线面特征结合的三维激光SLAM算法及性能评估

3D LiDAR SLAM algorithm and performance evaluation based on edge and surface features

针对自动驾驶的高精度定位中,传统激光同步定位和制图(SLAM)点云配准采用的迭代最临近点法(ICP)计算效率低、易误匹配的问题,提出1种3维激光SLAM点云配准方法:通过基于特征的ICP变种算法,对在不同时刻扫描的重叠点云进行配准;并分析有无全球卫星导航系统(GNSS)数据融合下的激光SLAM自主定位性能。实验结果表明,纯激光SLAM的位置误差存在线性发散,其值约为行车里程的1.98%,融入GNSS数据后,可以提高SLAM全局位姿的一致性,达到分米级定位精度。

Aiming at the problems of low computational efficiency and mismatching of the Iterative Closest Point(ICP)method used by the traditional Light Detection and Ranging(LiDAR)Simultaneous Location and Mapping(SLAM)point cloud registration algorithm in the high-precise positioning of automatic driving,the paper proposed a point cloud registration method of 3D LiDAR SLAM:the overlapping point clouds scanned at different times were registered by ICP variant algorithm based on features;and the autonomic positioning performances of LiDAR SLAM with or without Global Navigation Satellite System(GNSS)fusion were analyzed.Experimental result showed that the linear divergence of the positioning error of LiDAR SLAM could be 1.98%of the driving mileages,while the integration of GNSS data could help improve the overall position and attitude consistency of SLAM with the positioning accuracy of decimeter level.