基于FastDTW地磁匹配的快速闭环检测方法研究

Fast closed loop detection method research based onFastDTW geomagnetic matching

为了有效地解决大尺度建图时闭环检测匹配速度慢和闭环误报的问题,提出了一种FastDTW地磁匹配和激光雷达(LiDAR)点云匹配相结合的快速闭环检测方法。通过地磁匹配算法对地图中的位姿节点进行了过滤,极大地减小了闭环检测时的搜索空间,并且对搜索空间的约束有助于减少激光雷达即时定位与建图(SLAM)中由于高局部相似性引起的误检。通过在真实环境中收集激光雷达点云数据和IMU的三轴地磁数据进行实验,验证对比了逐帧法、分支定界法、动态时间规整(DTW)融合算法、FastDTW融合算法的闭环检测速度、误检次数和建图平均绝对误差。实验结果表明:与DTW融合算法相比,本文算法提高了17%的闭环检测速度,降低了16%的平移平均绝对误差,降低了10%的旋转平均绝对误差,并在四种算法中保持较低的误检次数。

In order to solve the problems of slow matching speed and false alarm of closed-loop detection in large-scale mapping,a fast closed-loop detection method combining FastDTW geomagnetic matching and LiDAR point cloud matching is proposed.The geomagnetic matching algorithm is used to filter the pose nodes in the map,which greatly reduces the search space in closed-loop detection,and the constraint on the search space helps to reduce the false detection caused by high local similarity in LiDAR simultaneous localization and mapping(SLAM).By collecting LiDAR point cloud data and IMU three-axis geomagnetic data in the real environment,the closed-loop detection speed,false detection times and map average absolute error of frame-by-frame method,branch and bound method,DTW fusion algorithm and FastDTW fusion algorithm are verified and compared.Experimental results show that,compared with DTW fusion algorithm,the proposed algorithm improves the closed-loop detection speed by 17%,reduces the average absolute error of translation by 16%,and reduces the average absolute error of rotation by 10%,and keeps a low number of false detections in the four algorithms.