• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2022, Vol. 58 ›› Issue (14): 161-169.doi: 10.3901/JME.2022.14.161

• 特邀专栏:大型构件视觉测量与机器人加工 • 上一篇    下一篇

扫码分享

用于引导机器人定位的汇聚式双目视觉算法实现

刘红帝1,2, 吕睿1,2, 田林雳1,2, 朱大虎1,2   

  1. 1. 武汉理工大学现代汽车零部件技术湖北省重点实验室 武汉 430070;
    2. 武汉理工大学汽车零部件技术湖北省协同创新中心 武汉 430070
  • 收稿日期:2021-05-02 修回日期:2021-09-15 出版日期:2022-07-20 发布日期:2022-09-07
  • 通讯作者: 田林雳(通信作者),女,1990年出生,博士,讲师。主要研究方向为机器视觉与汽车设计。E-mail:tll@whut.edu.cn
  • 作者简介:刘红帝,男,1995年出生,博士研究生。主要研究方向为光学成像与机器人视觉伺服。E-mail:stevenliu@whut.edu.cn
  • 基金资助:
    国家自然科学基金(51975443)和湖北省重点研发计划(2020BAA025)资助项目。

Realization of Convergent Binocular Vision Algorithm for Guiding Robot Localization

LIU Hongdi1,2, Lü Rui1,2, TIAN Linli1,2, ZHU Dahu1,2   

  1. 1. Hubei Key Laboratory of Advanced Automotive Components Technology, Wuhan University of Technology, Wuhan 430070;
    2. Hubei Collaborative Innovation Center for Automotive Components Technology, Wuhan University of Technology, Wuhan 430070
  • Received:2021-05-02 Revised:2021-09-15 Online:2022-07-20 Published:2022-09-07

摘要: 大型复杂构件机器人加工品质高度依赖于视觉引导的机器人定位精度。针对现有平行式双目相机视场范围有限、扫描-重构算法复杂等问题,提出一种汇聚式双目立体视觉算法用于引导加工型机器人进行精确定位。具体包括利用小孔成像原理结合三角形相似性建立汇聚式双目相机间成像模型,实现自主标定、特征匹配以及标记点空间位置信息提取;将数据通过手眼标定算法转换到机器人基坐标系下用于校正工件三维模型;进而自动规划运动路径,引导机器人对工件待加工区域进行精准定位加工。分别以规则陶瓷块和复杂铁轨两种工件进行试验验证,结果表明在视场范围内任意位姿摆放的工件平均定位误差均可控制在4%以内,与传统双目视觉算法相比具有定位精度高、运行速度快等优点,具有很好的应用前景。

关键词: 汇聚式双目视觉, 立体标定, 特征匹配, 模型校正, 机器人定位

Abstract: The machining quality of large and complex components highly depends on the visually-guided robot localization accuracy. Aiming at the problems of limited view field of existing parallel binocular cameras and complex scanning-reconstruction algorithms, a convergent binocular stereo vision algorithm is proposed to guide robot for precise localization. Firstly, both small hole imaging principle and triangle similarity are used to establish the imaging model of convergent binocular camera for the realization of autonomous calibration, feature matching, and extraction of the spatial position information of mark points. Secondly, the data is converted to the robot base coordinate system through the hand-eye calibration algorithm for correcting the three-dimensional model of the work-piece. Thirdly, the motion path is automatically planed to guide the robot for accurate localization. Two work-pieces, regular ceramic block and complex rail, are used for experimental verification. The results show that the average localization error of work-pieces placed in any posture within the field of view can be controlled within 4%, showing the advantages of high precision and fast running speed than the typical binocular vision algorithms, and has good application prospect.

Key words: convergent binocular vision, stereo calibration, feature matching, model correction, robot localization

中图分类号: