融合激光雷达和RGB-D相机建图
Mapping by integrating LiDAR and RGB-D camera
李少伟 1钟勇 1杨华山 1张树 1范周慧1
作者信息
- 1. 福建省汽车电子与电驱动技术重点实验室,福建 福州 350118
- 折叠
摘要
针对智能小车在未知环境的条件下,利用单一传感器同时定位与地图创建不能准确构建复杂环境地图的问题,提出采用一种RTABMAP算法,用于融合激光雷达和RGB-D相机建图,该算法采集了激光雷达、RGB-D相机和里程计的数据,将其存储在内存管理机制的节点中,提取这些节点的特征.通过匹配节点间的视觉词汇次数更新节点的权重,采用离散贝叶斯滤波估计进行回环检测,优化局部地图,最终构建全局地图.在安装有开源机器人操作系统(ROS)的智能小车上实验.结果表明,本研究方法在障碍物检测率方面与激光建图和RGB-D相机建图方法相比,提高了30.75%和18.63%;地图尺寸误差分别减少了 0.013 和 0.150 m;角度误差分别减少了 3°和 1°.
Abstract
To address the issue of using a single sensor for simultaneous localization and mapping in an un-known environment for intelligent vehicles,a RTABMAP algorithm was proposed for mapping by integrating LiDAR and RGB-D camera.The algorithm collects data from LiDAR,RGB-D camera,and odometer,and stores them in nodes of the memory management mechanism for feature extraction.The weight of the node is updated by matching the number of visual words between the nodes,and the discrete Bayesian filter estimation is used for loop detection to optimize the local map,and finally construct a global map.Experiments were car-ried out on a smart car equipped with an open source robot operating system(ROS).Results show that com-pared with laser mapping and RGB D camera mapping methods,the proposed method improves obstacle detec-tion rate by 30.75%and 18.63%;the map size error has been reduced by 0.013m and 0.150m respectively;the angle error has been reduced by 3 °and 1 °,respectively.
关键词
智能小车/RTABMAP算法/SLAM/传感器融合Key words
intelligent vehicles/RTABMAP algorithm/SLAM/sensor integration引用本文复制引用
出版年
2023
NETL
NSTL
万方数据