首页|基于筛选策略的动态环境下激光SLAM算法

基于筛选策略的动态环境下激光SLAM算法

Dynamic environment laser SLAM algorithm with a filtering strategy

扫码查看
原文链接
  • 维普
  • 万方数据
现有的同步定位与建图(SLAM)方法在理想条件下运行稳定,但在动态环境中会因移动物体特征点云的误匹配导致定位误差增大.为解决此问题,提出了一种动态点云检测算法.首先利用惯性测量装置信息对点云数据预处理,包含去畸变等操作;然后剔除地面点云,采用弯曲体素结构对非地面点云进行聚类;接着,通过匈牙利算法关联和匹配两帧之间的聚类,同时利用惯性信息统一坐标系;最后设计一种筛选策略,先用边界框交并比和质心速度粗略筛选动态聚类,再用z轴(高度)分布相似性进行精细筛选.实验结果表明,所提算法能够识别并滤除实验环境中的大部分动态点云聚类;与LIO-SAM 算法相比,四种场景下的定位均方根误差平均降低了 17.75%;平均精确率和召回率相比Removert分别提升 14.81%和 5.90%.
Most existing simultaneous localization and mapping(SLAM)methods perform stably under ideal conditions,but the localization accuracy decreases in dynamic environment due to mismatches of feature point clouds caused by moving objects.To address this issue,a dynamic point cloud detection algorithm is proposed.Firstly,the inertial measurement unit information is used to preprocess the point cloud data,including operations such as distortion removal.Secondly,the ground point clouds are eliminated,and the non-ground point clouds are clustered by using the curvature voxel structure.Thirdly,the Hungarian algorithm is used to associate and match the clusters between the two frames,while unifying the coordinate system with inertial information.Finally,a screening strategy is designed:coarse screening of dynamic clusters using the intersection-over-union of bounding boxes and centroid velocity,followed by fine screening based on the similarity of point cloud distribution along the z-axis(height).Experimental results show that the proposed algorithm can effectively identify and filter out most dynamic point cloud clusters in the environment.Compared to the LIO-SAM algorithm,the root mean square error of localization in the four scenarios is reduced by 17.75%on average.The average accuracy and recall rate are 14.81%and 5.90%higher than Removert,respectively.

LiDAR simultaneous localization and mappingdynamic environmentdynamic point cloud detectionfiltering strategy

徐晓苏、王睿、姚逸卿

展开 >

微惯性仪表与先进导航技术教育部重点实验室,南京 210096

东南大学 仪器科学与工程学院,南京 210096

激光同步定位与建图 动态环境 动态点云检测 筛选策略

国家自然科学基金装备重大基础研究项目教育部"春晖计划"合作科研项目中央高校基本科研业务费专项中央高校基本科研业务费专项

6192100451405-02A03HZKY202201282242023K300052242023K30006

2024

10.13695/j.cnki.12-1222/o3.2024.07.006

中国惯性技术学报
中国惯性技术学会

中国惯性技术学报

CSTPCD北大核心
影响因子：0.792
ISSN：1005-6734
年,卷(期)：2024.32(7)