中国机械工程

• 机械基础工程 • 上一篇    下一篇

基于单目图像序列的铸件三维重建方法

张洪鑫;王明珠;仇浩然;姜金刚   

  1. 哈尔滨理工大学机械动力工程学院,哈尔滨,150080
  • 出版日期:2018-12-06 发布日期:2018-12-06
  • 基金资助:
    国家自然科学基金资助项目(51675142)
    National Natural Science Foundation of China (No. 51675142)

3D Reconstruction Method of Castings Based on Monocular Image Sequences

ZHANG Hongxin;WANG Mingzhu;QIU Haoran;JIANG Jingang   

  1. School of Mechanical & Power Engineering,Harbin University of Science & Technology,Harbin,150080
  • Online:2018-12-06 Published:2018-12-06
  • Supported by:
    National Natural Science Foundation of China (No. 51675142)

摘要: 为了实现基于机器视觉的铸件毛刺自动检测与识别,提出了一种基于单目视觉的铸件三维重建方法。利用单目相机对目标进行无标定多角度旋转拍摄,获得一组单目图像序列;运用“两步匹配法”优化相邻图像特征点的匹配,剔除大量误匹配点,提高特征点匹配效率。对三维重建的点阵进行泊松表面重建,细化了铸件三维表面的细节。实验结果表明,该方法装置简单,无需对相机与目标的相对位置进行预先标定,能有效地恢复铸件三维表面的细节特征。该方法适用于机械加工环境下零件的三维表面重建。

关键词: 三维重建, 单目视觉, 特征点匹配, 泊松表面重建

Abstract: In order to realize automatic detection and identification of casting burrs based on machine vision, a 3D reconstruction method of castings was proposed based on monocular visions. The uncalibrated cameras were multi-angle rotated to grab a set of monocular image sequences of a single target. A “two-step matching” method was used to optimize the matching of feature points of adjacent images, as a result of eliminating a large number of false matching points and improving the matching efficiency of feature points. Poisson surface reconstructed was used to refine the details of 3D surface of the castings based on the 3D reconstructed point arrays. The experimental results show that the proposed method and device are simple, without the pre-calibration of the relative position between the camera and the target, which may effectively restore the detail features of the 3D surface of the castings. The method is suitable for 3D surface reconstruction of parts in machining environment.

Key words: 3D reconstruction, monocular vision, feature point matching, Poisson surface reconstruction

中图分类号: