One of the most important safety-related tasks in the rail industry is an early detection of defective rolling stock components. Railway wheels and wheel bearings are the two components prone to damages due to their interactions with brakes and railway track, which makes them a high priority when the rail industry investigates improvements in the current detection processes. One of the specific wheel defects is a flat wheel, which is often caused by a sliding wheel during a heavy braking application. The main contribution of this paper is the development of a computer vision method for automatically detecting the sliding wheels from images taken by wayside thermal cameras. As a byproduct, the process will also include a method for detecting hot bearings from the same images. We first discuss our automatic detection and segmentation method, which identifies the wheel and bearing portion of the image. Then, we develop a method, using histogram of oriented gradients to extract the features of these regions. These feature descriptors are later employed by support vector machine to build a fast classifier with a good detection rate, which can detect abnormalities in the wheel. At the end, we train our algorithm using simulated images of sliding wheels and test it on several thermal images collected in a revenue service by the Union Pacific Railroad in North America.
Read full abstract