Smart Curbs

What can a city bus see? With the MIT Senseable City Lab, I fitted public buses in Paris with computer-vision sensors that measured pedestrian activity and traffic along their routes. GDPR set the main rule: every image had to be processed on the device, and only anonymized metadata could leave the vehicle.

Each sensor was a Raspberry Pi running a compressed neural network. I wrote a fault-tolerant microservice pipeline for the whole data lifecycle (classification, local storage, sync), plus power management that woke the device when the bus started and put it to sleep when it stopped. Remote SSH access let me monitor the fleet and push updates over the air.

To prototype, I strapped a sensor to a bicycle and rode bus routes through Paris, which surfaced the real edge cases: Wi-Fi handoffs and battery quirks. Integrating with RATP, the Paris transit authority, turned up a stranger one: its API rejected negative coordinates, so data from some parts of the city simply failed. Instead of waiting for an upstream fix, I added a coordinate transform to the pipeline.

The sensors ran autonomously on public buses for five weeks. A web dashboard I built showed activity, traffic density, and pedestrian flow in near real time, making the case that transit fleets can double as mobile monitoring stations for the city.