Automatic detection and 3D pose reconstruction of loose bolts with rotation angle quantification using a calibration-free monocular camera

Loose or lost bolts in steel structures can compromise the safety and mechanical performance of connections. This paper presents a calibration-free monocular vision approach for detecting loose or lost bolts and quantifying the rotation angles of loose bolts. The approach integrates robust bolt detection, 3D pose reconstruction, and rotation angle estimation into a unified pipeline. The spatial pose of a bolt is recovered from a single 2D image using the homography matrix and the geometric features of the bolt. Laboratory experiments have been conducted to comprehensively evaluate the proposed approach. Results demonstrate high accuracy and robustness under varying distances, viewpoints, and imaging devices. This research enhances the capability of assessing bolt conditions and validates the proposed method through experiments on steel structures. Future work will extend the application of the approach to various bolt types and explore real-time deployment in complex field environments.