测绘通报 ›› 2019, Vol. 0 ›› Issue (3): 16-20.doi: 10.13474/j.cnki.11-2246.2019.0070

• 学术研究 • 上一篇    下一篇

一种用于图像特征提取的改进ORB-SLAM算法

张良桥1,2, 陈国良1,2, 许晓东1, 连达军3, 王睿1,2   

  1. 1. 中国矿业大学环境与测绘学院, 江苏 徐州 221116;
    2. 中国矿业大学国土环境与灾害监测国家测绘 地理信息局重点实验室, 江苏 徐州 221116;
    3. 苏州科技大学环境学院, 江苏 苏州 215009
  • 收稿日期:2018-05-24 修回日期:2018-12-27 出版日期:2019-03-25 发布日期:2019-04-02
  • 通讯作者: 陈国良。E-mail:chglcumt@163.com E-mail:chglcumt@163.com
  • 作者简介:张良桥(1994-),男,硕士生,研究方向为室内定位及视觉SLAM。E-mail:zhanglqcumt@163.com
  • 基金资助:
    国家重点研发计划(2016YFB0502105);江苏省自然科学基金(BK20161181);国家自然科学基金(41371423);江苏高校品牌专业建设工程(PPZY2015B144)

An improved ORB-SLAM algorithm for feature extraction

ZHANG Liangqiao1,2, CHEN Guoliang1,2, XU Xiaodong1, LIAN Dajun3, WANG Rui1,2   

  1. 1. School of Environment Science and Spatial Informatics, China University of Mining and Technology, Xuzhou 221116, China;
    2. NASG Key Laboratory of Land Environment and Disaster Monitoring, China University of Mining and Technology, Xuzhou 221116, China;
    3. Suzhou University of Science and Technology, Suzhou 215009, China
  • Received:2018-05-24 Revised:2018-12-27 Online:2019-03-25 Published:2019-04-02

摘要: 针对复杂室内环境下视觉SLAM定位存在实时性差、轨迹漂移等问题,本文提出了一种基于图像特征提取方法的ORB-SLAM算法。该算法在前端中提高图像特征检测与匹配的效率和精度,引入闭环检测策略优化相机位姿轨迹,提高定位精度。以不同来源图像对比分析不同特征提取算法SIFT、SURF、ORB的有效性,运用该算法估计机器人运动轨迹,与真实轨迹相对位姿误差为0.144 8 m,试验表明所提出的方法切实可行,具有较高的稳健性。

关键词: 视觉SLAM, 图像特征检测与匹配, ORB-SLAM, 回环检测, 位姿估计

Abstract: Aiming at the problem of poor real-time performance and trajectory drift in visual SLAM positioning in complex indoor environments,this paper proposes an ORB-SLAM algorithm based on image feature detection extraction method.The algorithm improves the efficiency and accuracy of image feature detection and matching in the front-end,introduces a closed-loop detection strategy to optimize camera pose trajectory,and improves positioning accuracy.The SIFT,SURF and ORB of different feature extraction algorithms are compared and analyzed in different sources.The robot motion trajectory is estimated by this algorithm.The relative pose error from the real trajectory is 0.144 8 m.Experiments show that the proposed method is feasible and robustness.

Key words: V-SLAM, image feature detection and matching, ORB-SLAM, loop detection, pose estimation

中图分类号: