基于环境的三维激光实时去动态物体方法  

Real-time removal of dynamic objects by 3D laser based on environment

作　　者：张志强[1,2,3] 张雷[1,2,4] 邹风山 杜振军[4] 刘明敏[4] ZHANG Zhi-qiang;ZHANG Lei;ZOU Feng-shan;DU Zhen-jun;LIU Ming-min(State Key Laboratory of Robotics,Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang 110016,China;Institutes for Robotics and Intelligent Manufacturing,Chinese Academy of Sciences,Shenyang 110169,China;University of Chinese Academy of Sciences,Beijing 100049,China;SIASUN Robot&Automation CO.,Ltd,Shenyang 110168,China)

机构地区：[1]中国科学院沈阳自动化研究所机器人学国家重点实验室,辽宁沈阳110016 [2]中国科学院机器人与智能制造创新研究院,辽宁沈阳110169 [3]中国科学院大学,北京100049 [4]沈阳新松机器人自动化股份有限公司,辽宁沈阳110168

出　　处：《计算机工程与设计》2024年第9期2627-2633,共7页Computer Engineering and Design

基　　金：国家重点研发计划基金项目(2020YFC2007703)。

摘　　要：为提高三维激光传入点云算法框架的点云质量,减少移动物体对点云地图精度的影响,提出一种能够在框架前端基于环境特征的实时去除动态物体的方法。将点云投影得到深度矩阵和高度矩阵(z值矩阵),并从z值矩阵中提取出环境特征;利用环境特征将前后帧点云快速对齐;利用对齐后的前后帧点云得到z值差分矩阵进行筛选,并标记动态点云;以动态点云为中心,利用深度矩阵进行聚类,并将聚类得到的未标记点云进行标记,传入后端时舍弃标记的点云。分别在Ours数据集上以及KITTI数据集上验证了该方法的效果。To improve the point cloud quality of the 3D laser input point cloud algorithm framework and reduce the impact of moving objects on the accuracy of the point cloud map,a method that removed dynamic objects in real time based on environmental characteristics at the front end of the framework was proposed.The point cloud was projected to obtain the depth matrix and z-value matrix,and the environmental features were extracted from the z-value matrix.Environmental features were used to quickly align the front and rear frame point clouds.Taking the dynamic point cloud as the center,the depth matrix was used for clustering,and the unlabeled point cloud obtained by clustering was labeled and the marked point cloud was discarded when it was passed to the backend.The effectiveness of the method is verified on the Ours dataset and the KITTI dataset respectively.

关 键 词：实时 环境特征 高度矩阵 深度矩阵 对齐前后帧 聚类 差分矩阵 

分 类 号：TP249[自动化与计算机技术—检测技术与自动化装置]