downward-facing chevron icon

Object Detection & Image Classification

New image processing technologies allow us to take large volumes of track imagery and create highly-detailed mappings of the rail network. This, in turn, allows us to ask smarter questions: Do some rail types perform better than others? Where are weather patterns driving rail break risk?

Answering these kinds of questions takes detailed rail and condition information from across the network. Automated rail scanning and labeling processes confirm existing records and fill in gaps to give a high-resolution picture of the network.

RailLinks® AI can process track features including joint bars, thermite welds, fasteners, ties, and other components. In addition to standard line-scan and downward-facing detection systems, we can also use forward-facing video to detect larger assets like cross bucks and signals.
Laptop displaying forward-facing track imagery, with various components outlined and labeled

Humans in the Loop

Humans are the bookends of the RailLinks AI workflow. Labelers intervene at critical stages of the process, including the creation of training data and the validation of unclear imagery. To manage this process, RailLinks AI has a range of tools for training labelers, assigning tasks, and measuring performance.
Graphic collage combining a photo of a man working at a computer in an office with several graphs showing performance metrics

Workflow Setup

Once a model is set up with a good baseline of training data, the model can be plugged into an automated workflow which can run on a schedule or whenever new data is uploaded. Unclear data can still be flagged and reviewed by human labelers.
Graphic showing many icons chained together into a single workflow. One of the icons is highlighted with a tooltip showing "Words Model - Completed", and there is faint word analysis imagery in the background.
A core strength of RailLinks AI is its ability to integrate with the rest of our data architecture. The results of a workflow can be automatically delivered to other parts of the RailLinks platform and viewed in combination with other data types.

Drone Inspections

VisioStack has joined with railway partners and the SBIR research program to develop innovative, drone-based solutions for inspecting track components. This opens new possibilities for conducting inspections in difficult or dangerous conditions, and for streamlining general inspections.
Autopilot & Track Scanning
A smart drone inspection requires the ability to follow a track and the ability to quickly process basic track geometry. VisioStack has completed projects on both of these fronts, creating software that can guide a drone along the length of a track while mapping ties, rails, and centerlines.
Tablet displaying a screenshot of drone autopilot software. The background of the screen shows a downward-facing video view from the drone, and various measurements and buttons are overlaid, along with a code log and a small picture-in-picture forward-facing view.
Crossing Measurement
There are thousands of safety-related incidents each year across the globe at railway crossings. When maintenance responsibilities are shared across multiple stakeholders—railways, municipalities, and regulatory bodies—crossing management requires focused accountability.
Photo of a drone hovering near a railroad crossbuck and gate, with a forest in the background.
Many incidents at crossings involve vehicles with low clearances, or areas where lines of sight and visibility are limited. We aim to reduce incidents through innovative assessment of risk, by measuring and reporting on these geometries through drone scanning.
Graphic showing the clearance of a crossing being calculated
VisioStack's AXIS (Aerial Crossing Inspection System) inspection process uses off-the-shelf drone technology to improve risk assessment at crossings. Combined with a railway's existing crossing asset information, AXIS delivers automation including flight pre-planning, execution & collection, data transfer, and generation of reports.