准在隧道中的退化问题,提出一种基于强度特征和强度地图的配准优化方法,自适应提取环境中的强度特征并根据强度特征的配准对车辆位姿进一步修正。试验结果表明,该方法较LOAM和HDL-Graph-SLAM具有更好的稳健性,能够在缺乏几何特征但强度特征丰富的隧道内实现稳定的定位和地图构建,定位精度提升了一个数量级。

Laser SLAM enables to locate the vehicle itself even in an unknown environment,and to efficiently sample the three-dimensional geospatial information of the traversed environment,which has been drawn wide attention in the field of autonomous driving in recent years.To improve the accuracy and performance of the laser SLAM system in a point cloud degenerated tunnel environment,we present an intensity enhanced laser SLAM approach based on LOAM.First,we improve the feature extraction module of LOAM.An adaptive feature extraction method based on spherical projection image is presented to extract line,facde,ground and reflectors from a single laser sweep.Besides,to solve the issue on point cloud registration degeneracy in tunnel environments,we presented intensity feature-basedregistration approach to fix the vehicle pose resulting from the geometric feature-based registration error.Reflecting features in the surrounding are adaptively extracted to ensure the adaptivity of our improved laser SLAM approach.The experimental results show that the proposed method presented the better and more robust performance especially in degenerated environments,e.g.,long straight tunnel,comparing with the performance of LOAM and HDL-Graph-SLAM.The accuracy of the proposed method was an order of magnitude larger than that of LOAM and HDL-Graph-SLAM.