基于DDP的激光回环检测算法  

Laser loop closure detection based on DDP descriptor

作　　者：吴青贵 杨盛毅[2,3] 朱力 何小飞 WU Qinggui;YANG Shengyi;ZHU Li;HE Xiaofei(School of Data Science and Information Engineering,Guizhou Minzu University,Guiyang 550025,China;Key Laboratory of Pattern Recognition and Intelligent Systems of Guizhou Province,Guizhou Minzu University,Guiyang 550025,China;School of Physics and Mechatronics Engineering,Guizhou Minzu University,Guiyang 550025,China)

机构地区：[1]贵州民族大学数据科学与信息工程学院,贵阳550025 [2]贵州民族大学贵州省模式识别与智能系统重点实验室,贵阳550025 [3]贵州民族大学物理与机电工程学院,贵阳550025

出　　处：《智能计算机与应用》2024年第4期102-107,共6页Intelligent Computer and Applications

基　　金：贵州省科学技术基金(黔科合基础[2017]1088)。

摘　　要：回环检测作为SLAM技术重要组成部分,能够提高轨迹估计的全局一致性。针对目前效果较好的扫描上下文算法(SC),本文提出一种基于DDP改进的回环检测算法。首先,根据点云分布对齐点云坐标系;其次,利用方位角与尺度信息构建DDP点云描述符;最后,提出高度维度描述符余弦距离与离散度描述符相关距离加权方法,计算候选帧DDP点云描述符相似度。实验结果表明,本文所提出的基于DDP改进的回环检测算法在KITTI数据集大多数序列中,精确率与召回率优于IRIS、NDD、SC和ISC算法,且能够满足激光雷达实时性要求。Loop closure detection,as an important part of SLAM technology,can improve the global consistency of trajectory estimation.In response to the issues existing in the currently effective Scan Context algorithm(SC),this paper proposes a loop closure detection algorithm based on DDP improvement.Firstly,align the point clouds coordinate system based on the point cloud distribution;secondly,construct DDP point cloud descriptors using azimuth and scale information;finally,propose a weighted method for cosine distance of height-dimensional descriptors and related distance of discreteness descriptor to calculate the similarity of candidate frame DDP point cloud descriptors.Experimental results show that the loop closure detection algorithm based on DDP improvement performs better in most sequences of the KITTI dataset,with higher precision and recall rates than IRIS,NDD,SC,and ISC algorithms,and can meet the real-time requirements of LiDAR.

关 键 词：SLAM 回环检测算法 点云描述符 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]