Railcar Identification AI

Bring real-time intelligence to every railcar on the property with Roboflow vision AI for railcar identification.

Read the car, not just the tag:

• Read reporting marks and car numbers from oblique, moving, weathered, and partially rusted views on yard tower, crossover, gate, and drone cameras with a model trained on your own footage, including snow, rain, night lighting, graffiti, and stencils faded to the primer.
• Classify each unit as tank, covered hopper, open hopper, boxcar, gondola, flatcar, autorack, refrigerated, intermodal well car, container, or chassis, and read the logo, hazmat placard, and load-limit stencil on the same pass.
• Fill the gaps where AEI tags are missing, damaged, or unread, and cross-check every read against UMLER so a bad character never becomes a bad record.

Place every car on the yard map:

• Map each identified car to a track and position on a 2D yard view using camera calibration and track geometry, so the yard manager sees where a car sits and not just that it arrived.
• Track dwell per car, per track, and per customer, and flag cars that exceed a dwell threshold before demurrage or per-diem charges accrue.
• Search the archive by reporting mark, car type, or logo to find the rare unit or the lessor's fleet across months of footage from hundreds of cameras.

Scale across yards and feed the systems you already run:

• Deploy one model across 50 streams today and 500 tomorrow with a deployment manager for edge devices at yards and terminals where connectivity drops.
• Write car ID, type, position, and timestamp to your yard management system, TMS, or gate system over REST or MQTT, and raise gate arms through the existing PLC when the ID matches the appointment.
• Give operations the record: every read, every confidence score, and every map placement, with the frame, for AAR interchange disputes and FRA 49 CFR documentation.

Bring intelligence to every railcar today.

‍

Frequently asked questions

What is railcar identification with Vision AI?

Railcar identification with Vision AI uses existing yard, gate, and drone cameras and a trained computer vision model (opens in a new tab) to read the reporting mark and car number on each railcar, classify the unit type, read logos and placards, and place the car on a 2D yard map for location and dwell tracking. The model is trained on your own footage so it handles oblique angles, motion blur, weathered paint, and night lighting, and it complements AEI tag readers where tags are missing or unreadable. It runs alongside your yard management system and railcar inspection workflows, with a record of every read that supports AAR interchange rules, reporting-mark conventions, and FRA 49 CFR record-keeping.

Can Vision AI read a car number from an oblique, moving, rusted railcar?

Reading a reporting mark off a car rolling past a tower camera at a 40 degree angle, half covered in graffiti, is one of the highest-stakes identification tasks in rail. A detection model finds the mark region on each car as it passes, an OCR stage reads the characters across many frames, and consensus logic votes across those frames so a single blurred or occluded read does not win. The result is validated against reporting-mark format and a UMLER lookup, and a low-confidence read is flagged for a person to confirm rather than written as fact. The same pipeline is used for shipping container OCR at intermodal gates.

Does railcar identification support AAR interchange rules and FRA 49 CFR?

Railcar identification supports the AAR Interchange Rules (the Association of American Railroads rules that govern how cars move between carriers and how reporting marks and car numbers are assigned and displayed), UMLER (the Umler equipment registry that holds the master record for every car in interchange), and FRA 49 CFR (the Federal Railroad Administration regulations covering car records, inspection, and reporting) by giving operations a timestamped, frame-backed record of each identification and map placement. At intermodal gates, the same reads support TSA and CBP appointment and manifest checks. Roboflow serves as the identification engine, and your operations, fleet, and IT teams own the acceptance thresholds, the UMLER reconciliation rules, and what happens when a read does not match.

Can it integrate with our yard management system, TMS, and AEI reader data?

Yes. Roboflow sends car ID, unit type, position, and dwell to yard management systems and TMS platforms over REST (opens in a new tab) or MQTT, merges reads with AEI reader data so a camera read fills in where a tag did not fire, and runs UMLER lookups to validate each mark. At gates, a matched ID can raise the arm through the existing PLC or SCADA layer on Allen-Bradley or Siemens controllers, and the same model can be extended with logo detection for lessor fleet searches across the archive.

‍