Published patent application, Indian Patent Office
Adaptive traffic signal control with real-time vehicle classification and weather data
A traffic control system I co-invented, which uses deep learning and reinforcement learning to optimize traffic flow and give emergency vehicles priority. It times signals with real-time vehicle classification, weather monitoring and a multi-factor priority index.
The problem
Fixed-time traffic signals ignore real-time conditions, which causes congestion, wasted fuel and dangerous delays for emergency vehicles. Existing adaptive systems address single factors, such as weather or vehicle count, rather than combining video analytics, weather, vehicle type and emergency priority.
How the system works
- IP cameras at intersections feed video to an edge AI control unit (NVIDIA Jetson AGX Orin, 2048-core Ampere GPU, 275 TOPS).
- A deep learning module classifies pedestrians, emergency vehicles, two-wheelers, auto-rickshaws, cars and buses in real time.
- A weather server with temperature, humidity and wind sensors gives two-wheelers priority in bad weather, for rider safety.
- A priority index per lane weighs vehicle types, counts, lane density, occupied area, weather severity and estimated passengers (car 3, bus 25, two-wheeler 1).
- Emergency vehicle preemption switches the signal, or shortens the wait, based on the vehicle's distance.
- Hardware: LED signal controllers on Arduino and STM32 microcontrollers, CSI camera interfaces, and I2C, SPI and UART sensor buses.
Examination
- Published by the Indian Patent Office as application 202341042839, with 10 claims.
- Responded to four prior-art citations (US20210409649A1, US20190333375A1, US20220108607A1 and US20160381325A1), arguing that none of them, alone or together, teaches the multi-factor priority system.
- Responded to a Section 3(k) objection ("computer program per se") by showing a tightly integrated hardware and software system with specific physical components.