A Method of Classifying Railway Sleepers and Surface Defects in Real Environment

Rail transport is an efficient and safe way to move large quantities of goods and people over long distances but it still suffers from maintenance issues, mainly due to assets of great extent, quantity, weight, and geographic dispersion. Because of this, some initiatives in automatic inspection of railway assets have been developed in recent years/in the last decade. In particular, the automatic inspection of railway sleepers still needs improvement and consolidation. This work presents a method for sleepers inventorying, identification of the type and defects based on image processing, heuristics and feature fusion in an unsupervised way. The Haar transform and integral images are used, as well as other image processing techniques such as edge detection, and entropy computation along with aspects of railroad topology. The algorithm was developed using real images of daily railway, previously unclassified, and that were subject to various noises and variations of a real railway operation. The method was validated through experiments with an image set comprising 32,917 sleepers in 10,116 images. The results are promising in which 97% accuracy is reached, for the identification of the type of sleepers, and 93% accuracy for the identification of visible defects in sleepers.