Automated Extraction of Structural Elements and Section Views from Bridge Drawings Using Deep Learning

Abstract

Technical drawings are essential for managing and maintaining existing bridges, especially in creating accurate digital models and integrating them into Building Information Modeling (BIM) workflows. However, digitizing older bridges, which often lack BIM models is challenging due to the time-consuming and error-prone process of manual data extraction from 2D drawings. Therefore, this study focuses on detecting key structural components, such as anchors, tendons, and section views, Which are critical for structural analysis, maintenance, and retrofitting. Using state-of-the-art object detection models, YOLOv8 and YOLOv11, the proposed pipeline achieves a mean average precision (mAP) of 97.4% for anchors, 95.7% for tendons, and 63.9% for section views. Optical Character Recognition (OCR) tool Pytesseract further enhances the process by accurately extracting IDs for anchors and tendons with 90% precision. Combining object detection with precise text extraction simplifies the creation of high-quality digital bridge models and improves BIM integration. Specifically, the extracted data provides semantic information on key structural components. This information can facilitate the process of creating a digital model and integrating it into BIM workflows.