Automated detection and quantification of structural component dimensions using segment anything model (SAM)-based segmentation

This paper presents a method for automatic detection and quantification of full cross-sectional dimensions of structural components using oblique photography and the SAM-dimension (Segment Anything Model-dimension) model. Unlike traditional methods that measure only a single cross-section, this approach enables full cross-sectional dimension detection across the entire component, enhancing efficiency and coverage. The method utilizes the camera's crosshair position within the component area to adjust the model's prompt module strategy, allowing operation without human intervention. A binary mask is generated by fusing the model's output with the original image. Additionally, the method incorporates local unit methods, connected domain analysis, and mask feature extraction for dimension quantification. Experimental results show the model's adaptability to varying distances and lighting, accurately segmenting square, circular, and variable cross-section components. The relative errors for column components' cross-sectional width and height are within 1.22 %, and for beam-type components, within 3.02 %, demonstrating robust and generalized performance.