Railway Fastener Defects Detection under Various Illumination Conditions using Fuzzy C-Means Part Model

Abstract

To ensure the safe operation of railways, computer vision and pattern recognition technology have been gradually applied to the routine inspection of railway track infrastructure. Rails are fixed to sleepers by railway fasteners, which are important components in railway track systems, and completely missing and partially worn railway fasteners may cause major accidents and train derailments. Because the acquisition of track images is carried out in the real world at any time of the day, the acquired track images have large illumination changes, and the fasteners in the images have slight deformations. To solve these problems, a fuzzy c-means part model (FCMPM) is proposed in this paper. The fastener part model is divided according to the fastener shape and solved by the fuzzy c-means clustering algorithm using the simplified and improved histogram of oriented gradients as the low-level feature. The part score is calculated based on the part’s deformation and then is seamlessly incorporated with cascade detection to determine whether the fastener has defects or not. The experimental results from the fastener defect detection show that the proposed FCMPM algorithm achieves good performance when analysing the collected fastener images and can meet the requirements of fastener defect detection for actual railway lines.