Scholarlink

<p>Bolt connected joints are widely used in mechanical assembly structures. Identifying the looseness condition of bolts is essential for maintaining the integrity of the whole structure. This paper presents a two-stage perspective distortion correction method based on computer vision, aimed at accurately detecting bolt loosening from images with significant angular tilt. In the first stage, perspective transformation is applied to correct the overall perspective distortion of the bolt images. In the second stage, the Faster R-CNN model is utilized to locate the bolt positions, while the Hough transform is applied to extract the marked line features. These features are then processed using the convex hull algorithm and affine transformation to correct the local perspective distortion. The experimental results demonstrate that the proposed method achieves a bolt loosening recognition accuracy exceeding 90% when the image shooting tilt angle is within 40°, by performing perspective correction at both the overall and local levels.</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

Computer Vision-Based Quantitative Detection of Bolt Loosening Using Two-Stage Perspective Distortion Correction Method

  • 期刊论文

  • 发表日期：

    2025年6月22日

查看全文
  • Ru Zhang,
  • Chaodong Guan,
  • Xiaodong Sui,
  • Nahai Ding,
  • Yang Ding,
  • Lianying Zhou

摘要

Bolt connected joints are widely used in mechanical assembly structures. Identifying the looseness condition of bolts is essential for maintaining the integrity of the whole structure. This paper presents a two-stage perspective distortion correction method based on computer vision, aimed at accurately detecting bolt loosening from images with significant angular tilt. In the first stage, perspective transformation is applied to correct the overall perspective distortion of the bolt images. In the second stage, the Faster R-CNN model is utilized to locate the bolt positions, while the Hough transform is applied to extract the marked line features. These features are then processed using the convex hull algorithm and affine transformation to correct the local perspective distortion. The experimental results demonstrate that the proposed method achieves a bolt loosening recognition accuracy exceeding 90% when the image shooting tilt angle is within 40°, by performing perspective correction at both the overall and local levels.