Millions of New Yorkers Ride the Bus. Blocked Lanes Were Making Them Late.
Bus lane violations don't just slow individual commuters. In a city where millions depend on buses every day, a single blocked lane creates delays that cascade across the entire system. Manual enforcement couldn't keep pace with the scale of the problem. Officers can't be everywhere. Violations happen faster than they can be documented.
The city needed enforcement that never goes off duty.
Built for the Street, Not the Lab
We partnered with a leading fleet safety solutions provider to design and deploy ABLE — Automated Bus Lane Enforcement — a computer vision system that runs directly on city buses using NVIDIA Jetson Xavier edge devices. No cloud required. No connectivity dependency. Everything processes in real time, on the bus, in conditions that would defeat a less resilient system.
We built four custom deep learning models, each trained on manually labeled datasets collected from simulated street drives:
- Curb Detection — assesses horizontal positioning relative to sidewalks
- Crosswalk Detection — enhances longitudinal localization using known street features
- Vehicle Detection — identifies and tracks vehicles encroaching on bus lanes
- Lane Detection — determines whether a violation is occurring in real time
We started with a pilot in Queens to test and refine the models before expanding to Manhattan and the Bronx. In dense urban corridors, GPS proved unreliable. We responded by developing visual navigation algorithms that replaced GPS-based localization entirely — ensuring consistent performance regardless of signal strength. The system adapted to the city, not the other way around.
91% Accuracy. 5,000+ Buses. Enforcement at Scale.
ABLE now runs on more than 5,000 New York City buses, meeting the city's strict certification requirements and delivering a 91% F1 score across diverse urban conditions. On-device processing eliminates cloud latency and reduces operational costs by removing bandwidth overhead entirely.
Bus lanes are clearer. Commutes are faster. And the enforcement infrastructure that makes it possible operates without human intervention across one of the most complex transit systems in the world.
This work established our client as the Metropolitan Transportation Authority's trusted provider of video enforcement solutions — a position built on AI that performs in the real world, not just in a demo.
That's the difference between a pilot that works in Queens and a platform that scales across New York City.