Scholarlink

<p>To accurately evaluate fabric appearance characteristics—such as smoothness, crease retention, and seam flatness—often affected by color and pattern interference, this study proposes an optimized algorithm based on binocular stereo vision. The goal is to capture the fine-grained surface topography of fabrics for precise analysis of wrinkle features. A binocular depth camera is used for data acquisition, and an enhanced RAFT-D algorithm is adopted for disparity estimation. By integrating pixel grayscale, gradient features, and local smoothness constraints, the algorithm establishes a robust similarity metric, enabling accurate matching of geometrically calibrated stereo image pairs and generating high-density, continuous disparity maps. According to depth reconstruction principles in computer vision, the disparity data are further converted into 3D point clouds representing the fabric surface. To validate the effectiveness of the proposed method, reconstruction results from a high-precision handheld 3D scanner are used as a reference. Both quantitative and qualitative evaluations show that the algorithm achieves an overall matching success rate of 90.8% across all fabric samples, demonstrating its superior accuracy and reliability in practical 3D fabric reconstruction applications.</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

RAFT-D: A Reconstruction Method Incorporating Depth Estimation for Enhancing Fabric Performance Evaluation

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • Sumin Ge,
  • Jia Li,
  • Zhilei Yuan,
  • Congqing Wang,
  • Jianbo Gu,
  • Pinghua xu

摘要

To accurately evaluate fabric appearance characteristics—such as smoothness, crease retention, and seam flatness—often affected by color and pattern interference, this study proposes an optimized algorithm based on binocular stereo vision. The goal is to capture the fine-grained surface topography of fabrics for precise analysis of wrinkle features. A binocular depth camera is used for data acquisition, and an enhanced RAFT-D algorithm is adopted for disparity estimation. By integrating pixel grayscale, gradient features, and local smoothness constraints, the algorithm establishes a robust similarity metric, enabling accurate matching of geometrically calibrated stereo image pairs and generating high-density, continuous disparity maps. According to depth reconstruction principles in computer vision, the disparity data are further converted into 3D point clouds representing the fabric surface. To validate the effectiveness of the proposed method, reconstruction results from a high-precision handheld 3D scanner are used as a reference. Both quantitative and qualitative evaluations show that the algorithm achieves an overall matching success rate of 90.8% across all fabric samples, demonstrating its superior accuracy and reliability in practical 3D fabric reconstruction applications.