Skip to content

228 Image Analysis

Published: 23 August 2019

By using image analysis, the quality of data in the railway databases, which are important for maintenance, can be improved. The project uses existing data from inspection systems and plant registries to investigate and demonstrate what parts of effective support for quality improvements can look like.

In this project, image analysis was used to investigate whether the maintenance of reindeer fencing can be facilitated by automating inspections using images / films. The results show that image analysis software can identify inclined poles in clean fencing on the measuring cart films.
The difference in the slope of a given bar between different times (films) can also be identified. Winter pictures give the best results. Manipulating the original images to get a variety of variations and then using deep learning is what worked best and about 200 original images have been required to get acceptable results. Striving for the posts as well as poles inside some vegetation is not possible now. Pictures with the camera used on the measuring trolley before 2018 are not good enough. However, the network is very difficult to see even in newer images. It is probably possible to improve the images with optical corrections.
Additional strategies to improve results are to combine images and utilise asset data. In order for the methodology to support maintenance to become condition based to a greater extent than today, it needs to be tested in, for example, a maintenance contract.

Stakeholders: Trafikverket, Sweco och Strukton

Project leader: Birre Nyström, Sweco.