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

The article delves into the critical issue of bolt loosening in mechanical structures, which can compromise the load-bearing capacity and safety of connected structures. Traditional methods of detecting bolt loosening, such as manual inspection and sensor detection, have limitations in efficiency and accuracy. The article introduces a computer vision-based approach that utilizes a two-stage perspective distortion correction method to address these challenges. The first stage involves using perspective transformation to correct the overall distortion of bolt images, while the second stage employs the Faster R-CNN model and Hough transform to correct local perspective distortions. This innovative method enables accurate detection of bolt loosening angles, even at large shooting angles, and offers a low-cost, efficient alternative to traditional methods. The article also presents an experimental study that validates the effectiveness of the proposed method under various shooting angles and lighting conditions, providing a comprehensive analysis of its performance and robustness.