Enhanced Datasets and AI Models for Monitoring of Grade Crossings

The safety of grade crossings is a major concern of transportation agencies and researchers due to the number of accidents every year. The Federal Transit Administration (FTA) reported 592 grade crossing collisions in 2022, resulting in 19 fatalities and 133 injuries. Many innovative technologies have been proposed to automate the monitoring of crossings. The goal of this project is to investigate the use of Artificial Intelligence (AI) and Deep Learning (DL) to monitor grade crossings and detect various hazardous conditions such as vehicles, pedestrians, cyclists, animals, warning lights, and others. In prior work, the research team developed a Convolutional Neural Network (CNN) model and trained it using a dataset of 1,364 images that was collected and manually labeled by the authors, reaching a validation accuracy of 98.90% to detect vehicles at grade crossings. However, there are limitations to the model stemming from the need to improve the size and balance of the data. The work in this proposal aims to address limitations in the current model and to make new advances by (1) increasing the number of photos in the dataset using real video streams; (2) using captures from a train simulator videogame environment; (3) addressing the issue of imbalanced dataset for training and validation; and (4) hyper-optimizing the model for accuracy and real-time performance. This project relates directly to the strategic research goal of the University Transportation Center for Railway Safety (UTCRS) of reducing fatalities and injuries at highway-rail grade crossings (HRGCs); and relates to the railway operation systems research area of autonomous systems for grade crossing safety. The outcomes of this research will advance knowledge in automated monitoring of hazards at grade crossings, and result in a model that can be implemented in cameras for automated hazard monitoring at grade crossings.

• Record URL:
  • https://www.utrgv.edu/railwaysafety/_files/documents/research/operations/exhibit-d_utcrs_utrgv_grade_crossing_ai_models_ali_2024.pdf
• Supplemental Notes:
  • The University of California Riverside is a partner for this project.

Language

• English

Project

• Status: Active
• Funding: $76921
• Contract Numbers:

  69A3552348340

• Sponsor Organizations:

  Office of the Assistant Secretary for Research and Technology

  University Transportation Centers Program
  Department of Transportation
  Washington, DC  United States  20590

  University of Texas Rio Grande Valley

  1201 W. University Dr
  Edinburg, TX  United States  78539
• Managing Organizations:

  University of Texas Rio Grande Valley

  1201 W. University Dr
  Edinburg, TX  United States  78539
• Project Managers:

  Stearns, Amy

• Performing Organizations:

  University of Texas Rio Grande Valley

  1201 W. University Dr
  Edinburg, TX  United States  78539
• Principal Investigators:

  Ali, Gasser

  Tarawneh, Constantine

• Start Date: 20240601
• Expected Completion Date: 20250831
• Actual Completion Date: 0
• USDOT Program: University Transportation Centers Program

Subject/Index Terms

• TRT Terms: Artificial intelligence; Image analysis; Machine learning; Monitoring; Railroad grade crossings; Railroad safety; Railroad tracks
• Subject Areas: Data and Information Technology; Railroads; Safety and Human Factors;

Filing Info

• Accession Number: 01924850
• Record Type: Research project
• Source Agency: University Transportation Center for Railway Safety
• Contract Numbers: 69A3552348340
• Files: UTC, RIP
• Created Date: Jul 22 2024 8:04AM