field tested · real intersection · kolkata

Autonomous Traffic Control

a CV pipeline that reads live traffic video, counts vehicles per lane, and gives green time proportional to demand. tested at a real intersection. built from scratch, twice.

github

the algorithm

adaptive phase scheduling

green time is proportional to lane density. the busiest pair of approaches gets more time. empty lanes don't block busy ones.

Signal Controller · Live Simulation

running

N/S Approaches

green

North12 vehicles

South9 vehicles

E/W Approaches

red

East5 vehicles

West3 vehicles

phase timer7.2s remaining · NS phase active

fixed timing would give equal time to all lanes regardless of demand, wasting cycles on empty approaches

detection output

yolo sees every lane.

YOLOv8 draws bounding boxes per vehicle and per lane mask. the count feeds directly into the timing model.

YOLOv8

detection model

Kolkata

tested live IRL

YOLO DetectionLane DensityQueue TimingImplemented IRL

five years apart

2019 prototype

hardware-first

a Raspberry Pi driving a physical intersection simulator. IR sensors and Logitech webcams estimated lane density. real hardware in the loop, real GPIO signals.

built with Debayan. got picked up by the Times of India.

Raspberry PiPythonIR sensorsLogitech webcams

hardware-driven

2024 revamp

computer vision

rebuilt as a full CV pipeline. video in, detections out, per-lane counts, adaptive timing decisions. tested on a real intersection in Kolkata. solo rebuild.

the goal was the same as 2019: stop empty lanes from blocking busy ones. the approach became a lot more accurate.

PythonYOLOv8OpenCVSolo build

software-driven

press

times of india.

the 2019 prototype got picked up by the Times of India. for a student project that started as a Raspberry Pi experiment, that was a big moment.

the coverage focused on the practical angle: a low-cost system that could help at intersections where traditional infrastructure is expensive or unavailable.

Times of India · 2019

Times of India feature on the traffic control project

team · 2019 build

two people, one Raspberry Pi, one real intersection.

Kush Sharma

builder

Debayan

builder