Development of a Low Power, Low Cost Rural Railway Intersection Smart Detection and Warning System

This project explores a different approach to provide preemptive warning for train detection at grade-crossings to increase safety and reduce motor vehicle congestion. The development of a novel, low cost, low power, and off rail right-of-way (ROW) detection and warning system will be presented. A background of track circuits, which is the rail industries standard for train detection, will also be provided to highlight the benefits and challenges of the rail industry installing a system at every grade-crossings that lack any type of active warning. The benefits of using thermal imaging instead of traditional video for computer vision will also be discussed. Then, the development of a detection algorithm using convolutional neural networks and deep learning methods on a custom dataset of images created by two different thermal sensors will be overviewed. The two sensors tested are the MLX90640 and FLIR Lepton 3.0. Both react to heat radiation emitted from objects but in very different ways from a technology perspective. This difference in technology creates a large gap in cost and power between the two and was therefore worth testing to analyze the trade off in accuracy achieved in creating an outdoor detection model. Finally, the wireless communication between the detection and warning devices is discussed as well as the network communication for detected data to be logged to an online server and updated on a mobile application. Testing was done at a location approximately one mile east of the Kirby Union Pacific Railyard in San Antonio, TX from a distance that was off the property of ROW.


  • English

Media Info

  • Media Type: Digital/other
  • Edition: Final Research Report
  • Features: Figures; Photos; References; Tables;
  • Pagination: 66p

Subject/Index Terms

Filing Info

  • Accession Number: 01775407
  • Record Type: Publication
  • Report/Paper Numbers: 19SAUTSA03
  • Contract Numbers: 69A3551747106
  • Files: TRIS
  • Created Date: Jun 25 2021 6:37PM