【摘要】 针对自动驾驶三维建图中存在的建图不准确以及位姿飘移的问题，利用激光雷达里程计消除惯性测量单元（Inertial Measurement Unit, nIMU）累计误差并通过IMU预积分去除激光雷达点云畸变，形成激光雷达与IMU的紧耦合建图系统；通过增加回环检测因子、激光雷达里程计因子以及IMU预积分因子进行后端图优化，旨在提升定位建图的全局一致性，减小位姿估计误差，降低累计漂移误差。最后，在学校园区实地场景以及利用开源数据集KITTI进行实验验证，实验表明，在选取的学校园区实地场景下，改进算法APE误差均值相较于原算法降低了11.04%,APE均方根误差较于原算法降低了17.35%；改进算法在KITTI数据集场景下平均APE误差下降了10.04%，均方误差方面相较于原算法平均下降了12.04%。研究结果表明，改进的建图方法能够有效提高建图的位姿估计精度与地图构建精度。更多还原

【Abstract】 To address the issues of inaccurate mapping and position drift in 3D autonomous driving maps, LIDAR odometry was utilized to counteract cumulative errors of the inertial measurement unit(IMU), and corrections for LIDAR point cloud distortions were made through IMU pre-integration. This approach enabled the creation of a mapping system where LIDAR and IMU were tightly integrated. Subsequently, the back-end map was enhanced by the incorporation of loopback detection, LIDAR odometry, and IMU pre-integration factors, aiming to bolster the global consistency of the positioning map and minimize cumulative drift errors. The optimization of the back-end map sought to enhance global localization consistency, reduce positioning errors, and curtail cumulative drift. Experimental validation was conducted in a school campus environment and with the use of the KITTI open-source dataset. The results demonstrate that in the school campus scenario, an 11. 04% reduction in average APE error and a 17. 35% decrease in RMSE are achieved by the refined algorithm compared to the baseline algorithm. For the KITTI dataset scenario, a reduction of 10. 04% in both average APE error and RMSE, and a 12. 04% decrease in mean square error are observed, underscoring the efficacy of the enhanced mapping technique in elevating position estimation and map construction precision.更多还原