Project
Autonomous Traffic Control
2024 revamp
Computer vision traffic control built for real intersections
I rebuilt the system as a full CV pipeline that reads live traffic video, estimates lane demand, and schedules phases for throughput. The focus was simple: reduce wasted time at lights in places where infrastructure is weak and traffic is unpredictable.
This version was tested at a real intersection in Kolkata. It uses YOLOv8 for vehicle detection and a lightweight timing model to decide which lane gets green and for how long. The rebuild was done solo.
Pipeline
Video in, detections out, per lane counts, then timing decisions that balance queues.
Goal
Make signals react to real traffic so empty lanes are not blocking busy ones.
2019 prototype
Hardware first traffic control
The first version was a Raspberry Pi driven intersection simulator. We used Python, IR sensors, and Logitech webcams to estimate lane density and drive phase timing with real hardware in the loop.
It was featured in the Times of India, which was a huge moment for a student project. This was built in 2019 with Debayan as a collaborator.
Collaborators
