Vision-based displacement monitoring for the integral lifting of a large-span spatial truss structure

Abstract Large-span spatial truss structures are widely used for their high performance, making it crucial to ensure proper installation quality. Currently, the lifting process of the large-span spatial truss structure is guided by visual observation, which is both imprecise and time-consuming. This paper presents a vision-based monitoring system designed to guide the lifting process of large-span spatial truss structures in real-time. A deep learning-based target detector initially tracks the trajectory of welded balls during the lifting process. To precisely extract the center point of each welded ball within the detected bounding box, we propose a novel ellipse recognition algorithm that uses morphological operations and the least squares method to identify the center coordinates at the pixel level. The binocular vision system then converts these pixel coordinates into world coordinates and calculates the displacement of the welded ball during the jacking process. A field test of the lifting of a large-span structure at a high-speed railway station verifies the effectiveness of the proposed. Test results indicate that the proposed method provides relatively accurate displacement estimates with a 2.5% relative error.