基于激光雷达-视觉融合的3D多模态建图研究  

Research on 3DMultimodal Mapping Based on Lidar-Vision Fusion

作　　者：杨旭东 赖惠鸽[1] 康文 王鹏 陶焓 李少东 Yang Xudong;Lai Huige;Kang Wen;Wang Peng;Tao Han;Li Shaodong(School of Mechanical Engineering,Ningxia University,Yinchuan 750021,Ningxia,China)

机构地区：[1]宁夏大学大学机械工程学院,宁夏银川750021

出　　处：《应用激光》2024年第4期113-122,共10页Applied Laser

基　　金：国家自然科学基金资助项目(51765056)。

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%。This paper addresses the challenges of uneven frame construction,large errors,and poor reconstruction of 3Dmaps for indoor environments using mobile robots.To address these issues,we introduce Camera Radar Net(CRN),a novel 3Dmap 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-dimensional 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 verification.The experimental results show that compared with lidar inertial odometer and visual inertial odometer,the proposed method reduces the size error of 3Dmap 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 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]