Technical Reports

Robust Anomaly Detection for Vision-Based Inspection of Railway Components

  • 22
  • Jun
  • 2015
AUTHOR: Federal Railroad Administration
KEYWORDS: visual track inspection; computer vision; concrete tie conditions; fastener inspection; tie inspection; crack detection; deep learning; distributed computing
ABSTRACT: The Computer Vision Laboratory at the University of Maryland has completed a two-year research project in which it developed a library of tools and algorithms that are used to inspect railway tracks with machine vision technology. This technology has been integrated into a distributed computing framework and a user-friendly review package with a client-server interface. This framework is built on top of high-quality open-source C++ software libraries and can run on many different platforms, including Windows, Linux, OS X, and iOS. The software has been thoroughly tested and ENSCO, Inc. has successfully used these tools during the FRA-sponsored project that evaluated the degradation rates of concrete ties on Amtrakā€™s Northeast Corridor. Our final deliverable contains algorithms for crack detection, fastener detection and classification, as well as semantic segmentation for material classification and anomaly detection.