An Approach to Engineering Drawing Organization: Title Block Detection and Processing

During the last 20 years the world has been meeting digitalization challenges in many areas of life and business. Companies convert their business processes to digital form as much as possible, what allows reducing costs and increasing profits. The paper considers a case of the company aimed at manufacturing and servicing spare parts for mining equipment. The company has collected a large amount of engineering drawings in a paper form that need to be organized in a digital form to simplify and support quick access to these documents. We propose an approach to engineering drawing organization based on modern artificial intelligence technologies for detection of main engineering drawing elements as well as extracting metadata from them. Based on the metadata we propose organizing engineering drawings into a structured digital collection with a possibility of quick access to them. All metadata in drawings are usually located in the title block. We show how we detect the title block as well as process it and recognize needed text-based information. The open-source Google Tesseract framework is used for text recognition in English, Finnish, Spanish, and Portuguese languages. For Cyrillic text recognition our own neural network-based model has been developed and trained on Russian GOST engineering drawings. We developed a drawing organization service and evaluated it based on the company engineering drawings database. The evaluation shows a good accuracy for documents that have a good quality (about 74% for title block detection as well as metadata identification).