Scholarlink

<p>Accurate binocular camera calibration is essential for reliable three-dimensional reconstruction in scientific imaging, industrial inspection, optical measurement and other applications that depend on precise geometric information. However, feature-point localization remains vulnerable to noise, blur and distortion, which limits calibration accuracy, weakens robustness and reduces performance in uncontrolled real-world imaging conditions across practical systems. Here we present a geometric-fusion calibration framework that combines saddle-point preservation, unbiased cell centroids, and binocular epipolar constraints to improve the stability and reliability of global feature points in structured-light reconstruction. Experiments show that the method improves corner localization under image degradation, reduces reprojection error by 17%, and produces more accurate point clouds in real-world reconstruction tasks. These results support more reliable precision three-dimensional imaging for medical endoscopy, immersive display systems, embodied perception platforms and other compact active sensing technologies that require accurate geometric reconstruction in challenging environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Geometry-aware super-resolution fusion calibration for binocular structured light 3D reconstruction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月21日

查看全文
  • Hongyan Cao,
  • Dayong Qiao,
  • Mengya Han,
  • Wangke Yu,
  • Benquan Wang,
  • Yijie Shen

摘要

Accurate binocular camera calibration is essential for reliable three-dimensional reconstruction in scientific imaging, industrial inspection, optical measurement and other applications that depend on precise geometric information. However, feature-point localization remains vulnerable to noise, blur and distortion, which limits calibration accuracy, weakens robustness and reduces performance in uncontrolled real-world imaging conditions across practical systems. Here we present a geometric-fusion calibration framework that combines saddle-point preservation, unbiased cell centroids, and binocular epipolar constraints to improve the stability and reliability of global feature points in structured-light reconstruction. Experiments show that the method improves corner localization under image degradation, reduces reprojection error by 17%, and produces more accurate point clouds in real-world reconstruction tasks. These results support more reliable precision three-dimensional imaging for medical endoscopy, immersive display systems, embodied perception platforms and other compact active sensing technologies that require accurate geometric reconstruction in challenging environments.