The organization and identification of train wagons still heavily relies on humans reading their spray painted UIC numbers. A solution that can read these numbers automatically can be used in a variety of situations, such as damage identification and tracking. Although systems have been created for this task in the private sector, no public research or data is available. The first contribution of this thesis is a new UIC number dataset that was released to the public. The second contribution is a new system that is capable of locating and reading UIC numbers from low quality images. For this purpose, a modular framework, supporting an Extremal Regions and a FASText character region detector was implemented. Furthermore, the FASText method was extended for the use of color images. For grouping regions into lines, a new O(n 2 * log(n)) algorithm was designed for the UIC domain. The new method offers a significant performance increase over the established O(n 8) algorithm designed by Neumann & Matas. For decoding the lines into text, TesseractOCR and a Support Vector Machine are provided. Finally, an independent component was designed to group lines into UIC numbers, allowing a user to use the remaining pipeline for a different text detection system.