Automatic detection of plan symbols in railway equipment engineering using a machine learning approach

Deian Stoitchkov1, Peer Breier1, Martin Slepicka1, Cengiz Genc2, Felix Harmsen2, Tobias Köhler2, Simon Vilgertshofer1, André Borrmann1
1 Technical University of Munich, Germany
2 Signon Deutschland GmbH, Berlin, Germany


DOI: 10.35490/EC3.2019.167
Abstract: Exact data in the form of technical drawings and plans of built assets are a significant requirement for the successful operation and reconstruction of such assets. When the consistency between this data and the real world situation cannot be assured, the data is not reliable and needs to be updated by comparing plans and reality. Depending on the size and number of assets this may involve an enormous amount of manual effort. In the scope of this research , an approach for supporting and automating such a process by utilizing concepts developed in the field of machine learning was developed. This paper focuses on the interpretation of technical drawings in terms of detecting and classifying plan symbols as this is a time intensive and error prone process when done manually. It is described how the capabilities of Convolutional Neural Networks are employed in analyzing images to automatically detect important plan symbols in the ?eld of Train Traffic Control and Supervision Systems and how those networks are trained without the need for a time consuming-manual labeling process.
Keywords: Machine Learning, CNN, Infrastructure, Railway Engineering
Pages: 92 - 99
Paper:
http://ec-3.org/conf2019/contribution_167_final/

Presentation Video: