体素化广义迭代最近点的回环检测算法研究

Research on loop detection algorithm for voxelized generalized iterative closest point

同步定位与地图构建(SLAM)是大规模定位和地图构建的关键技术之一,针对室外大场景环境建图时存在累计误差较大导致定位精度不高、地图出现重影和漂移的问题,提出了一种基于体素化广义迭代最近点(VGICP)优化的回环检测算法。该方法扩展了广义迭代最近点(GICP)算法,计算体素内局部多个点,在保证准确性的同时避免了代价高昂的最邻近搜索。将所提方法加入扫描上下文(SC)-LeGO—激光雷达测距和实时测绘(LOAM)完整框架中,并利用KITTI数据集05序列测试。实验结果表明:优化算法估计的轨迹和真实轨迹有较高的重合度,绝对位姿误差(APE)和相对姿态误差(RPE)的最大值分别下降了46.4%,18.8%;均方值误差下降了17.7%,19.9%;优化算法可以进一步提高建图精度并降低姿态漂移误差。

Synchronous localization and mapping(SLAM)is one of the key technologies for large-scale localization and mapping.Aiming at the large cumulative error problem in outdoor large scene environment mapping,which leads to low positioning precision and map ghosting and drift,a loop detection algorithm based on voxelized generalized iterative closest point(VGICP)optimization is proposed.This method extends the generalized iterative closest point(GICP)algorithm to calculate multiple local points in voxels,which ensures accuracy and avoids costly nearest neighbor search.The proposed method is added to the complete SC-LeGO-LiDAR odometry and mapping in real-time(LOAM)framework and tested using the KITTI dataset 05 sequence.The experimental results show that the trajectory estimated by the optimization algorithm has a high coincidence with the real trajectory,and the maximum absolute pose error(APE)and relative pose error(RPE)have decreased by 46.4%and 18.8%respectively;The mean square error decreases by 17.7%and 19.9%.The optimization algorithm can further improve the mapping precision and reduce the attitude drift error.