NHAI · Delhi–Mumbai Expressway
NHAI ATMS
Advanced Traffic Management System: real-time AI video analytics and incident response across India's longest expressway.
- Client
- NHAI, Delhi–Mumbai Expressway
- Domain
- Highway traffic management
- My role
- Integration and data pipeline, Master System Integrator team
Context
- Operators couldn't watch hundreds of camera feeds
- Vendor systems didn't talk to each other
- Response teams were sent without exact details
Recognition
12 Feb 2024
Successfully demonstrated the ATMS project to:
- Shri Nitin GadkariMinister of Road Transport and Highways, India
- Shri Gajendra Singh ShekhawatMember of Lok Sabha, India
- Shri Manohar Lal KhattarChief Minister of Haryana, India
- Feb 2023Expressway inaugurated by the Prime Minister
The problem
- Too many feeds for human eyesOperators could not reliably watch every camera, so incidents were often reported by other drivers first.
- Systems that didn't talkCameras, speed sensors, number-plate readers and signboards each produced data on their own, with no single operational picture.
- Slow, vague alertsResponse teams were dispatched without a precise location or incident type.
- A network that keeps growingEvery new stretch of expressway would widen the gap unless monitoring became automatic.
How it works
- 01VIDS: at a T-junction, one pole carries three cameras, one per direction; one LPU runs AI on all three feeds and turns detections into events.
- 02VSDS: a gantry with a radar and a camera over each of 4 lanes; one LPU reads speed from the radar and number plates (ANPR) from the cameras.
- 03ECB: anyone at a roadside emergency calling box presses SOS and the call rings straight on the application server.
- 04Both LPUs push their data to the application server in the control room.
- 05Every camera is recorded on the video recorder (VRS); the server pulls live or recorded video for any event.
- 06VMS: the operator writes a message on the application server and publishes it to a sign on a gantry over the highway.
- 07ADAS: ambulances and patrol vehicles stay connected to the server: alerts, live location and voice calls.
What we built
- One pipeline for every vendor
- Cameras, sensors and signboards from many vendors
- One event-driven pipeline, one operational picture
- Device health tracked for every unit
- AI on every camera
- Wrong-way driving and stopped vehicles
- Speeding and congestion
- Alert with location and a live camera view
- Live map control room
- Every device and incident on one map
- Problems visible the moment they start
- Reports and audit trail for management
- Talk to drivers
- Warnings pushed to roadside signboards
- Roadside emergency calls into the same console
The hard parts
- Systems we didn't own. Make many vendors' systems work as one.
- Trusted alerts. False alarms teach people to ignore the screen.
- The trade-off. Prioritised operator-verified alerts over full automation - fewer false dispatches at the cost of some latency on edge cases.
Before and after
Incident detection
Reported by other drivers
Detected automatically from camera feeds
Incident reporting
Manual, after the fact
Real time
Speed detection
Not tracked live
Live speed detection system
Coverage
Stand-alone vendor systems
580+ km on three corridors in one control room
My part
As part of the Master System Integrator team, I built the centralised event-driven pipeline that brings every vendor's system together, and integrated six AI and computer-vision modules (incident detection, speed detection, signboards, number-plate recognition, driver assistance and pothole detection) into the command centre.
What I took away
- Agree the data contract before the features.
- Earn trust in alerts one true positive at a time.
- On a highway, minutes decide outcomes.
Want the detail behind the numbers - architecture, failure modes, what I’d change? Happy to walk through it.
Talk about this project →