AI System Assesses Rail Crossing Safety Using Multi-modal Data

Paimon Goulart, Chansong Lim, N\'icolas Roque dos Santos, Yue Dong, Sheldon Peterson, Jia Chen, Evangelos E. Papalexakis· July 3, 2026 View original

Summary

This research proposes an AI system that leverages multi-modal data, including visual cues from images and structured accident reports, to assess railway crossing safety. The proof-of-concept pipeline identifies high-risk and low-risk crossings with high accuracy, aligning with expert and Federal Railroad Administration safety scores.

Researchers have developed a novel AI system designed to evaluate the safety of railway crossings by integrating diverse data sources. The system processes visual information from images of crossings alongside structured historical accident data, such as official reports. The goal is to create a robust safety assessment and scoring mechanism that aligns with both expert human judgment and the Federal Railroad Administration's (FRA) safety scoring standards. The proposed proof-of-concept pipeline demonstrates significant capability in identifying high-risk and low-risk crossings, achieving a macro F1 score of 0.757. Furthermore, the system estimates FRA-based safety scores with a low Root Mean Square Error (RMSE) of 0.078 and a correlation of 0.492, utilizing a fine-tuned compact Vision-Language Model (VLM) pipeline. Qualitative results also show strong alignment with domain expert assessments, indicating its potential for practical application.

Why it matters

Improving railway crossing safety is critical, and this AI system offers a data-driven approach to proactively identify and assess risks, potentially preventing accidents and saving lives.

How to implement this in your domain

  1. 1Explore integrating visual inspection data with historical safety records for infrastructure risk assessment.
  2. 2Pilot multi-modal AI models for safety analysis in specific high-risk operational areas.
  3. 3Collaborate with domain experts to validate AI-generated safety scores and refine model parameters.
  4. 4Develop data collection strategies for visual and structured data at critical infrastructure points.
  5. 5Investigate the potential for real-time safety monitoring using similar multi-modal AI approaches.

Who benefits

TransportationInfrastructurePublic SafetyInsurance

Key takeaways

  • Multi-modal AI can effectively assess railway crossing safety by combining visual and structured data.
  • The proposed system aligns well with expert opinion and official safety scoring.
  • Integrating accident history significantly improves safety prediction capabilities.
  • This approach offers a proactive method for identifying high-risk crossings.

Original post by Paimon Goulart, Chansong Lim, N\'icolas Roque dos Santos, Yue Dong, Sheldon Peterson, Jia Chen, Evangelos E. Papalexakis

"arXiv:2607.01365v1 Announce Type: new Abstract: Given one or more images of a railway crossing, can we leverage visual cues that allow us to robustly estimate how safe it is? Can we improve our ability to do so by introducing structured data (such as official accident reports) ab…"

View on X

Originally posted by Paimon Goulart, Chansong Lim, N\'icolas Roque dos Santos, Yue Dong, Sheldon Peterson, Jia Chen, Evangelos E. Papalexakis on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses