Efficient Railway Analysis Using Video

Camera-based systems for rail maintenance have had a long history. However, traditional rail inspection methods require trains that are run during down time, have sensitive sensing/imaging equipment with high costs, or may require low speeds for analysis. There is a tremendous opportunity to develop more low-cost analysis and maintenance systems which leverage recent advances increased computational power and access to data through deep learning for computer vision. This research aims to provide a low-cost camera-based railway analysis system using a single forward-facing locomotive camera. Rather than providing detailed rail, tie and fastening wear or fatigue measurements, the system is targeted general railroad track condition and health at higher-level through identification of localized anomalies (vegetation, poor drainage, and surface ballast fouling) to target for more in-depth examination. Existing railway semantic segmentation datasets will be used to pre-train detection algorithms while a new rail anomaly dataset will be collected for fine-tuning the network model. Evaluation of popular deep learning techniques on a variety of low-cost hardware options will characterize the cost-performance trade-off.

    Project

    • Status: Active
    • Funding: $58656
    • Sponsor Organizations:

      Office of the Assistant Secretary for Research and Technology

      University Transportation Centers Program
      Department of Transportation
      Washington, DC  United States  20590
    • Project Managers:

      Teng, Hualiang

    • Start Date: 20210801
    • Expected Completion Date: 20220801
    • Actual Completion Date: 0
    • USDOT Program: University Transportation Centers Program

    Subject/Index Terms

    Filing Info

    • Accession Number: 01787402
    • Record Type: Research project
    • Source Agency: University Transportation Center on Improving Rail Transportation Infrastructure Sustainability and Durability
    • Files: UTC, RIP
    • Created Date: Nov 5 2021 8:15PM