• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2024, Vol. 60 ›› Issue (6): 296-305.doi: 10.3901/JME.2024.06.296

• 运载工程 • 上一篇    下一篇

扫码分享

结构化环境下基于点线特征的VI-SLAM系统

郭翼彪1, 周云水1, 黄圣杰1, 刘硕1, 谢国涛1,2, 秦晓辉1,2   

  1. 1. 湖南大学机械与运载工程学院 长沙 410082;
    2. 湖南大学无锡智能控制研究院 无锡 214072
  • 收稿日期:2023-05-08 修回日期:2023-11-28 出版日期:2024-03-20 发布日期:2024-06-07
  • 通讯作者: 秦晓辉,男,1988年出生,博士,副研究员。主要研究方向为智能汽车,面向自动驾驶的SLAM。E-mail:qxh8880507@163.com
  • 作者简介:郭翼彪,男,1998年出生。主要研究方向为视觉SLAM、多传感器融合的SLAM。E-mail:biaoyg2580@163.com
  • 基金资助:
    国家自然科学基金(52102456, 52172384)、长沙市自然科学基金(kq2202162)和汽车车身先进设计制造国家重点实验室开放课题(32115013)资助项目。

VI-SLAM System Based on Point-line Features in Structured Environment

GUO Yibiao1, ZHOU Yunshui1, HUANG Shengjie1, LIU Shuo1, XIE Guotao1,2, QIN Xiaohui1,2   

  1. 1. College of Mechanical and Vehicle Engineering, Hunan University, Changsha 410082;
    2. Wuxi Intelligent Control of Research InstituteWICRI of Hunan University, Wuxi 214072
  • Received:2023-05-08 Revised:2023-11-28 Online:2024-03-20 Published:2024-06-07

摘要: 现阶段室内等结构化环境中的机器人定位技术备受关注,视觉-惯性传感器融合的同时定位与建图(visual-inertial simultaneous localization and mapping,VI-SLAM)系统凭借其成本低、体积小、互补性高等优点得到广泛应用。针对现有VI-SLAM系统中相机和IMU的旋转外参在线初始化困难、室内环境中的结构化特征利用不充分等问题,提出一种结构化环境下基于点线特征的双目VI-SLAM系统。该系统基于结构化环境中的线特征,采用先静止、后运动的两步法,在线初始化相机和IMU之间的旋转外参,并通过融合视觉提供的点、线特征的重投影误差约束和IMU提供的预积分约束,共同优化定位系统的状态量。在EuRoC室内无人机数据集和真实地下停车场中的试验表明,两步初始化旋转外参算法有效且准确,可为优化环节提供良好的初始值,通过与多种视觉定位算法进行对比,验证了该系统拥有更高的定位精度。

关键词: 同时定位与建图, 视觉-惯性, 点线特征, 在线初始化

Abstract: Robot localization technology in structured environments such as indoor has attracted much attention at present. The visual-inertial simultaneous localization and mapping (VI-SLAM) system has been widely used with its low-cost, small-size and high-complementarity. A stereo VI-SLAM system based on point-line features in structured environment is proposed to overcome the difficulty in camera-IMU extrinsic online calibration and insufficient utilization of structured features in the existing VI-SLAM system. Based on the line features in the structured environment, the system uses a two-step method of first stationary and then moving to online initialize the camera-IMU extrinsic parameters, and jointly optimizes the state variables of the localization system by fusing the re-projection error constraints of the point-line features provided by vision and the pre-integration constraints provided by IMU. Experiments on EuRoC indoor UAV datasets and real underground parking lot show that the two-step initialization extrinsic parameters algorithm is effective and accurate to provide good initial value for optimization. Compared with other localization algorithms, the system has higher localization accuracy.

Key words: SLAM, visual-inertial, point-line feature, online initialization

中图分类号: