The project aimed to address the need for a comprehensive, nationwide digital record of infrastructure surrounding railway tracks, including noise barriers, track accesses, embankments, level crossings, overpasses, and underpasses. These elements are, amongst others, crucial for noise mapping, emergency management, and fulfillment of reporting obligations. The project focussed on the automated detection of noise barrier walls. One of the major challenge was to bring together the available digital data from diverse sources, which varied significantly in information content and timeliness. To overcome this, the project sought to employ machine learning techniques to automatically capture and assess the location and technical properties of infrastructure objects.

To tackle the challenge, a comprehensive solution leveraging public data, advanced ML techniques and models, as well as a seamless integration into existing software solutions, was developed. A nationwide dataset consisting of aerial flight survey data surrounding rail tracks was created as a robust data foundation for machine learning models and further use by the client. Based on this, an efficient processing workflow, using state-of-the-art computer vision techniques to detect and segment noise barrier walls from these aerial flight survey data was established.

The final step was the integration of the computer vision model into the required software interface via an open-source plugin. This integration ensures users can easily access additional functions, models, and updates, maintaining compatibility with new software versions. Our solution delivers consistent, accurate, and up-to-date information, enhancing planning and operational efficiency for infrastructure object analysis.

The newly developed processing tool and resulting dataset now serve multiple critical functions. The Eisenbahn-Bundesamt (EBA) evaluates the dataset for noise mapping for the current reporting period. Additionally, other infrastructure users or authorities can employ the tool for updates or calculations as needed.

The insights gained into the requirements, available methods, and datasets for the automated recording of trackside infrastructure enable the DZSF to strategically coordinate further developments in this area. This initiative not only provides the necessary impetus for advancements but also contributes to the overall goal of establishing a standardized nationwide approach to infrastructure recording. This standardization is key to improving efficiency, accuracy, and reliability in infrastructure management across Germany.

Official Project Website