Incident ModelRail Security Intelligence

Track Trespass Detection.

Stop trains before tragedies occur. DHI differentiates between authorized maintenance crews and civilian trespassers, alarming dispatch the second an unauthorized body enters the rail envelope.

<150ms
Edge Processing
On-Prem
Inferencing
IR / Thermal
Low-Light Ready

Logic Validation

Track Trespass Detection requires precise computer vision engineering.

Verify the specific neural thresholds and VMS integration triggers used to automate this safety protocol.

Edge Processing

<150ms

Stop trains before tragedies occur. DHI differentiates between authorized maintenance crews and civilian trespassers, alarming dispatch the second an unauthorized body enters the rail envelope.

Inferencing

On-Prem

Authorized vs Unauthorized: The model reads high-visibility PPE to distinguish authorized hi-vis maintenance crews from civilian trespassers, so routine track work does not trigger alerts but an unprotected body in the rail envelope does.

Integration fit

Automated Dispatch Paging

Bypass the VMS entirely if needed. DHI can trigger physical sirens, flashers, or route IP pages directly to the train control center the moment track occupancy is compromised.

Failure mode coverage

Rail Security Intelligence

Motion Path Analysis: Directional human travel toward the rails is separated from ambient movement like blowing debris or animals, keeping nuisance alarms low.

Use-case fit

IR / Thermal

Low-Light Augmentation: Inference runs directly on IR and thermal camera feeds, so detection holds up in unlit tunnels and overnight yards.

Direct answer

What makes track trespass detection different from ordinary motion detection?

Track trespass detection has to understand the mapped rail envelope. DHI is built to confirm a human body inside the restricted area, distinguish likely authorized activity from unauthorized intrusion when the scene supports it, and route the event to dispatch or local warning systems quickly.

Rail-envelope fit

The first setup should map the trackbed, fence line, bridge approach, tunnel portal, or yard edge so the model knows the actual danger boundary.

Nuisance review

The pilot should separate wildlife, wind, debris, passing trains, and authorized maintenance from confirmed human occupancy.

Dispatch fit

The useful event includes camera, zone, time, direction, and urgency context so control can warn, slow, dispatch, or escalate.

Edge fit

Corridor cameras can have constrained backhaul, so local inference reduces dependence on a cloud path for the first alert.

Pilot assumptions to validate

  • The pilot zone has a clear mapped rail envelope and enough camera coverage to confirm human occupancy.
  • Authorized work crews, maintenance windows, and PPE rules are documented before testing.
  • Dispatch or field response teams agree how each alert type should be handled.

Rail corridor fit

Trespass detection has to work where bandwidth and visibility are both constrained.

Rail trespass is different from station platform safety because the camera may sit on a pole, in a tunnel mouth, or along a yard fence with limited backhaul. The page is tuned around corridor occupancy and dispatcher escalation.

Primary scene

Ideal pilot zones include known fence breaches, bridge approaches, tunnel portals, and yard edges where a person can enter the rail envelope without passing through a staffed station.

Operator action

The useful alert is not just motion. It is a confirmed human body inside a mapped track envelope, with location metadata precise enough for dispatch to slow movement, trigger a field response, or warn operators.

Pilot measure

A rail pilot should track nuisance alerts from wildlife, wind, work crews, and passing trains separately from confirmed human intrusion so the team can prove the model is reducing noise, not adding it.

Detection Logic

  • Authorized vs Unauthorized: The model reads high-visibility PPE to distinguish authorized hi-vis maintenance crews from civilian trespassers, so routine track work does not trigger alerts but an unprotected body in the rail envelope does.
  • Motion Path Analysis: Directional human travel toward the rails is separated from ambient movement like blowing debris or animals, keeping nuisance alarms low.
  • Low-Light Augmentation: Inference runs directly on IR and thermal camera feeds, so detection holds up in unlit tunnels and overnight yards.
VMS Integration Stack

Automated Dispatch Paging

Bypass the VMS entirely if needed. DHI can trigger physical sirens, flashers, or route IP pages directly to the train control center the moment track occupancy is compromised.

Genetec Certified
Milestone Ready

Industrial Privacy & Sovereignty

DHI's models do not stream raw video to the public cloud. All safety inferencing occurs on-premise within your local network, supporting privacy review and tighter control for critical infrastructure environments.