基于SR-Context的激光雷达点云闭环检测算法  

A SR-Context Loop-Closure Detection Algorithm of Lidar Point Clouds

作　　者：李炯 邵金菊[2] 王任栋 赵凯 梁震 Li Jiong;Shao Jinju;Wang Rendong;Zhao Kai;Liang Zhen(95848 Army of P.L.A.,Xiaogan,Hubei 432019,China;School of Transportation and Vehicle Engineering,Shandong University of Technology,Zibo,Shandong255000,China;Institute of Military Transportation,Army Military Transportation University,Tianjin 300161,China)

机构地区：[1]中国人民解放军95848部队,湖北孝感432019 [2]山东理工大学交通与车辆工程学院,山东淄博255000 [3]陆军军事交通学院军事交通运输研究所,天津300161

出　　处：《光学学报》2021年第22期207-217,共11页Acta Optica Sinica

基　　金：天津市应用基础与前沿技术研究计划(15JCQNJC01000)。

摘　　要：针对传统激光雷达闭环检测算法受动态障碍物干扰较大、关键帧搜索以及特征匹配耗时较长的问题,采用MF-RANSAC算法并改进ScanContext,提出一种鲁棒性更强、耗时更短的SR-Context激光雷达闭环检测算法。首先,利用区域生长算法对扇形栅格化后的点云进行分割;随后,不依赖于目标识别和跟踪,借助动态区域多点选取和多属性查询对应点,提出一种MF-RANSAC算法快速实现动态目标剔除;最后,通过简化特征匹配计算和删除闭环历史匹配帧的方式改进ScanContext算法,对去除动态目标后的栅格提取特征实现闭环检测。分别在KITTI数据集与实车数据集下进行测试,实验结果表明,本文算法在城市动态环境下能够快速准确实现闭环检测进而提高激光雷达建图精度,且平均耗时仅为ScanContext算法耗时的40%。Traditional light detection and ranging(lidar)loop-closure detection algorithms are greatly interfered with by dynamic obstacles,and key-frame search and feature matching take a long time.In response,this paper proposed a less time-consuming SR-Context lidar loop-closure detection algorithm with stronger robustness based on the multiple-features random sample consensus(MF-RANSAC)algorithm and an improved ScanContext algorithm.Firstly,the region growing algorithm was used to cluster the point clouds that had undergone fan-shaped rasterization.Then,an MF-RANSAC algorithm was proposed to eliminate dynamic targets quickly.This algorithm was based on multi-point selection in a dynamic region and query of corresponding points with multiple attributes rather than target recognition and tracking.Finally,the ScanContext algorithm was improved by simplifying the feature matching calculation and deleting the loop-closure historical matching frames.Loop-closure detection of the point clouds of the current frame after elimination of dynamic targets was thus achieved.Tests were carried out on a KITTI dataset and a real vehicle dataset.The experimental results show that the proposed method delivers quick and accurate loop-closure detection in dynamic urban environments and thereby improves lidar mapping accuracy.The average time it takes is only 40%of that of the ScanContext algorithm.

关 键 词：传感器 闭环检测 智能车 动态实时定位与建图 激光雷达 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]