Automatic Detection of Rail Surface Cracks with a Superpixel-Based Data-Driven Framework

Abstract

A bilevel superpixel-based framework for the vision inspection of rail conditions and automatically detecting rail surface cracks is proposed in this paper. The simple linear iterative clustering (SLIC) algorithm is applied to generate superpixels from raw rail images. Bag-of-words (BoW) features are extracted from each superpixel with DAISY descriptors and are used to develop the superpixel classifier for identifying cracks. Five classification algorithms, the support vector machines (SVM), neural networks (NN), random forests (RF), logistic regression (LR), and boosted tree (BT), are considered in the classifier development, and their performances are comparatively analyzed. The comparison shows that the RF classifier provides the best performance. The effectiveness of the proposed crack-detection framework is validated by rail images collected from rail systems in China. The computational results demonstrate that the proposed framework can automatically detect rail surface cracks and obtain their boundaries on images captured from different angles and distances.