中国机械工程 ›› 2026, Vol. 37 ›› Issue (9): 2101-2114.DOI: 10.3969/j.issn.1004-132X.2026.09.003

• 智能诊断与运维专栏 • 上一篇    

轴孔类零件单目内窥成像位姿检测研究

李蓉(), 金子源, 王华建, 刘坚, 尹韶辉   

  1. 湖南大学国家高效磨削工程技术研究中心, 长沙, 410082
  • 收稿日期:2024-08-22 修回日期:2026-08-06 出版日期:2026-09-25 发布日期:2026-10-21
  • 通讯作者: 李蓉
  • 作者简介:李蓉*(通信作者),女,1976年生,副教授。研究方向为视觉检测与质量控制、故障诊断与信号处理。E-mail:jyylirong@hnu.edu.cn。
  • 基金资助:
    湖南省科技创新计划(2023GK2008);湖南省科技创新计划(2023RC1046);岳麓山工业创新中心项目(2023YCII0112);国家重点研发计划(2024YFB3411400)

Research on Pose Detection of Shaft and Hole Parts Based on Monocular Endoscopic Imaging

LI Rong(), JIN Ziyuan, WANG Huajian, LIU Jian, YIN Shaohui   

  1. National Engineering Research Center for High Efficiency Grinding,Hunan University,Changsha,410082
  • Received:2024-08-22 Revised:2026-08-06 Online:2026-09-25 Published:2026-10-21
  • Contact: LI Rong

摘要:

为保障复杂轴类零件内孔表面缺陷检测的内窥成像质量并避免内窥镜与工件碰撞划伤,提出了一种轴孔类零件单目内窥成像位姿检测方法。针对单目内窥成像干扰轮廓多的特点,提出一种基于长度约束与最小包络圆约束的干扰轮廓去除方法。基于改进最大类间方差阈值分割方法对图像进行阈值分割,采用Suzuki轮廓提取算法进行边缘轮廓提取,采用所提出的干扰轮廓去除方法进行目标轮廓筛选并采用径向搜索提取目标内孔轮廓点集。针对轮廓点集离散且存在误差点,以及内窥光源位姿偏移较大时光照不均导致目标轮廓不完整的问题,提出了一种基于轮廓恢复的椭圆检测算法。针对位姿求解的二义性问题,基于成像仅有圆特征、灰度信息丰富且位姿检测时由机械臂移动工件、内孔中心靠近光轴的特点,提出一种基于角度约束与灰度分布的二义性去除方法。实验结果验证了提出的轴孔类零件单目内窥成像视觉位姿检测方法的有效性,该方法在X、Y轴方向上的位置检测精度为0.062 mm,绕X、Y轴旋转的姿态检测精度在1.4°以内,均达到了企业的要求。

关键词: 位姿检测, 内窥成像, 椭圆检测, 二义性去除, 边缘提取, 图像分割

Abstract:

To ensure the quality of endoscopic imaging for detecting surface defects in inner holes of complex shaft parts and to avoid collision and scratching between the endoscope and the workpiece, a method was proposed herein for detecting the pose of shaft-hole parts based on monocular endoscopic imaging. Considering the characteristics of monocular endoscope imaging with many interference contours, a method was proposed for removing interference contours based on length constraint and minimum enclosing circle constraint. The improved Otsu method was adopted for image threshold segmentation, the Suzuki contour extraction algorithm was chosen for edge contour extraction, the proposed interference contour removal method was used for target contour selection, and the radial search was employed to extract the contour point set of the target inner hole. Considering that the contour point set was discrete and contained error points, and that incomplete target contours were caused by uneven illumination when the endoscopic light source was significantly offset in position and orientation, an ellipse detection algorithm was proposed based on contour restoration. Finally, considering the ambiguity problem in pose solving, a disambiguation method was proposed based on angle constraint and grayscale distribution,which leveraged the characteristics of endoscopic imaging:only circular features, rich grayscale information, the part being moved by the manipulator during pose detection, and the center of the inner hole being close to the optical axis. Experimental results verify the effectiveness of the proposed pose detection algorithm herein. The position detection accuracy in the X and Y axis directions is 0.062 mm, and the attitude detection accuracy around the X and Y axes is within 1.4°, which meets the requirements of enterprises.

Key words: pose detection, endoscopic imaging, ellipse detection, ambiguity removal, edge extraction, image segmentation

中图分类号: