【摘要】 针对激光SLAM在城市测绘中存在的累积误差大、稳健性差的问题,本文提出了一种面向城市高精度制图的激光紧耦合SLAM方法,该方法引入杆状和面状特征进行点云配准,降低了城市环境下SLAM的累积误差,并通过GNSS角点位置约束,提高了全局地图构建的准确性。本文在4组城市常见场景(开放园区、地下车库、城市公园、街区道路)中对所提方法进行了验证,并与目前主流的LOAM、LeGO-LOAM和LIO-SAM方法进行了对比,试验结果表明LOAM和LeGO-LOAM在复杂城市场景中稳定性较差,LIO-SAM和本文所提方法成功实现了4组场景的制图。与LIO-SAM相比,本文所提方法仅采用激光惯导紧耦合时,轨迹绝对位置误差较LIO-SAM降低了32.25%,结合GNSS位置因子后进一步降低了92.03%(轨迹精度均优于10 cm)。此外,开放园区的控制点精度评定表明本文所提方法的点云绝对坐标精度优于5 cm。更多还原

【Abstract】 Aiming to reduce the cumulative error and improve the robustness of SLAM system in accurate urban mapping, a tightly coupled laser SLAM algorithm that combined LiDAR, inertial measurement unit(IMU), and global navigation satellite system(GNSS) was developed. The proposed method achieved high accuracy point cloud registration by adding pole-like and plane features that reduced cumulative errors in SLAM. In addition, a GNSS corner-based constraint was used to improve the accuracy of the global map construction. This study compared the proposed method with three mainstream SLAM methods(i.e., LOAM, LeGO-LOAM, and LIO-SAM) in four common urban scenes(i.e., open park, underground garage, urban park, and road). The test results showed that LOAM and LeGO-LOAM have poor stabilities in complex urban scenes. The LIO-SAM and proposed method have successfully realized the mapping of all scenes. Compared to LIO-SAM, the absolute position error(APE) of the proposed method has improved by 32.25% without the GNSS position factor and has improved by 92.03% with the GNSS position factor(APE<10 cm). Moreover, the absolute coordinate error of generated point cloud by the proposed method in the open park scene was less than 5 cm, which demonstrates the proposed method can fulfill the requirements of centimeter-level urban mapping.更多还原