Treffer: Research on a bolt loosening visual detection method based on geometric features of bolt groups without marked points.

A new methodology that employs unmanned aerial vehicles (UAVs) for photographic inspections is proposed to identify bolt loosening on steel bridges by quantitatively assessing the rotational angle of each bolt or nut. This approach capitalises on the constancy of the geometric centreline of the nuts affixed to structures, irrespective of camera position or orientation variations. Nuts can be located and key points can be discerned from the captured images by integrating fast region-based convolutional neural network (Fast R-CNN) and real-time multi-person pose estimation (RTM-Pose) models and adopting a top-down detection strategy. A pivotal component of this methodology is using a straight line that traverses the centre of the nuts as a reference, mitigating systematic errors arising from discrepancies in shooting angles and ensuring the accuracy and reliability of the detection outcomes. This technique is validated by imaging bolts from multiple angles, confirming its efficacy and precision in practical applications. Additionally, the data model is enhanced and a predictive key-point correction method is introduced to improve the accuracy of key-point detection. Consequently, systematic errors induced by varying shooting angles can be minimised. Experimental findings reveal absolute errors ranging from 0.1°-1.2° due to different shooting angles, with relative errors between -2.43% and 3.97%. Most measurements exhibit relative errors within ±4%. This methodology establishes a solid foundation for applying UAVs in real-world scenarios to detect bolt loosening on steel bridges. [ABSTRACT FROM AUTHOR]