Development of Railroad Trespassing Database Using Artificial Intelligence
archive: archived pipeline: cataloged verified
Get this paper ↗ (full text — opens at the source; we link to it, we don't host it)
Summary
This report details the development of a proof-of-concept Railroad Trespassing Database using Artificial Intelligence (AI) to address the critical safety issue of trespassing on railroad rights-of-way (ROW) and highway-rail grade crossings. Trespassing accounts for approximately 95% of all railroad industry fatalities in the United States. While railroads have increasingly deployed video surveillance cameras, the manual processing of this "big data" is labor-intensive and inefficient. The research, sponsored by the Federal Railroad Administration (FRA) and conducted by Rutgers University, aimed to fill this gap by creating an automated system to detect, record, and analyze trespass events, thereby providing data-driven insights for engineering, enforcement, and education strategies. The researchers developed an AI algorithm utilizing You Only Look Once (YOLO) for object detection and DeepSORT for object tracking. This system was deployed to analyze over 50,000 hours of video data, comprising 27,000 hours of live streams and 1,176 hours of recorded footage from 11 locations across six states (New Jersey, Virginia, North Carolina, Connecticut, Louisiana, and Illinois). The AI automatically extracted metadata for each trespass event, including date, time, trespasser type (e.g., pedestrian, car, truck), weather conditions, and trajectory. To ensure data integrity, the research team manually validated all AI-detected events. The study focused heavily on two year-long case studies: a grade crossing in New Jersey and a ROW location in North Carolina, alongside shorter analyses of other sites. The AI algorithm detected over 29,000 trespass events across all studied locations. The New Jersey grade crossing case study identified 21,202 events, while the North Carolina ROW study identified 476 events. The analysis revealed detailed temporal and spatial patterns of trespassing behavior. For instance, the New Jersey data highlighted specific times and days with high trespass rates and identified "near-miss" events where trespassers were within 10 seconds of a train’s arrival. The North Carolina analysis identified specific origin and destination zones for trespassers, including instances of trespassing for social events like graduation photos. The study also analyzed traffic exposure and signal activations to contextualize the risk levels associated with different types of trespassers. The significance of this work lies in the creation of a comprehensive, publicly available dataset and a web-based dashboard that allows stakeholders to query and visualize trespass trends. By automating the extraction of detailed metadata from video feeds, the project demonstrates a scalable method for understanding non-fatal trespass behaviors, which are precursors to accidents. The findings support the development of targeted mitigation strategies, such as specific enforcement blitzes, improved channelization, or landscaping, justified by empirical data rather than anecdotal evidence. This approach provides a foundation for more effective, data-informed safety interventions to reduce trespassing fatalities and incidents.
Key finding
An AI-based system successfully detected over 29,000 trespass events from more than 28,000 hours of video data across 11 locations, enabling detailed temporal and spatial analysis of trespassing behaviors to inform safety mitigation strategies.
Methodology
field_study
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via bulk_ingest_rosap on 2026-05-23 (6 acquisition events logged).
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | rosap | — | — | 2 | 2026-05-23 |
| archive | success | — | — | — | 1 | 2026-05-23 |
| extract | success | cached | — | — | 2 | 2026-06-10 |
| clean | success | — | — | — | 1 | 2026-06-01 |
| chunk | success | — | — | — | 1 | 2026-06-01 |
| embed | success | — | — | — | 1 | 2026-06-02 |
| enrich | success | — | — | — | 1 | 2026-05-23 |
| promote | success | — | — | — | 1 | 2026-05-23 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 3 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 24 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.