基于激光雷达-视觉融合的3D多模态建图研究
Research on 3D Multimodal Mapping Based on Lidar-Vision Fusion
杨旭东 1赖惠鸽 1康文 1王鹏 1陶焓 1李少东1
作者信息
- 1. 宁夏大学大学机械工程学院,宁夏银川 750021
- 折叠
摘要
针对移动机器人室内环境三维地图构建不齐帧、误差大和重建不佳等问题,提出激光雷达和RGB-Depth相机融合(camera radar net,CRN)方法,这是一种新的三维地图构建方法.在CRN中,提出一种雷达-视觉惯性里程计融合(Lidar-Visual Inertial Odometry via Smoothing and Mapping,LVIO-SAM)方法,该方法将优化估计二维移动平台空间位姿.然后通过误差卡尔曼滤波器(Error State Kalman Filter,ESKF)算法将空间位姿数据与轮式里程计进行动态优化,得到良好的建图效果.最后,使用移动机器人进行试验验证.试验结果显示,与激光雷达惯性里程计和视觉惯性里程计相比,所提方法在构建室内环境中,三维地图尺寸误差减少了 22％,里程计精度提高了 0.19％.
Abstract
This paper addresses the challenges of uneven frame construction,large errors,and poor reconstruction of 3D maps for indoor environments using mobile robots.To address these issues,we introduce Camera Radar Net(CRN),a novel 3D map construction method that integrates LiDAR and RGB-Depth cameras.In CRN,a fusion algorithm of Lidar-Visual Inertial Odometry via Smoothing and Mapping(LVIO-SAM)is proposed,which will optimally estimate the spatial pose of the two-di-mensional mobile platform.Then,the spatial pose data and the wheeled odometer are dynamically optimized by the Error State Kalman Filter(ESKF)algorithm to obtain a good mapping effect.Finally,a mobile robot was used for experimental verifica-tion.The experimental results show that compared with lidar inertial odometer and visual inertial odometer,the proposed method reduces the size error of 3D map by 22％and improves the odometer accuracy by 0.19％in the construction of indoor environment.
关键词
三维建图/里程计融合/位姿估计/误差卡尔曼滤波器/移动机器人Key words
3D mapping/odometer integration/pose estimation/error state Kalman filter/mobile robot引用本文复制引用
出版年
2024
NETL
NSTL
万方数据