AI-DRIVEN ADAPTIVE TRAFFIC MANAGEMENT SYSTEM

Abstract

Increasing traffic demand has made efficient signal management a significant challenge, particularly at busy urban intersections. Conventional traffic signals generally rely on predetermined timing plans and therefore cannot adjust their operation according to the changing number of vehicles on different approaches. This can result in unnecessary waiting, uneven utilization of green time, longer queues, and increased fuel consumption. To address this issue, this work proposes an AI-driven adaptive traffic management system capable of modifying signal timing according to the observed traffic condition. The proposed system uses a camera to acquire live traffic footage, while OpenCV performs video processing and defines separate regions for the four approaches of the intersection. YOLOv8 is employed to detect vehicles and obtain individual vehicle counts for each direction. These counts are converted into traffic-density information and supplied to an adaptive timing algorithm that determines the appropriate green duration. The resulting control commands are transmitted wirelessly to ESP32 controllers, which operate the corresponding red, yellow, and green signal LEDs. By combining computer vision, deep-learning-based vehicle detection, adaptive decision making, and embedded control, the proposed approach provides a flexible alternative to fixed-time traffic operation. The system is intended to improve green-time utilization, reduce unnecessary vehicle waiting, and support the development of intelligent traffic infrastructure.