Innovations Deserving Exploratory Analysis--The Transit IDEA Program. IDEA 107. Rail Track Geometry Estimation via Fusion of Inertia and Vision Data with Track Deflection Compensation

Rail transit agencies need frequent, accurate measurement of track geometry to keep operations safe, yet conventional inspection relies on either slow manual surveys or costly dedicated geometry vehicles, so most networks are inspected infrequently. This project developed the Rail Integrity and Geometry Sensor (RIGS), a compact, low-cost sensing module that estimates the four governing track geometry parameters of gauge, curvature, superelevation, and vertical misalignment from a moving platform. A single unit integrates a camera, two inertial measurement units (IMUs), a two-dimensional light detection and ranging (LiDAR) scanner, and a global positioning system (GPS) receiver on a single-board computer, and it can be mounted on an inspection cart, a hi-rail truck, or a regularly scheduled service vehicle. A vision pipeline based on a U-Net segmentation network and homography rectification produces continuous gauge and curvature estimates, while an inertial pipeline reconstructs platform displacement. The two streams are aligned in time by cross-correlation and fused with a particle filter, and a deflection-compensation formulation recovers the rail vertical profile under the load of the moving platform. A full-scale 80-foot research testbed incorporating concrete and timber tie sections was constructed and surveyed with a survey-grade LiDAR system to provide reference geometry. All four parameters estimated by RIGS showed good agreement with the reference, confirming that low-cost sensing combined with modern computer vision and probabilistic fusion can approach the accuracy of dedicated geometry cars at a small fraction of their cost. The principal contributions are an end-to-end vision and inertial fusion pipeline, a deflection-compensation formulation that recovers the rail vertical profile by combining relative and absolute motion measurements, and a full-scale instrumented testbed with survey-grade reference geometry for quantitative validation. Remaining development work includes validation at operating speed over long distances, ruggedization for the environmental conditions encountered in rail operations, hardening of the vision models to distinguish running rails from guard rails and switch rails, and higher-precision positioning for milepost-accurate defect reporting. The path forward centers on a dedicated field campaign in which the module will be mounted on a hi-rail inspection truck and on partner-agency vehicles, with New Jersey Transit and the Chicago Transit Authority positioned as early agency partners.

• Record URL:
  • http://apps.trb.org/cmsfeed/TRBNetProjectDisplay.asp?ProjectID=5915

Language

• English

Project

• Status: Completed
• Funding: $75,000.00
• Contract Numbers:

  Project J-04, IDEA 107

• Sponsor Organizations:

  Federal Transit Administration

  1200 New Jersey Avenue, SE
  Washington, DC  United States  20590
• Managing Organizations:

  Transportation Research Board

  500 Fifth Street, NW
  Washington, DC  United States  20001

  Transit Cooperative Research Program (TCRP)

  Transportation Research Board
  500 Fifth Street, NW
  Washington, DC  United States  20001
• Project Managers:

  Jawed, Inam

• Performing Organizations:

  Texas A&M Transportation Institute

  ,    
• Principal Investigators:

  Najafi, Amirali

• Start Date: 20250102
• Expected Completion Date: 0
• Actual Completion Date: 20260630

Subject/Index Terms

• TRT Terms: Artificial intelligence; Computer vision; Data fusion; Deflection; Geometry; Image analysis; Railroad tracks
• Subject Areas: Data and Information Technology; Maintenance and Preservation; Railroads;

Filing Info

• Accession Number: 01960243
• Record Type: Research project
• Source Agency: Transportation Research Board
• Contract Numbers: Project J-04, IDEA 107
• Files: TRB, RIP
• Created Date: Jul 8 2025 5:04PM