摘要
针对传统激光同时定位与建图在动态环境中位姿估计累计误差大、地图中存在动态目标错误点云的问题,本文提出了一种基于可视点法实时剔除动态目标的激光-惯导SLAM方法(DM-LIO)。该方法使用IMU测量值为基于可视点法的动态目标剔除模块提供先验位姿,并引入基于弯曲体素空间的点云聚类方法,以解决在低分辨率可视点法下动态点不能被完全捕捉的问题,从而实现了在算法前端剔除激光点云中的动态目标。本文通过自主搭建室内真机实验平台和使用公开数据集两种方式对算法性能进行评估。真机实验结果表明本文提出的DM-LIO能够对多个动态目标以及非先验动态目标进行实时剔除;在公开数据集Urbanloco上的测试结果表明,在高动态的环境下DM-LIO的绝对轨迹误差相较于LIO-SAM减少了60%以上,验证了该算法在高动态环境中具有良好的定位精度。
To address the problems of conventional LiDAR simultaneous localization and mapping(SLAM)in dynamic environments with large cumulative errors in pose estimation and dynamic object error point clouds in the map,this paper presents a tightly coupled LiDARinertial SLAM(DM-LIO)method for real-time removal of dynamic objects based on the visible point method.This method by utilizes the IMU measurements to provide a priori poses for the dynamic object removal module based on the visible point method,and also introduces a point cloud clustering method based on curved voxel space to solve the problem that dynamic points of viewable point method cannot be fully captured at low resolutions,which enables the rejection of dynamic objects in laser point clouds at the front end of algorithm.The performance of algorithm is evaluated by both building a real indoor experimental platform and using a public dataset.The results of real-world experiments show that the proposed DM-LIO method is able to remove multiple dynamic objects as well as non-a priori dynamic objects online;the test results based on the public dataset of Urbanloco show that the absolute trajectory error of DM-LIO is reduced by more than 60%compared to LIO-SAM in highly dynamic environments,which verifies that the algorithm possesses good positioning accuracy in a highly dynamic environment.
作者
陈耀华
何丽
王宏伟
冉腾
刘哲凝
Chen Yaohua;He Li;Wang Hongwei;Ran Teng;Liu Zhening(College of Intelligent Manufacturing and Modern Industry,Xinjiang University,Urumqi 830047,China)
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2023年第9期246-254,共9页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(62063033)
新疆维吾尔自治区重点研发计划项目(2022B01050-2)资助。
