AUTOMATIC TITLE BLOCK DETECTION AND INFORMATION EXTRACTION OF BUILDING DRAWINGS
Duan Li · HAL (Le Centre pour la Communication Scientifique Directe) · 2025
Information from building drawings is important in the architecture, engineering, and construction (AEC) sector for building construction, maintenance, compliance checks, and error checks. However, manual information extraction from building drawings is time-consuming and significantly increases project costs, especially when handling large volumes of drawings. This paper proposes automatic detection and information extraction from one of the key areas of building drawings: title blocks. By integrating Faster RCNN and GPT-4o, the proposed pipeline can detect title blocks from drawings, extract information from the detected title blocks, and store the extracted data in a database. A user interface has been established so users can retrieve building drawings straightforwardly and efficiently. Our model demonstrated strong performance in both vector and historical (scanned hand-written) drawings with an accuracy of 88.2% in title block detection, which has been a challenge in research due to the fact that historical drawings are blurred and noisy.